diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/launch_plugin.py b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/launch_plugin.py
index 3b09713e12..90af6b7053 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/launch_plugin.py
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/launch_plugin.py
@@ -103,7 +103,7 @@ def default_setup(self):
                        fct_mod=(self.reset_makeopts,(),{}),
                        allowed=['auto', 'none', 'sse4', 'avx2','512y','512z'])
         self.add_param('cudacpp_backend', 'CPP', include=False, hidden=False,
-                       allowed=['Fortan', 'CPP', 'CUDA'])
+                       allowed=['Fortran', 'CPP', 'CUDA'])
         self['vector_size'] = 16 # already setup in default class (just change value)
         self['aloha_flag'] = '--fast-math'
         self['matrix_flag'] = '-O3'
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/AUTHORS b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/AUTHORS
index 8541e954b9..71519d1ad8 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/AUTHORS
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/AUTHORS
@@ -10,6 +10,7 @@ generates includes the following authors:
   Stephan Hageboeck (CERN)
   Olivier Mattelaer (Universite Catholique de Louvain, original author)
   Stefan Roiser (CERN, original author)
+  Jorgen Teig (CERN)
   Andrea Valassi (CERN, original author)
   Zenny Wettersten (CERN)
 
@@ -28,5 +29,4 @@ acknowledged collaboration with the following collaborators:
   Taran Singhania (PES University Bangalore)
   David Smith (CERN)
   Carl Vuosalo (University of Wisconsin-Madison)
-  Joergen Teig (CERN)
 
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/COPYRIGHT b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/COPYRIGHT
index a134b5fef9..9036d9260a 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/COPYRIGHT
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/COPYRIGHT
@@ -15,6 +15,7 @@ The full development team currently includes the following authors :
   Stephan Hageboeck (CERN)
   Olivier Mattelaer (Universite Catholique de Louvain, original author)
   Stefan Roiser (CERN, original author)
+  Jorgen Teig (CERN)
   Andrea Valassi (CERN, original author)
   Zenny Wettersten (CERN)
 See https://github.com/madgraph5/madgraph4gpu for more details. For the full
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/cpp_model_parameters_cc.inc b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/cpp_model_parameters_cc.inc
index 3c231bdbd6..54ce4c64cf 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/cpp_model_parameters_cc.inc
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/cpp_model_parameters_cc.inc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 %(info_lines)s
@@ -15,7 +15,7 @@
 #include <iomanip>
 #include <iostream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using namespace mg5amcGpu;
 #else
 using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/cpp_model_parameters_h.inc b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/cpp_model_parameters_h.inc
index 0250c160ed..94b8dd6444 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/cpp_model_parameters_h.inc
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/cpp_model_parameters_h.inc
@@ -25,7 +25,7 @@
 #include "read_slha.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -85,7 +85,7 @@ namespace mg5amcCpu
 #include <limits>
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -155,7 +155,7 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -172,7 +172,7 @@ namespace mg5amcCpu
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wunused-variable"  // e.g. <<warning: unused variable ‘mdl_G__exp__2’ [-Wunused-variable]>>
 #pragma GCC diagnostic ignored "-Wunused-parameter" // e.g. <<warning: unused parameter ‘G’ [-Wunused-parameter]>>
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #pragma nv_diagnostic push
 #pragma nv_diag_suppress 177 // e.g. <<warning #177-D: variable "mdl_G__exp__2" was declared but never referenced>>
 #endif
@@ -196,7 +196,7 @@ namespace mg5amcCpu
 %(eftspecial2)s
       return out;
     }
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #pragma GCC diagnostic pop
 #pragma nv_diagnostic pop
 #endif
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/Bridge.h b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/Bridge.h
index bf8b5e024d..f9ed70dfde 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/Bridge.h
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/Bridge.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Roiser (Nov 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef BRIDGE_H
 #define BRIDGE_H 1
@@ -14,16 +14,23 @@
 #include "MemoryAccessMomenta.h"  // for MemoryAccessMomenta::neppM
 #include "MemoryBuffers.h"        // for HostBufferMomenta, DeviceBufferMomenta etc
 
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
+
+#include <sys/stat.h> // bypass std::filesystem #803
+
 #include <algorithm>
 #include <cassert>
 #include <cmath>
 #include <cstring>
-#include <filesystem>
 #include <iostream>
 #include <memory>
 #include <type_traits>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +90,7 @@ namespace mg5amcCpu
     Bridge& operator=( const Bridge& ) = delete;
     Bridge& operator=( Bridge&& ) = delete;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     /**
      * Set the gpublocks and gputhreads for the gpusequence - throws if evnt != gpublocks*gputhreads
      * (this is needed for BridgeKernel tests rather than for actual production use in Fortran)
@@ -150,7 +157,7 @@ namespace mg5amcCpu
     unsigned int m_nevt; // number of events
     int m_nGoodHel;      // the number of good helicities (-1 initially when they have not yet been calculated)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     int m_gputhreads; // number of gpu threads (default set from number of events, can be modified)
     int m_gpublocks;  // number of gpu blocks (default set from number of events, can be modified)
     DeviceBuffer<FORTRANFPTYPE, sizePerEventMomenta> m_devMomentaF;
@@ -187,12 +194,12 @@ namespace mg5amcCpu
   // Forward declare transposition methods
   //
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
   template<typename Tin, typename Tout>
   __global__ void dev_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt );
 
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
 
   template<typename Tin, typename Tout>
   void hst_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt );
@@ -209,7 +216,7 @@ namespace mg5amcCpu
   Bridge<FORTRANFPTYPE>::Bridge( unsigned int nevtF, unsigned int nparF, unsigned int np4F )
     : m_nevt( nevtF )
     , m_nGoodHel( -1 )
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     , m_gputhreads( 256 )                  // default number of gpu threads
     , m_gpublocks( m_nevt / m_gputhreads ) // this ensures m_nevt <= m_gpublocks*m_gputhreads
     , m_devMomentaF( m_nevt )
@@ -233,7 +240,7 @@ namespace mg5amcCpu
   {
     if( nparF != CPPProcess::npar ) throw std::runtime_error( "Bridge constructor: npar mismatch" );
     if( np4F != CPPProcess::np4 ) throw std::runtime_error( "Bridge constructor: np4 mismatch" );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( ( m_nevt < s_gputhreadsmin ) || ( m_nevt % s_gputhreadsmin != 0 ) )
       throw std::runtime_error( "Bridge constructor: nevt should be a multiple of " + std::to_string( s_gputhreadsmin ) );
     while( m_nevt != m_gpublocks * m_gputhreads )
@@ -249,20 +256,28 @@ namespace mg5amcCpu
 #else
     std::cout << "WARNING! Instantiate host Bridge (nevt=" << m_nevt << ")" << std::endl;
     m_pmek.reset( new MatrixElementKernelHost( m_hstMomentaC, m_hstGs, m_hstRndHel, m_hstRndCol, m_hstMEs, m_hstSelHel, m_hstSelCol, m_nevt ) );
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
     // Create a process object, read param card and set parameters
     // FIXME: the process instance can happily go out of scope because it is only needed to read parameters?
     // FIXME: the CPPProcess should really be a singleton? what if fbridgecreate is called from several Fortran threads?
     CPPProcess process( /*verbose=*/false );
     std::string paramCard = "../../Cards/param_card.dat";
-    if( !std::filesystem::exists( paramCard ) )
-    {
-      paramCard = "../" + paramCard;
-    }
+    /*
+#ifdef __HIPCC__
+    if( !std::experimental::filesystem::exists( paramCard ) ) paramCard = "../" + paramCard;
+#else
+    if( !std::filesystem::exists( paramCard ) ) paramCard = "../" + paramCard;
+#endif
+    */
+    //struct stat dummybuffer; // bypass std::filesystem #803
+    //if( !( stat( paramCard.c_str(), &dummyBuffer ) == 0 ) ) paramCard = "../" + paramCard; //
+    auto fileExists = []( std::string& fileName )
+    { struct stat buffer; return stat( fileName.c_str(), &buffer ) == 0; };
+    if( !fileExists( paramCard ) ) paramCard = "../" + paramCard; // bypass std::filesystem #803
     process.initProc( paramCard );
   }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::set_gpugrid( const int gpublocks, const int gputhreads )
   {
@@ -276,7 +291,7 @@ namespace mg5amcCpu
   }
 #endif
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::gpu_sequence( const FORTRANFPTYPE* momenta,
                                             const FORTRANFPTYPE* gs,
@@ -291,14 +306,14 @@ namespace mg5amcCpu
     constexpr int neppM = MemoryAccessMomenta::neppM;
     if constexpr( neppM == 1 && std::is_same_v<FORTRANFPTYPE, fptype> )
     {
-      checkCuda( cudaMemcpy( m_devMomentaC.data(), momenta, m_devMomentaC.bytes(), cudaMemcpyHostToDevice ) );
+      gpuMemcpy( m_devMomentaC.data(), momenta, m_devMomentaC.bytes(), gpuMemcpyHostToDevice );
     }
     else
     {
-      checkCuda( cudaMemcpy( m_devMomentaF.data(), momenta, m_devMomentaF.bytes(), cudaMemcpyHostToDevice ) );
+      gpuMemcpy( m_devMomentaF.data(), momenta, m_devMomentaF.bytes(), gpuMemcpyHostToDevice );
       const int thrPerEvt = CPPProcess::npar * CPPProcess::np4; // AV: transpose alg does 1 element per thread (NOT 1 event per thread)
       //const int thrPerEvt = 1; // AV: try new alg with 1 event per thread... this seems slower
-      dev_transposeMomentaF2C<<<m_gpublocks * thrPerEvt, m_gputhreads>>>( m_devMomentaF.data(), m_devMomentaC.data(), m_nevt );
+      gpuLaunchKernel( dev_transposeMomentaF2C, m_gpublocks * thrPerEvt, m_gputhreads, m_devMomentaF.data(), m_devMomentaC.data(), m_nevt );
     }
     if constexpr( std::is_same_v<FORTRANFPTYPE, fptype> )
     {
@@ -341,7 +356,7 @@ namespace mg5amcCpu
   }
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::cpu_sequence( const FORTRANFPTYPE* momenta,
                                             const FORTRANFPTYPE* gs,
@@ -396,7 +411,7 @@ namespace mg5amcCpu
   // - C++ array: momenta[npagM][npar][np4][neppM] with nevt=npagM*neppM (AOSOA)
   //
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename Tin, typename Tout>
   __global__ void dev_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt )
   {
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/BridgeKernels.cc b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/BridgeKernels.cc
index d58066c9c1..eaf4037a24 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/BridgeKernels.cc
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/BridgeKernels.cc
@@ -1,17 +1,18 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "BridgeKernels.h"
 
+#include "GpuAbstraction.h"
 #include "MemoryAccessMomenta.h"
 
 #include <sstream>
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -45,7 +46,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu
 {
 
@@ -96,7 +97,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/BridgeKernels.h b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/BridgeKernels.h
index 15eb4bff4d..3efef8ce97 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/BridgeKernels.h
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/BridgeKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef BRIDGEKERNELS_H
 #define BRIDGEKERNELS_H 1
@@ -12,7 +12,7 @@
 #include "MatrixElementKernels.h"
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -49,7 +49,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A Bridge wrapper class encapsulating matrix element calculations on a CPU host
   class BridgeKernelHost final : public BridgeKernelBase
   {
@@ -89,7 +89,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A Bridge wrapper class encapsulating matrix element calculations on a GPU device
   class BridgeKernelDevice : public BridgeKernelBase
   {
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/CommonRandomNumberKernel.cc b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/CommonRandomNumberKernel.cc
index 985b39f576..010bc4cbd0 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/CommonRandomNumberKernel.cc
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/CommonRandomNumberKernel.cc
@@ -1,15 +1,16 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "CommonRandomNumbers.h"
+#include "GpuAbstraction.h"
 #include "MemoryBuffers.h"
 #include "RandomNumberKernels.h"
 
 #include <cassert>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/CrossSectionKernels.cc b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/CrossSectionKernels.cc
index 0b355a3c8d..c15b39844d 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/CrossSectionKernels.cc
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/CrossSectionKernels.cc
@@ -1,10 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "CrossSectionKernels.h"
 
+#include "GpuAbstraction.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessWeights.h"
 #include "MemoryBuffers.h"
@@ -77,7 +78,7 @@ debug_me_is_abnormal( const fptype& me, size_t ievtALL )
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -185,7 +186,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/CrossSectionKernels.h b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/CrossSectionKernels.h
index 7933ca4bbf..4d9659e04e 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/CrossSectionKernels.h
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/CrossSectionKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef CROSSSECTIONKERNELS_H
 #define CROSSSECTIONKERNELS_H 1
@@ -13,7 +13,7 @@
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -96,7 +96,7 @@ namespace mg5amcCpu
   //--------------------------------------------------------------------------
 
   /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating the calculation of event statistics on a GPU device
   class CrossSectionKernelDevice : public CrossSectionKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/CurandRandomNumberKernel.cc b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/CurandRandomNumberKernel.cc
index eb56333b03..08a16f6f2c 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/CurandRandomNumberKernel.cc
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/CurandRandomNumberKernel.cc
@@ -1,9 +1,9 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "MemoryBuffers.h"
 #include "RandomNumberKernels.h"
 
@@ -22,7 +22,7 @@ inline void assertCurand( curandStatus_t code, const char *file, int line, bool
 }
 #endif /* clang-format on */
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -36,7 +36,7 @@ namespace mg5amcCpu
   {
     if( m_isOnDevice )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if( !m_rnarray.isOnDevice() )
         throw std::runtime_error( "CurandRandomNumberKernel on device with a host random number array" );
 #else
@@ -114,7 +114,7 @@ namespace mg5amcCpu
     /*
     printf( "\nCurandRandomNumberKernel::generateRnarray size = %d\n", (int)m_rnarray.size() );
     fptype* data = m_rnarray.data();
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( m_rnarray.isOnDevice() )
     {
       data = new fptype[m_rnarray.size()]();
@@ -123,7 +123,7 @@ namespace mg5amcCpu
 #endif
     for( int i = 0; i < ( (int)m_rnarray.size() / 4 ); i++ )
       printf( "[%4d] %f %f %f %f\n", i * 4, data[i * 4], data[i * 4 + 2], data[i * 4 + 2], data[i * 4 + 3] );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( m_rnarray.isOnDevice() ) delete[] data;
 #endif
     */
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/EventStatistics.h b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/EventStatistics.h
index 48b51e0a49..b425a5bade 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/EventStatistics.h
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/EventStatistics.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef EventStatistics_H
 #define EventStatistics_H 1
@@ -16,7 +16,7 @@
 #include <limits>
 #include <string>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/GpuAbstraction.h b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/GpuAbstraction.h
new file mode 100644
index 0000000000..9c467b1e04
--- /dev/null
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/GpuAbstraction.h
@@ -0,0 +1,69 @@
+// Copyright (C) 2020-2023 CERN and UCLouvain.
+// Licensed under the GNU Lesser General Public License (version 3 or later).
+// Created by: J. Teig (Jul 2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+
+#ifndef MG5AMC_GPUABSTRACTION_H
+#define MG5AMC_GPUABSTRACTION_H 1
+
+#include <cassert>
+
+//--------------------------------------------------------------------------
+
+#ifdef __CUDACC__
+
+#define gpuError_t cudaError_t
+#define gpuPeekAtLastError cudaPeekAtLastError
+#define gpuGetErrorString cudaGetErrorString
+#define gpuSuccess cudaSuccess
+
+#define gpuMallocHost( ptr, size ) checkGpu( cudaMallocHost( ptr, size ) )
+#define gpuMalloc( ptr, size ) checkGpu( cudaMalloc( ptr, size ) )
+
+#define gpuMemcpy( dstData, srcData, srcBytes, func ) checkGpu( cudaMemcpy( dstData, srcData, srcBytes, func ) )
+#define gpuMemcpyHostToDevice cudaMemcpyHostToDevice
+#define gpuMemcpyDeviceToHost cudaMemcpyDeviceToHost
+#define gpuMemcpyToSymbol( type1, type2, size ) checkGpu( cudaMemcpyToSymbol( type1, type2, size ) )
+
+#define gpuFree( ptr ) checkGpu( cudaFree( ptr ) )
+#define gpuFreeHost( ptr ) checkGpu( cudaFreeHost( ptr ) )
+
+#define gpuSetDevice cudaSetDevice
+#define gpuDeviceSynchronize cudaDeviceSynchronize
+#define gpuDeviceReset cudaDeviceReset
+
+#define gpuLaunchKernel( kernel, blocks, threads, ... ) kernel<<<blocks, threads>>>( __VA_ARGS__ )
+#define gpuLaunchKernelSharedMem( kernel, blocks, threads, sharedMem, ... ) kernel<<<blocks, threads, sharedMem>>>( __VA_ARGS__ )
+
+//--------------------------------------------------------------------------
+
+#elif defined __HIPCC__
+
+#define gpuError_t hipError_t
+#define gpuPeekAtLastError hipPeekAtLastError
+#define gpuGetErrorString hipGetErrorString
+#define gpuSuccess hipSuccess
+
+#define gpuMallocHost( ptr, size ) checkGpu( hipHostMalloc( ptr, size ) ) // HostMalloc better
+#define gpuMalloc( ptr, size ) checkGpu( hipMalloc( ptr, size ) )
+
+#define gpuMemcpy( dstData, srcData, srcBytes, func ) checkGpu( hipMemcpy( dstData, srcData, srcBytes, func ) )
+#define gpuMemcpyHostToDevice hipMemcpyHostToDevice
+#define gpuMemcpyDeviceToHost hipMemcpyDeviceToHost
+#define gpuMemcpyToSymbol( type1, type2, size ) checkGpu( hipMemcpyToSymbol( type1, type2, size ) )
+
+#define gpuFree( ptr ) checkGpu( hipFree( ptr ) )
+#define gpuFreeHost( ptr ) checkGpu( hipHostFree( ptr ) )
+
+#define gpuSetDevice hipSetDevice
+#define gpuDeviceSynchronize hipDeviceSynchronize
+#define gpuDeviceReset hipDeviceReset
+
+#define gpuLaunchKernel( kernel, blocks, threads, ... ) kernel<<<blocks, threads>>>( __VA_ARGS__ )
+#define gpuLaunchKernelSharedMem( kernel, blocks, threads, sharedMem, ... ) kernel<<<blocks, threads, sharedMem>>>( __VA_ARGS__ )
+
+//--------------------------------------------------------------------------
+
+#endif
+
+#endif // MG5AMC_GPUABSTRACTION_H
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/CudaRuntime.h b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/GpuRuntime.h
similarity index 62%
rename from epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/CudaRuntime.h
rename to epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/GpuRuntime.h
index 64ce52f4b3..93579ef08b 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/CudaRuntime.h
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/GpuRuntime.h
@@ -1,49 +1,50 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
-// Created by: S. Roiser (Jul 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Created by: J. Teig (Jun 2023, based on earlier work by S. Roiser) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
-#ifndef MG5AMC_CUDARUNTIME_H
-#define MG5AMC_CUDARUNTIME_H 1
+#ifndef MG5AMC_GPURUNTIME_H
+#define MG5AMC_GPURUNTIME_H 1
 
 // MG5AMC on GPU uses the CUDA runtime API, not the lower level CUDA driver API
 // See https://docs.nvidia.com/cuda/cuda-runtime-api/driver-vs-runtime-api.html#driver-vs-runtime-api
 
-#include <cassert>
+#include "GpuAbstraction.h"
+
 #include <iostream>
 
 //--------------------------------------------------------------------------
 
 // See https://stackoverflow.com/a/14038590
-#ifdef __CUDACC__ /* clang-format off */
-#define checkCuda( code ) { assertCuda( code, __FILE__, __LINE__ ); }
-inline void assertCuda( cudaError_t code, const char* file, int line, bool abort = true )
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
+#define checkGpu( code ) { assertGpu( code, __FILE__, __LINE__ ); }
+inline void assertGpu( gpuError_t code, const char* file, int line, bool abort = true )
 {
-  if( code != cudaSuccess )
+  if( code != gpuSuccess )
   {
-    printf( "ERROR! assertCuda: '%s' (%d) in %s:%d\n", cudaGetErrorString( code ), code, file, line );
-    if( abort ) assert( code == cudaSuccess );
+    printf( "ERROR! assertGpu: '%s' (%d) in %s:%d\n", gpuGetErrorString( code ), code, file, line );
+    if( abort ) assert( code == gpuSuccess );
   }
 }
 #endif /* clang-format on */
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
+  // Instantiate a GpuRuntime at the beginnining of the application's main to
+  // invoke gpuSetDevice(0) in the constructor and book a gpuDeviceReset() call in the destructor
   // *** FIXME! This will all need to be designed differently when going to multi-GPU nodes! ***
-  struct CudaRuntime final
+  struct GpuRuntime final
   {
-    CudaRuntime( const bool debug = true )
+    GpuRuntime( const bool debug = true )
       : m_debug( debug ) { setUp( m_debug ); }
-    ~CudaRuntime() { tearDown( m_debug ); }
-    CudaRuntime( const CudaRuntime& ) = delete;
-    CudaRuntime( CudaRuntime&& ) = delete;
-    CudaRuntime& operator=( const CudaRuntime& ) = delete;
-    CudaRuntime& operator=( CudaRuntime&& ) = delete;
+    ~GpuRuntime() { tearDown( m_debug ); }
+    GpuRuntime( const GpuRuntime& ) = delete;
+    GpuRuntime( GpuRuntime&& ) = delete;
+    GpuRuntime& operator=( const GpuRuntime& ) = delete;
+    GpuRuntime& operator=( GpuRuntime&& ) = delete;
     bool m_debug;
 
     // Set up CUDA application
@@ -62,8 +63,8 @@ namespace mg5amcGpu
       */
       // Replace cudaFree(0) by cudaSetDevice(0), even if it is not really needed either
       // (but see https://developer.nvidia.com/blog/cuda-pro-tip-always-set-current-device-avoid-multithreading-bugs)
-      if( debug ) std::cout << "__CudaRuntime: calling cudaSetDevice(0)" << std::endl;
-      checkCuda( cudaSetDevice( 0 ) ); // SLOW!
+      if( debug ) std::cout << "__GpuRuntime: calling GpuSetDevice(0)" << std::endl;
+      checkGpu( gpuSetDevice( 0 ) ); // SLOW!
     }
 
     // Tear down CUDA application (call cudaDeviceReset)
@@ -72,14 +73,13 @@ namespace mg5amcGpu
     // See https://docs.nvidia.com/cuda/cuda-memcheck/index.html#leak-checking
     static void tearDown( const bool debug = true )
     {
-      if( debug ) std::cout << "__CudaRuntime: calling cudaDeviceReset()" << std::endl;
-      checkCuda( cudaDeviceReset() );
+      if( debug ) std::cout << "__GpuRuntime: calling GpuDeviceReset()" << std::endl;
+      checkGpu( gpuDeviceReset() );
     }
   };
-
 }
 #endif
 
 //--------------------------------------------------------------------------
 
-#endif // MG5AMC_CUDARUNTIME_H
+#endif // MG5AMC_GPURUNTIME_H
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MadgraphTest.h b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MadgraphTest.h
index ef40624c88..6054185300 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MadgraphTest.h
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MadgraphTest.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Hageboeck (Dec 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MADGRAPHTEST_H_
 #define MADGRAPHTEST_H_ 1
@@ -14,7 +14,11 @@
 
 #include <array>
 #include <cmath>
-#include <filesystem>
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
 #include <fstream>
 #include <iomanip>
 #include <memory>
@@ -22,7 +26,7 @@
 #include <string>
 #include <vector>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using mg5amcGpu::CPPProcess;
 #else
 using mg5amcCpu::CPPProcess;
@@ -201,7 +205,7 @@ class MadgraphTest : public testing::TestWithParam<TestDriverBase*>
 
 // Since we link both the CPU-only and GPU tests into the same executable, we prevent
 // a multiply defined symbol by only compiling this in the non-CUDA phase:
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
 /// Compare momenta and matrix elements.
 /// This uses an implementation of TestDriverBase to run a madgraph workflow,
@@ -219,7 +223,14 @@ TEST_P( MadgraphTest, CompareMomentaAndME )
   const char* dumpEventsC = getenv( "CUDACPP_RUNTEST_DUMPEVENTS" );
   const bool dumpEvents = ( dumpEventsC != 0 ) && ( std::string( dumpEventsC ) != "" );
   const std::string refFileName = testDriver->getRefFileName();
+  /*
+#ifdef __HIPCC__
+  const std::string dumpFileName = std::experimental::filesystem::path( refFileName ).filename();
+#else
   const std::string dumpFileName = std::filesystem::path( refFileName ).filename();
+#endif
+  */
+  const std::string dumpFileName = refFileName; // bypass std::filesystem #803
   std::ofstream dumpFile;
   if( dumpEvents )
   {
@@ -307,6 +318,6 @@ TEST_P( MadgraphTest, CompareMomentaAndME )
   }
 }
 
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
 
 #endif /* MADGRAPHTEST_H_ */
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MatrixElementKernels.cc b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MatrixElementKernels.cc
index 74b5239ebf..81699dfea9 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MatrixElementKernels.cc
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MatrixElementKernels.cc
@@ -1,12 +1,12 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "MatrixElementKernels.h"
 
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h" // Includes the abstraction for Nvidia/AMD compilation
 #include "MemoryAccessMomenta.h"
 #include "MemoryBuffers.h"
 
@@ -14,7 +14,7 @@
 
 //============================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu
 {
 
@@ -150,7 +150,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
@@ -209,13 +209,13 @@ namespace mg5amcGpu
     PinnedHostBufferHelicityMask hstIsGoodHel( ncomb );
     DeviceBufferHelicityMask devIsGoodHel( ncomb );
     // ... 0d1. Compute good helicity mask on the device
-    computeDependentCouplings<<<m_gpublocks, m_gputhreads>>>( m_gs.data(), m_couplings.data() );
+    gpuLaunchKernel( computeDependentCouplings, m_gpublocks, m_gputhreads, m_gs.data(), m_couplings.data() );
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-    sigmaKin_getGoodHel<<<m_gpublocks, m_gputhreads>>>( m_momenta.data(), m_couplings.data(), m_matrixElements.data(), m_numerators.data(), m_denominators.data(), devIsGoodHel.data() );
+    gpuLaunchKernel( sigmaKin_getGoodHel, m_gpublocks, m_gputhreads, m_momenta.data(), m_couplings.data(), m_matrixElements.data(), m_numerators.data(), m_denominators.data(), devIsGoodHel.data() );
 #else
-    sigmaKin_getGoodHel<<<m_gpublocks, m_gputhreads>>>( m_momenta.data(), m_couplings.data(), m_matrixElements.data(), devIsGoodHel.data() );
+    gpuLaunchKernel( sigmaKin_getGoodHel, m_gpublocks, m_gputhreads, m_momenta.data(), m_couplings.data(), m_matrixElements.data(), devIsGoodHel.data() );
 #endif
-    checkCuda( cudaPeekAtLastError() );
+    checkGpu( gpuPeekAtLastError() );
     // ... 0d2. Copy back good helicity mask to the host
     copyHostFromDevice( hstIsGoodHel, devIsGoodHel );
     // ... 0d3. Copy back good helicity list to constant memory on the device
@@ -226,19 +226,19 @@ namespace mg5amcGpu
 
   void MatrixElementKernelDevice::computeMatrixElements( const unsigned int channelId )
   {
-    computeDependentCouplings<<<m_gpublocks, m_gputhreads>>>( m_gs.data(), m_couplings.data() );
+    gpuLaunchKernel( computeDependentCouplings, m_gpublocks, m_gputhreads, m_gs.data(), m_couplings.data() );
 #ifndef MGONGPU_NSIGHT_DEBUG
     constexpr unsigned int sharedMemSize = 0;
 #else
     constexpr unsigned int sharedMemSize = ntpbMAX * sizeof( float );
 #endif
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-    sigmaKin<<<m_gpublocks, m_gputhreads, sharedMemSize>>>( m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), channelId, m_numerators.data(), m_denominators.data(), m_selhel.data(), m_selcol.data() );
+    gpuLaunchKernelSharedMem( sigmaKin, m_gpublocks, m_gputhreads, sharedMemSize, m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), channelId, m_numerators.data(), m_denominators.data(), m_selhel.data(), m_selcol.data() );
 #else
-    sigmaKin<<<m_gpublocks, m_gputhreads, sharedMemSize>>>( m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), m_selhel.data(), m_selcol.data() );
+    gpuLaunchKernelSharedMem( sigmaKin, m_gpublocks, m_gputhreads, sharedMemSize, m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), m_selhel.data(), m_selcol.data() );
 #endif
-    checkCuda( cudaPeekAtLastError() );
-    checkCuda( cudaDeviceSynchronize() );
+    checkGpu( gpuPeekAtLastError() );
+    checkGpu( gpuDeviceSynchronize() );
   }
 
   //--------------------------------------------------------------------------
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MatrixElementKernels.h b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MatrixElementKernels.h
index 23e84757a2..72bd8f195b 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MatrixElementKernels.h
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MatrixElementKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MATRIXELEMENTKERNELS_H
 #define MATRIXELEMENTKERNELS_H 1
@@ -10,7 +10,7 @@
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -81,7 +81,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating matrix element calculations on a CPU host
   class MatrixElementKernelHost final : public MatrixElementKernelBase, public NumberOfEvents
   {
@@ -130,7 +130,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating matrix element calculations on a GPU device
   class MatrixElementKernelDevice : public MatrixElementKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessAmplitudes.h b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessAmplitudes.h
index 573b3bbbc9..ffb76e93de 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessAmplitudes.h
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessAmplitudes.h
@@ -15,7 +15,7 @@
 #define MGONGPU_TRIVIAL_AMPLITUDES 1
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessCouplings.h b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessCouplings.h
index 1afc589b11..b4b76f3842 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessCouplings.h
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessCouplings.h
@@ -15,7 +15,7 @@
 #include "MemoryBuffers.h"       // for HostBufferCouplings::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessCouplingsFixed.h b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessCouplingsFixed.h
index dc0d93afff..ffcdf4dbef 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessCouplingsFixed.h
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessCouplingsFixed.h
@@ -14,7 +14,7 @@
 //#include "MemoryAccessHelpers.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessDenominators.h b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessDenominators.h
index 3bce635718..66f2d32a6b 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessDenominators.h
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessDenominators.h
@@ -10,7 +10,7 @@
 #include "MemoryAccessGs.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessGs.h b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessGs.h
index 31311aa375..4c726b30f3 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessGs.h
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessGs.h
@@ -13,7 +13,7 @@
 #include "MemoryBuffers.h" // for HostBufferMatrixElements::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessHelpers.h b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessHelpers.h
index c82a6c7635..db73e4e064 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessHelpers.h
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessHelpers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessHelpers_H
 #define MemoryAccessHelpers_H 1
@@ -105,7 +105,7 @@ class KernelAccessHelper : public MemoryAccessHelper<T>
     }
     else
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
       //printf( "kernelAccessRecord: ievt=%d threadId=%d\n", ievt, threadIdx.x );
       return T::ieventAccessRecord( buffer, ievt ); // NB fptype and fptype_sv coincide for CUDA
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessMatrixElements.h b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessMatrixElements.h
index f32e6fea5b..3741011971 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessMatrixElements.h
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessMatrixElements.h
@@ -13,7 +13,7 @@
 #include "MemoryBuffers.h" // for HostBufferMatrixElements::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessMomenta.h b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessMomenta.h
index 29266de32c..3be229d392 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessMomenta.h
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessMomenta.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessMomenta_H
 #define MemoryAccessMomenta_H 1
@@ -13,7 +13,7 @@
 #include "MemoryAccessVectors.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -30,7 +30,7 @@ namespace mg5amcCpu
 
     // Number of Events Per Page in the momenta AOSOA memory buffer layout
     // (these are all best kept as a compile-time constants: see issue #23)
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
     // -----------------------------------------------------------------------------------------------
     // --- GPUs: neppM is best set to a power of 2 times the number of fptype's in a 32-byte cacheline
     // --- This is relevant to ensure coalesced access to momenta in global memory
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessNumerators.h b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessNumerators.h
index b152183b28..18991f4fa6 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessNumerators.h
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessNumerators.h
@@ -10,7 +10,7 @@
 #include "MemoryAccessGs.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessRandomNumbers.h b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessRandomNumbers.h
index e2988d39f3..40cb089135 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessRandomNumbers.h
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessRandomNumbers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessRandomNumbers_H
 #define MemoryAccessRandomNumbers_H 1
@@ -11,7 +11,7 @@
 #include "CPPProcess.h"
 #include "MemoryAccessHelpers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using mg5amcGpu::CPPProcess;
 #else
 using mg5amcCpu::CPPProcess;
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessVectors.h b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessVectors.h
index e9b197368e..08faccff0f 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessVectors.h
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessVectors.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessVectors_H
 #define MemoryAccessVectors_H 1
@@ -10,7 +10,7 @@
 
 #include "mgOnGpuVectors.h"
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu // this is only needed for CPU SIMD vectorization
 {
 
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessWavefunctions.h b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessWavefunctions.h
index 5428aaf933..33bef4559e 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessWavefunctions.h
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryAccessWavefunctions.h
@@ -15,7 +15,7 @@
 #define MGONGPU_TRIVIAL_WAVEFUNCTIONS 1
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryBuffers.h b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryBuffers.h
index 48306a9d41..f29b8c5357 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryBuffers.h
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MemoryBuffers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021, based on earlier work by S. Hageboeck) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryBuffers_H
 #define MemoryBuffers_H 1
@@ -11,12 +11,12 @@
 #include "mgOnGpuCxtypes.h"
 
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "Parameters_%(model_name)s.h"
 
 #include <sstream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -87,7 +87,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   constexpr bool HostBufferALIGNED = false;   // ismisaligned=false
   constexpr bool HostBufferMISALIGNED = true; // ismisaligned=true
 
@@ -119,7 +119,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA pinned host buffer
   template<typename T>
   class PinnedHostBufferBase : public BufferBase<T>
@@ -128,18 +128,18 @@ namespace mg5amcCpu
     PinnedHostBufferBase( const size_t size )
       : BufferBase<T>( size, false )
     {
-      checkCuda( cudaMallocHost( &( this->m_data ), this->bytes() ) );
+      gpuMallocHost( &( this->m_data ), this->bytes() );
     }
     virtual ~PinnedHostBufferBase()
     {
-      checkCuda( cudaFreeHost( this->m_data ) );
+      gpuFreeHost( this->m_data );
     }
   };
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA device buffer
   template<typename T>
   class DeviceBufferBase : public BufferBase<T>
@@ -148,18 +148,18 @@ namespace mg5amcCpu
     DeviceBufferBase( const size_t size )
       : BufferBase<T>( size, true )
     {
-      checkCuda( cudaMalloc( &( this->m_data ), this->bytes() ) );
+      gpuMalloc( &( this->m_data ), this->bytes() );
     }
     virtual ~DeviceBufferBase()
     {
-      checkCuda( cudaFree( this->m_data ) );
+      gpuFree( this->m_data );
     }
   };
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for a given number of events
   template<typename T, size_t sizePerEvent, bool ismisaligned>
   class HostBuffer : public HostBufferBase<T, ismisaligned>, virtual private NumberOfEvents
@@ -175,7 +175,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA pinned host buffer for a given number of events
   template<typename T, size_t sizePerEvent>
   class PinnedHostBuffer : public PinnedHostBufferBase<T>, virtual private NumberOfEvents
@@ -191,7 +191,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA device buffer for a given number of events
   template<typename T, size_t sizePerEvent>
   class DeviceBuffer : public DeviceBufferBase<T>, virtual private NumberOfEvents
@@ -213,7 +213,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for momenta random numbers
   constexpr size_t sizePerEventRndNumMomenta = MemoryBuffers::np4 * MemoryBuffers::nparf;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for momenta random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumMomenta, HostBufferALIGNED> HostBufferRndNumMomenta;
 #else
@@ -232,7 +232,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer with ONE fptype per event
   constexpr size_t sizePerEventOneFp = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer with ONE fptype per event
   typedef HostBuffer<fptype, sizePerEventOneFp, HostBufferALIGNED> HostBufferOneFp;
 #else
@@ -257,7 +257,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for Gs
   constexpr size_t sizePerEventGs = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventGs, HostBufferALIGNED> HostBufferGs;
 #else
@@ -276,7 +276,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for numerators
   constexpr size_t sizePerEventNumerators = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventNumerators, HostBufferALIGNED> HostBufferNumerators;
 #else
@@ -296,7 +296,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for denominators
   constexpr size_t sizePerEventDenominators = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventDenominators, HostBufferALIGNED> HostBufferDenominators;
 #else
@@ -315,7 +315,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for random numbers
   constexpr size_t sizePerEventCouplings = MemoryBuffers::ndcoup * MemoryBuffers::nx2;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventCouplings, HostBufferALIGNED> HostBufferCouplings;
 #else
@@ -333,7 +333,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for momenta
   constexpr size_t sizePerEventMomenta = MemoryBuffers::np4 * MemoryBuffers::npar;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for momenta
   typedef HostBuffer<fptype, sizePerEventMomenta, HostBufferALIGNED> HostBufferMomenta;
   //typedef HostBuffer<fptype, sizePerEventMomenta, HostBufferMISALIGNED> HostBufferMomenta; // TEST MISALIGNMENT!
@@ -352,7 +352,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for sampling weights
   constexpr size_t sizePerEventWeights = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for sampling weights
   typedef HostBuffer<fptype, sizePerEventWeights, HostBufferALIGNED> HostBufferWeights;
 #else
@@ -370,7 +370,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for matrix elements
   constexpr size_t sizePerEventMatrixElements = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for matrix elements
   typedef HostBuffer<fptype, sizePerEventMatrixElements, HostBufferALIGNED> HostBufferMatrixElements;
 #else
@@ -385,7 +385,7 @@ namespace mg5amcCpu
   // A base class encapsulating a memory buffer for the helicity mask
   typedef BufferBase<bool> BufferHelicityMask;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for the helicity mask
   typedef HostBufferBase<bool, HostBufferALIGNED> HostBufferHelicityMask;
 #else
@@ -403,7 +403,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for wavefunctions
   constexpr size_t sizePerEventWavefunctions = MemoryBuffers::nw6 * MemoryBuffers::nx2;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for wavefunctions
   typedef HostBuffer<fptype, sizePerEventWavefunctions, HostBufferALIGNED> HostBufferWavefunctions;
 #else
@@ -421,7 +421,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for helicity random numbers
   constexpr size_t sizePerEventRndNumHelicity = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for helicity random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumHelicity, HostBufferALIGNED> HostBufferRndNumHelicity;
 #else
@@ -439,7 +439,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for color random numbers
   constexpr size_t sizePerEventRndNumColor = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for color random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumColor, HostBufferALIGNED> HostBufferRndNumColor;
 #else
@@ -457,7 +457,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for helicity selection
   constexpr size_t sizePerEventSelectedHelicity = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for helicity selection
   typedef HostBuffer<int, sizePerEventSelectedHelicity, HostBufferALIGNED> HostBufferSelectedHelicity;
 #else
@@ -475,7 +475,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for color selection
   constexpr size_t sizePerEventSelectedColor = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for color selection
   typedef HostBuffer<int, sizePerEventSelectedColor, HostBufferALIGNED> HostBufferSelectedColor;
 #else
@@ -487,7 +487,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<class Tdst, class Tsrc>
   void copyDeviceFromHost( Tdst& dst, const Tsrc& src ) // keep the same order of arguments as in memcpy
   {
@@ -504,13 +504,13 @@ namespace mg5amcCpu
       throw std::runtime_error( sstr.str() );
     }
     // NB (PR #45): cudaMemcpy involves an intermediate memcpy to pinned memory if host array is a not a pinned host array
-    checkCuda( cudaMemcpy( dst.data(), src.data(), src.bytes(), cudaMemcpyHostToDevice ) );
+    gpuMemcpy( dst.data(), src.data(), src.bytes(), gpuMemcpyHostToDevice );
   }
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<class Tdst, class Tsrc>
   void copyHostFromDevice( Tdst& dst, const Tsrc& src ) // keep the same order of arguments as in memcpy
   {
@@ -527,7 +527,7 @@ namespace mg5amcCpu
       throw std::runtime_error( sstr.str() );
     }
     // NB (PR #45): cudaMemcpy involves an intermediate memcpy to pinned memory if host array is a not a pinned host array
-    checkCuda( cudaMemcpy( dst.data(), src.data(), src.bytes(), cudaMemcpyDeviceToHost ) );
+    gpuMemcpy( dst.data(), src.data(), src.bytes(), gpuMemcpyDeviceToHost );
   }
 #endif
 
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/RamboSamplingKernels.cc b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/RamboSamplingKernels.cc
index da68aa9255..79abbcc4f8 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/RamboSamplingKernels.cc
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/RamboSamplingKernels.cc
@@ -1,11 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "RamboSamplingKernels.h"
 
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "MemoryAccessMomenta.h"
 #include "MemoryAccessRandomNumbers.h"
 #include "MemoryAccessWeights.h"
@@ -14,7 +14,7 @@
 
 #include <sstream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -92,7 +92,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingKernelDevice::RamboSamplingKernelDevice( const fptype energy,               // input: energy
                                                         const BufferRndNumMomenta& rndmom, // input: random numbers in [0,1]
                                                         BufferMomenta& momenta,            // output: momenta
@@ -135,7 +135,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   getMomentaInitialDevice( const fptype energy,
                            fptype* momenta )
@@ -147,17 +147,17 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   void
   RamboSamplingKernelDevice::getMomentaInitial()
   {
-    getMomentaInitialDevice<<<m_gpublocks, m_gputhreads>>>( m_energy, m_momenta.data() );
+    gpuLaunchKernel( getMomentaInitialDevice, m_gpublocks, m_gputhreads, m_energy, m_momenta.data() );
   }
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   getMomentaFinalDevice( const fptype energy,
                          const fptype* rndmom,
@@ -171,11 +171,11 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   void
   RamboSamplingKernelDevice::getMomentaFinal()
   {
-    getMomentaFinalDevice<<<m_gpublocks, m_gputhreads>>>( m_energy, m_rndmom.data(), m_momenta.data(), m_weights.data() );
+    gpuLaunchKernel( getMomentaFinalDevice, m_gpublocks, m_gputhreads, m_energy, m_rndmom.data(), m_momenta.data(), m_weights.data() );
   }
 #endif
 
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/RamboSamplingKernels.h b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/RamboSamplingKernels.h
index 184089efd7..7c214cd74b 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/RamboSamplingKernels.h
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/RamboSamplingKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef RAMBOSAMPLINGKERNELS_H
 #define RAMBOSAMPLINGKERNELS_H 1
@@ -10,7 +10,7 @@
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -93,7 +93,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating RAMBO phase space sampling on a GPU device
   class RamboSamplingKernelDevice final : public SamplingKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/RandomNumberKernels.h b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/RandomNumberKernels.h
index 188a72c2c9..21d63beeac 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/RandomNumberKernels.h
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/RandomNumberKernels.h
@@ -1,14 +1,14 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef RANDOMNUMBERKERNELS_H
 #define RANDOMNUMBERKERNELS_H 1
 
 #include "mgOnGpuConfig.h"
 
-// NB This must come AFTER mgOnGpuConfig.h which contains our definition of __global__ when __CUDACC__ is not defined
+// NB This must come AFTER mgOnGpuConfig.h which contains our definition of __global__ when MGONGPUCPP_GPUIMPL is not defined
 #ifndef MGONGPU_HAS_NO_CURAND
 //#include "curand.h"
 struct curandGenerator_st; // forward definition from curand.h
@@ -16,7 +16,7 @@ struct curandGenerator_st; // forward definition from curand.h
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/check_sa.cc b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/check_sa.cc
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/cpp_hel_amps_h.inc b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/cpp_hel_amps_h.inc
index 562af241af..b9840f1374 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/cpp_hel_amps_h.inc
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/cpp_hel_amps_h.inc
@@ -5,7 +5,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for %(output_name)s by
 %(info_lines)s
@@ -26,7 +26,7 @@
 //#include <iomanip>
 //#include <iostream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/cudacpp.mk b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/cudacpp.mk
index b399eb36b0..c570d2418c 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/cudacpp.mk
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/cudacpp.mk
@@ -1,7 +1,7 @@
 # Copyright (C) 2020-2023 CERN and UCLouvain.
 # Licensed under the GNU Lesser General Public License (version 3 or later).
 # Created by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-# Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+# Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #=== Determine the name of this makefile (https://ftp.gnu.org/old-gnu/Manuals/make-3.80/html_node/make_17.html)
 #=== NB: use ':=' to ensure that the value of CUDACPP_MAKEFILE is not modified further down after including make_opts
@@ -42,10 +42,10 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure common compiler flags for C++ and CUDA
+#=== Configure common compiler flags for C++ and CUDA/HIP
 
 INCFLAGS = -I.
-OPTFLAGS = -O3 # this ends up in CUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
+OPTFLAGS = -O3 # this ends up in GPUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
 
 # Dependency on src directory
 MG5AMC_COMMONLIB = mg5amc_common
@@ -121,24 +121,45 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure the CUDA compiler
+#=== Configure the GPU compiler (CUDA or HIP)
+
+# FIXME! (AV 24.01.2024)
+# In the current implementation (without separate builds for C++ and CUDA/HIP), we first check for cudacc and hipcc in CUDA_HOME and HIP_HOME.
+# If CUDA_HOME or HIP_HOME are not set, try to determine them from the path to cudacc and hipcc.
+# While convoluted, this is currently necessary to allow disabling CUDA/HIP builds by setting CUDA_HOME or HIP_HOME to invalid paths.
+# This will (probably?) be fixed when separate C++ and CUDA/HIP builds are implemented (PR #775).
 
 # If CXX is not a single word (example "clang++ --gcc-toolchain...") then disable CUDA builds (issue #505)
-# This is because it is impossible to pass this to "CUFLAGS += -ccbin <host-compiler>" below
+# This is because it is impossible to pass this to "GPUFLAGS += -ccbin <host-compiler>" below
 ifneq ($(words $(subst ccache ,,$(CXX))),1) # allow at most "CXX=ccache <host-compiler>" from outside
   $(warning CUDA builds are not supported for multi-word CXX "$(CXX)")
   override CUDA_HOME=disabled
 endif
 
-# If CUDA_HOME is not set, try to set it from the location of nvcc
+# If CUDA_HOME is not set, try to set it from the path to nvcc
 ifndef CUDA_HOME
-  CUDA_HOME = $(patsubst %%bin/nvcc,%%,$(shell which nvcc 2>/dev/null))
+  CUDA_HOME = $(patsubst %%/bin/nvcc,%%,$(shell which nvcc 2>/dev/null))
   $(warning CUDA_HOME was not set: using "$(CUDA_HOME)")
 endif
 
-# Set NVCC as $(CUDA_HOME)/bin/nvcc if it exists
+# If HIP_HOME is not set, try to set it from the path to hipcc
+ifndef HIP_HOME
+  HIP_HOME = $(patsubst %%/bin/hipcc,%%,$(shell which hipcc 2>/dev/null))
+  $(warning HIP_HOME was not set: using "$(HIP_HOME)")
+endif
+
+# FIXME! (AV 24.01.2024)
+# In the current implementation (without separate builds for C++ and CUDA/HIP),
+# builds are performed for HIP only if CUDA is not found in the path.
+# If both CUDA and HIP are installed, HIP builds can be triggered by unsetting CUDA_HOME.
+# This will be fixed when separate C++ and CUDA/HIP builds are implemented (PR #775).
+
+#--- Option 1: CUDA exists -> use CUDA
+
+# Set GPUCC as $(CUDA_HOME)/bin/nvcc if it exists
 ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
-  NVCC = $(CUDA_HOME)/bin/nvcc
+
+  GPUCC = $(CUDA_HOME)/bin/nvcc
   USE_NVTX ?=-DUSE_NVTX
   # See https://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/index.html
   # See https://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/
@@ -148,7 +169,7 @@ ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
   # Examples: use 60 for P100 (Piz Daint), 80 for A100 (Juwels Booster, NVidia raplab/Curiosity).
   MADGRAPH_CUDA_ARCHITECTURE ?= 70
   ###CUARCHFLAGS = -gencode arch=compute_$(MADGRAPH_CUDA_ARCHITECTURE),code=compute_$(MADGRAPH_CUDA_ARCHITECTURE) -gencode arch=compute_$(MADGRAPH_CUDA_ARCHITECTURE),code=sm_$(MADGRAPH_CUDA_ARCHITECTURE) # Older implementation (AV): go back to this one for multi-GPU support #533
-  ###CUARCHFLAGS = --gpu-architecture=compute_$(MADGRAPH_CUDA_ARCHITECTURE) --gpu-code=sm_$(MADGRAPH_CUDA_ARCHITECTURE),compute_$(MADGRAPH_CUDA_ARCHITECTURE) # Newer implementation (SH): cannot use this as-is for multi-GPU support #533
+  ###CUARCHFLAGS = --gpu-architecture=compute_$(MADGRAPH_CUDA_ARCHITECTURE) --gpu-code=sm_$(MADGRAPH_CUDA_ARCHITECTURE),compute_$(MADGRAPH_CUDA_ARCHITECTURE)  # Newer implementation (SH): cannot use this as-is for multi-GPU support #533
   comma:=,
   CUARCHFLAGS = $(foreach arch,$(subst $(comma), ,$(MADGRAPH_CUDA_ARCHITECTURE)),-gencode arch=compute_$(arch),code=compute_$(arch) -gencode arch=compute_$(arch),code=sm_$(arch))
   CUINC = -I$(CUDA_HOME)/include/
@@ -158,41 +179,79 @@ ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
     CURANDLIBFLAGS = -L$(CUDA_HOME)/lib64/ -lcurand # NB: -lcuda is not needed here!
   endif
   CUOPTFLAGS = -lineinfo
-  CUFLAGS = $(foreach opt, $(OPTFLAGS), -Xcompiler $(opt)) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
-  ###CUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
-  ###NVCC_VERSION = $(shell $(NVCC) --version | grep 'Cuda compilation tools' | cut -d' ' -f5 | cut -d, -f1)
-  CUFLAGS += -std=c++17 # need CUDA >= 11.2 (see #333): this is enforced in mgOnGpuConfig.h
+  ###GPUFLAGS = $(OPTFLAGS) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
+  GPUFLAGS = $(foreach opt, $(OPTFLAGS), -Xcompiler $(opt)) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
+  ###GPUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
+  ###GPUCC_VERSION = $(shell $(GPUCC) --version | grep 'Cuda compilation tools' | cut -d' ' -f5 | cut -d, -f1)
+  GPUFLAGS += -std=c++17 # need CUDA >= 11.2 (see #333): this is enforced in mgOnGpuConfig.h
   # Without -maxrregcount: baseline throughput: 6.5E8 (16384 32 12) up to 7.3E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 160 # improves throughput: 6.9E8 (16384 32 12) up to 7.7E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 128 # improves throughput: 7.3E8 (16384 32 12) up to 7.6E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 96 # degrades throughput: 4.1E8 (16384 32 12) up to 4.5E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 64 # degrades throughput: 1.7E8 (16384 32 12) flat at 1.7E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 160 # improves throughput: 6.9E8 (16384 32 12) up to 7.7E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 128 # improves throughput: 7.3E8 (16384 32 12) up to 7.6E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 96 # degrades throughput: 4.1E8 (16384 32 12) up to 4.5E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 64 # degrades throughput: 1.7E8 (16384 32 12) flat at 1.7E8 (65536 128 12)
+  CUBUILDRULEFLAGS = -Xcompiler -fPIC -c
+  CCBUILDRULEFLAGS = -Xcompiler -fPIC -c -x cu
+  CUDATESTFLAGS = -lcuda
+
+  # Set the host C++ compiler for GPUCC via "-ccbin <host-compiler>"
+  # (NB issue #505: this must be a single word, "clang++ --gcc-toolchain..." is not supported)
+  GPUFLAGS += -ccbin $(shell which $(subst ccache ,,$(CXX)))
+
+  # Allow newer (unsupported) C++ compilers with older versions of CUDA if ALLOW_UNSUPPORTED_COMPILER_IN_CUDA is set (#504)
+  ifneq ($(origin ALLOW_UNSUPPORTED_COMPILER_IN_CUDA),undefined)
+  GPUFLAGS += -allow-unsupported-compiler
+  endif
+
 else ifneq ($(origin REQUIRE_CUDA),undefined)
+
   # If REQUIRE_CUDA is set but no cuda is found, stop here (e.g. for CI tests on GPU #443)
-  $(error No cuda installation found (set CUDA_HOME or make nvcc visible in PATH))
+  $(error No cuda installation found (set CUDA_HOME or make GPUCC visible in PATH))
+
+#--- Option 2: CUDA does not exist, HIP exists -> use HIP
+
+# Set GPUCC as $(HIP_HOME)/bin/hipcc if it exists
+else ifneq ($(wildcard $(HIP_HOME)/bin/hipcc),)
+
+  GPUCC = $(HIP_HOME)/bin/hipcc
+  #USE_NVTX ?=-DUSE_NVTX # should maybe find something equivalent to this in HIP?
+  HIPARCHFLAGS = -target x86_64-linux-gnu --offload-arch=gfx90a
+  HIPINC = -I$(HIP_HOME)/include/
+  # Note: -DHIP_FAST_MATH is equivalent to -use_fast_math in HIP 
+  # (but only for single precision line 208: https://rocm-developer-tools.github.io/HIP/hcc__detail_2math__functions_8h_source.html)
+  # Note: CUOPTFLAGS should not be used for HIP, it had been added here but was then removed (#808)
+  GPUFLAGS = $(OPTFLAGS) $(INCFLAGS) $(HIPINC) $(HIPARCHFLAGS) -DHIP_FAST_MATH -DHIP_PLATFORM=amd -fPIC
+  ###GPUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
+  GPUFLAGS += -std=c++17
+  ###GPUFLAGS+= --maxrregcount 255 # (AV: is this option valid on HIP and meaningful on AMD GPUs?)
+  CUBUILDRULEFLAGS = -fPIC -c
+  CCBUILDRULEFLAGS = -fPIC -c -x hip
+
+else ifneq ($(origin REQUIRE_HIP),undefined)
+
+  # If REQUIRE_HIP is set but no HIP is found, stop here (e.g. for CI tests on GPU #443)
+  $(error No hip installation found (set HIP_HOME or make GPUCC visible in PATH))
+
+#--- Option 3: CUDA does not exist, HIP does not exist -> switch off both CUDA and HIP
+
 else
-  # No cuda. Switch cuda compilation off and go to common random numbers in C++
+
+  # No cudacc and no hipcc: switch CUDA and HIP compilation off and go to common random numbers in C++
   $(warning CUDA_HOME is not set or is invalid: export CUDA_HOME to compile with cuda)
-  override NVCC=
+  $(warning HIP_HOME is not set or is invalid: export HIP_HOME to compile with hip)
+  override GPUCC=
   override USE_NVTX=
   override CUINC=
   override CURANDLIBFLAGS=
-endif
-export NVCC
-export CUFLAGS
 
-# Set the host C++ compiler for nvcc via "-ccbin <host-compiler>"
-# (NB issue #505: this must be a single word, "clang++ --gcc-toolchain..." is not supported)
-CUFLAGS += -ccbin $(shell which $(subst ccache ,,$(CXX)))
-
-# Allow newer (unsupported) C++ compilers with older versions of CUDA if ALLOW_UNSUPPORTED_COMPILER_IN_CUDA is set (#504)
-ifneq ($(origin ALLOW_UNSUPPORTED_COMPILER_IN_CUDA),undefined)
-CUFLAGS += -allow-unsupported-compiler
 endif
 
+# Export GPUCC (so that it can also be used in cudacpp_src.mk?)
+export GPUCC
+export GPUFLAGS
+
 #-------------------------------------------------------------------------------
 
-#=== Configure ccache for C++ and CUDA builds
+#=== Configure ccache for C++ and CUDA/HIP builds
 
 # Enable ccache if USECCACHE=1
 ifeq ($(USECCACHE)$(shell echo $(CXX) | grep ccache),1)
@@ -201,15 +260,15 @@ endif
 #ifeq ($(USECCACHE)$(shell echo $(AR) | grep ccache),1)
 #  override AR:=ccache $(AR)
 #endif
-ifneq ($(NVCC),)
-  ifeq ($(USECCACHE)$(shell echo $(NVCC) | grep ccache),1)
-    override NVCC:=ccache $(NVCC)
+ifneq ($(GPUCC),)
+  ifeq ($(USECCACHE)$(shell echo $(GPUCC) | grep ccache),1)
+    override GPUCC:=ccache $(GPUCC)
   endif
 endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure PowerPC-specific compiler flags for C++ and CUDA
+#=== Configure PowerPC-specific compiler flags for C++ and CUDA/HIP
 
 # PowerPC-specific CXX compiler flags (being reviewed)
 ifeq ($(UNAME_P),ppc64le)
@@ -225,9 +284,9 @@ else
   ######CXXFLAGS+= -fno-semantic-interposition # no benefit (neither alone, nor combined with -flto)
 endif
 
-# PowerPC-specific CUDA compiler flags (to be reviewed!)
+# PowerPC-specific CUDA/HIP compiler flags (to be reviewed!)
 ifeq ($(UNAME_P),ppc64le)
-  CUFLAGS+= -Xcompiler -mno-float128
+  GPUFLAGS+= -Xcompiler -mno-float128
 endif
 
 #-------------------------------------------------------------------------------
@@ -235,9 +294,11 @@ endif
 #=== Configure defaults and check if user-defined choices exist for OMPFLAGS, AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
 
 # Set the default OMPFLAGS choice
-ifneq ($(shell $(CXX) --version | egrep '^Intel'),)
+ifneq ($(findstring hipcc,$(GPUCC)),)
+override OMPFLAGS = # disable OpenMP MT when using hipcc #802
+else ifneq ($(shell $(CXX) --version | egrep '^Intel'),)
 override OMPFLAGS = -fopenmp
-###override OMPFLAGS = # disable OpenMP MT on Intel (was ok without nvcc but not ok with nvcc before #578)
+###override OMPFLAGS = # disable OpenMP MT on Intel (was ok without GPUCC but not ok with GPUCC before #578)
 else ifneq ($(shell $(CXX) --version | egrep '^(clang)'),)
 override OMPFLAGS = -fopenmp
 ###override OMPFLAGS = # disable OpenMP MT on clang (was not ok without or with nvcc before #578)
@@ -293,7 +354,10 @@ endif
 
 # Set the default RNDGEN (random number generator) choice
 ifeq ($(RNDGEN),)
-  ifeq ($(NVCC),)
+  ifeq ($(GPUCC),)
+    override RNDGEN = hasNoCurand
+  # Edgecase for HIP compilation
+  else ifeq ($(findstring hipcc,$(GPUCC)),hipcc)
     override RNDGEN = hasNoCurand
   else ifeq ($(RNDGEN),)
     override RNDGEN = hasCurand
@@ -310,7 +374,7 @@ export OMPFLAGS
 
 #-------------------------------------------------------------------------------
 
-#=== Set the CUDA/C++ compiler flags appropriate to user-defined choices of AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
+#=== Set the CUDA/HIP/C++ compiler flags appropriate to user-defined choices of AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
 
 # Set the build flags appropriate to OMPFLAGS
 $(info OMPFLAGS=$(OMPFLAGS))
@@ -368,13 +432,13 @@ CXXFLAGS+= $(AVXFLAGS)
 $(info FPTYPE=$(FPTYPE))
 ifeq ($(FPTYPE),d)
   CXXFLAGS += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
-  CUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
+  GPUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
 else ifeq ($(FPTYPE),f)
   CXXFLAGS += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
-  CUFLAGS  += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
+  GPUFLAGS  += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
 else ifeq ($(FPTYPE),m)
   CXXFLAGS += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
-  CUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
+  GPUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
 else
   $(error Unknown FPTYPE='$(FPTYPE)': only 'd', 'f' and 'm' are supported)
 endif
@@ -383,7 +447,7 @@ endif
 $(info HELINL=$(HELINL))
 ifeq ($(HELINL),1)
   CXXFLAGS += -DMGONGPU_INLINE_HELAMPS
-  CUFLAGS  += -DMGONGPU_INLINE_HELAMPS
+  GPUFLAGS  += -DMGONGPU_INLINE_HELAMPS
 else ifneq ($(HELINL),0)
   $(error Unknown HELINL='$(HELINL)': only '0' and '1' are supported)
 endif
@@ -392,7 +456,7 @@ endif
 $(info HRDCOD=$(HRDCOD))
 ifeq ($(HRDCOD),1)
   CXXFLAGS += -DMGONGPU_HARDCODE_PARAM
-  CUFLAGS  += -DMGONGPU_HARDCODE_PARAM
+  GPUFLAGS  += -DMGONGPU_HARDCODE_PARAM
 else ifneq ($(HRDCOD),0)
   $(error Unknown HRDCOD='$(HRDCOD)': only '0' and '1' are supported)
 endif
@@ -444,11 +508,11 @@ ifeq ($(UNAME_S),Darwin)
   override CULIBFLAGSRPATH2 =
 else
   # RPATH to cuda/cpp libs when linking executables
-  override CXXLIBFLAGSRPATH = -Wl,-rpath,$(LIBDIRRPATH)
-  override CULIBFLAGSRPATH = -Xlinker -rpath,$(LIBDIRRPATH)
+  override CXXLIBFLAGSRPATH = -Wl,-rpath=$(LIBDIRRPATH)
+  override CULIBFLAGSRPATH = -Xlinker -rpath=$(LIBDIRRPATH)
   # RPATH to common lib when linking cuda/cpp libs
-  override CXXLIBFLAGSRPATH2 = -Wl,-rpath,'$$ORIGIN'
-  override CULIBFLAGSRPATH2 = -Xlinker -rpath,'$$ORIGIN'
+  override CXXLIBFLAGSRPATH2 = -Wl,-rpath='$$ORIGIN'
+  override CULIBFLAGSRPATH2 = -Xlinker -rpath='$$ORIGIN'
 endif
 
 # Setting LD_LIBRARY_PATH or DYLD_LIBRARY_PATH in the RUNTIME is no longer necessary (neither on Linux nor on Mac)
@@ -461,7 +525,7 @@ override RUNTIME =
 cxx_main=$(BUILDDIR)/check.exe
 fcxx_main=$(BUILDDIR)/fcheck.exe
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 cu_main=$(BUILDDIR)/gcheck.exe
 fcu_main=$(BUILDDIR)/fgcheck.exe
 else
@@ -492,14 +556,15 @@ $(BUILDDIR)/.build.$(TAG):
 	@touch $(BUILDDIR)/.build.$(TAG)
 
 # Generic target and build rules: objects from CUDA compilation
-ifneq ($(NVCC),)
+# NB: CCBUILDRULEFLAGS includes "-x cu" for nvcc and "-x hip" for hipcc (#810)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/%%.o : %%.cu *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $(CUBUILDRULEFLAGS) $< -o $@
 
 $(BUILDDIR)/%%_cu.o : %%.cc *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c -x cu $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $(CCBUILDRULEFLAGS) $< -o $@
 endif
 
 # Generic target and build rules: objects from C++ compilation
@@ -509,11 +574,14 @@ $(BUILDDIR)/%%.o : %%.cc *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	$(CXX) $(CPPFLAGS) $(CXXFLAGS) -fPIC -c $< -o $@
 
 # Apply special build flags only to CrossSectionKernel.cc and gCrossSectionKernel.cu (no fast math, see #117 and #516)
+# Added edgecase for HIP compilation
 ifeq ($(shell $(CXX) --version | grep ^nvc++),)
 $(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS := $(filter-out -ffast-math,$(CXXFLAGS))
 $(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS += -fno-fast-math
-ifneq ($(NVCC),)
-$(BUILDDIR)/gCrossSectionKernels.o: CUFLAGS += -Xcompiler -fno-fast-math
+ifeq ($(findstring nvcc,$(GPUCC)),nvcc)
+  $(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -Xcompiler -fno-fast-math
+else
+  $(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -fno-fast-math
 endif
 endif
 
@@ -530,10 +598,10 @@ ifeq ($(RNDGEN),hasCurand)
 $(BUILDDIR)/CurandRandomNumberKernel.o: CXXFLAGS += $(CUINC)
 endif
 
-# Avoid "warning: builtin __has_trivial_... is deprecated; use __is_trivially_... instead" in nvcc with icx2023 (#592)
+# Avoid "warning: builtin __has_trivial_... is deprecated; use __is_trivially_... instead" in GPUCC with icx2023 (#592)
 ifneq ($(shell $(CXX) --version | egrep '^(Intel)'),)
-ifneq ($(NVCC),)
-CUFLAGS += -Xcompiler -Wno-deprecated-builtins
+ifneq ($(GPUCC),)
+GPUFLAGS += -Wno-deprecated-builtins
 endif
 endif
 
@@ -541,8 +609,8 @@ endif
 # This patch does remove the warning, but I prefer to keep it disabled for the moment...
 ###ifneq ($(shell $(CXX) --version | egrep '^(clang|Apple clang|Intel)'),)
 ###$(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS += -Wno-overriding-t-option
-###ifneq ($(NVCC),)
-###$(BUILDDIR)/gCrossSectionKernels.o: CUFLAGS += -Xcompiler -Wno-overriding-t-option
+###ifneq ($(GPUCC),)
+###$(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -Xcompiler -Wno-overriding-t-option
 ###endif
 ###endif
 
@@ -569,7 +637,7 @@ MG5AMC_CXXLIB = mg5amc_$(processid_short)_cpp
 cxx_objects_lib=$(BUILDDIR)/CPPProcess.o $(BUILDDIR)/MatrixElementKernels.o $(BUILDDIR)/BridgeKernels.o $(BUILDDIR)/CrossSectionKernels.o
 cxx_objects_exe=$(BUILDDIR)/CommonRandomNumberKernel.o $(BUILDDIR)/RamboSamplingKernels.o
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 MG5AMC_CULIB = mg5amc_$(processid_short)_cuda
 cu_objects_lib=$(BUILDDIR)/gCPPProcess.o $(BUILDDIR)/gMatrixElementKernels.o $(BUILDDIR)/gBridgeKernels.o $(BUILDDIR)/gCrossSectionKernels.o
 cu_objects_exe=$(BUILDDIR)/gCommonRandomNumberKernel.o $(BUILDDIR)/gRamboSamplingKernels.o
@@ -581,11 +649,17 @@ $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: cxx_objects_lib += $(BUILDDIR)/fbridge.o
 $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib)
 	$(CXX) -shared -o $@ $(cxx_objects_lib) $(CXXLIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: $(BUILDDIR)/fbridge_cu.o
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: cu_objects_lib += $(BUILDDIR)/fbridge_cu.o
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cu_objects_lib)
-	$(NVCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+# Bypass std::filesystem completely to ease portability on LUMI #803
+#ifneq ($(findstring hipcc,$(GPUCC)),)
+#	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB) -lstdc++fs
+#else
+#	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+#endif
 endif
 
 #-------------------------------------------------------------------------------
@@ -602,16 +676,16 @@ $(cxx_main): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PAT
 $(cxx_main): $(BUILDDIR)/check_sa.o $(LIBDIR)/lib$(MG5AMC_CXXLIB).so $(cxx_objects_exe) $(BUILDDIR)/CurandRandomNumberKernel.o
 	$(CXX) -o $@ $(BUILDDIR)/check_sa.o $(OMPFLAGS) -ldl -pthread $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe) $(BUILDDIR)/CurandRandomNumberKernel.o $(CURANDLIBFLAGS)
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(cu_main): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(cu_main): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(cu_main): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(cu_main): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 else ifneq ($(shell $(CXX) --version | grep ^nvc++),) # support nvc++ #531
 $(cu_main): LIBFLAGS += -L$(patsubst %%bin/nvc++,%%lib,$(subst ccache ,,$(CXX))) -lnvhpcatm -lnvcpumath -lnvc
 endif
 $(cu_main): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(cu_main): $(BUILDDIR)/gcheck_sa.o $(LIBDIR)/lib$(MG5AMC_CULIB).so $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o
-	$(NVCC) -o $@ $(BUILDDIR)/gcheck_sa.o $(CUARCHFLAGS) $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o $(CURANDLIBFLAGS)
+	$(GPUCC) -o $@ $(BUILDDIR)/gcheck_sa.o $(CUARCHFLAGS) $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o $(CURANDLIBFLAGS)
 endif
 
 #-------------------------------------------------------------------------------
@@ -635,19 +709,27 @@ $(fcxx_main): LIBFLAGS += -L$(shell dirname $(shell $(FC) --print-file-name libg
 endif
 $(fcxx_main): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(fcxx_main): $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler.o $(LIBDIR)/lib$(MG5AMC_CXXLIB).so $(cxx_objects_exe)
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(BUILDDIR)/fcheck_sa.o $(OMPFLAGS) $(BUILDDIR)/fsampler.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe) -lstdc++
+else
 	$(CXX) -o $@ $(BUILDDIR)/fcheck_sa.o $(OMPFLAGS) $(BUILDDIR)/fsampler.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe)
+endif
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(fcu_main): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(fcu_main): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(fcu_main): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(fcu_main): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 endif
 ifeq ($(UNAME_S),Darwin)
 $(fcu_main): LIBFLAGS += -L$(shell dirname $(shell $(FC) --print-file-name libgfortran.dylib)) # add path to libgfortran on Mac #375
 endif
 $(fcu_main): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(fcu_main): $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBDIR)/lib$(MG5AMC_CULIB).so $(cu_objects_exe)
-	$(NVCC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe)
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) -lstdc++ -L$(shell dirname $(shell $(GPUCC) -print-prog-name=clang))/../../lib -lamdhip64
+else
+	$(GPUCC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe)
+endif
 endif
 
 #-------------------------------------------------------------------------------
@@ -659,7 +741,7 @@ $(BUILDDIR)/testxxx.o: testxxx_cc_ref.txt
 $(testmain): $(BUILDDIR)/testxxx.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/testxxx.o # Comment out this line to skip the C++ test of xxx functions
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/testxxx_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/testxxx_cu.o: INCFLAGS += $(GTESTINC)
 $(BUILDDIR)/testxxx_cu.o: testxxx_cc_ref.txt
@@ -672,7 +754,7 @@ $(BUILDDIR)/testmisc.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/testmisc.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/testmisc.o # Comment out this line to skip the C++ miscellaneous tests
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/testmisc_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/testmisc_cu.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/testmisc_cu.o
@@ -684,12 +766,12 @@ $(BUILDDIR)/runTest.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/runTest.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/runTest.o
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/runTest_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/runTest_cu.o: INCFLAGS += $(GTESTINC)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(testmain): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(testmain): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(testmain): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(testmain): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 else ifneq ($(shell $(CXX) --version | grep ^nvc++),) # support nvc++ #531
 $(testmain): LIBFLAGS += -L$(patsubst %%bin/nvc++,%%lib,$(subst ccache ,,$(CXX))) -lnvhpcatm -lnvcpumath -lnvc
 endif
@@ -713,14 +795,23 @@ $(testmain): LIBFLAGS += -lgomp
 endif
 endif
 
-ifeq ($(NVCC),) # link only runTest.o
+# Bypass std::filesystem completely to ease portability on LUMI #803
+#ifneq ($(findstring hipcc,$(GPUCC)),)
+#$(testmain): LIBFLAGS += -lstdc++fs
+#endif
+
+ifeq ($(GPUCC),) # link only runTest.o
 $(testmain): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(testmain): $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib) $(cxx_objects_exe) $(GTESTLIBS)
 	$(CXX) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) -ldl -pthread $(LIBFLAGS)
 else # link both runTest.o and runTest_cu.o
 $(testmain): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(testmain): $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) $(GTESTLIBS)
-	$(NVCC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) -lcuda
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) $(CUDATESTFLAGS) -lstdc++ -lpthread  -L$(shell dirname $(shell $(GPUCC) -print-prog-name=clang))/../../lib -lamdhip64
+else
+	$(GPUCC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) $(CUDATESTFLAGS)
+endif
 endif
 
 # Use target gtestlibs to build only googletest
@@ -829,9 +920,9 @@ ifeq ($(USECCACHE),1)
 	ccache --version | head -1
 endif
 	@echo ""
-	@echo NVCC=$(NVCC)
-ifneq ($(NVCC),)
-	$(NVCC) --version
+	@echo GPUCC=$(GPUCC)
+ifneq ($(GPUCC),)
+	$(GPUCC) --version
 endif
 	@echo ""
 	@echo CXX=$(CXX)
@@ -850,7 +941,7 @@ endif
 
 # Target: check (run the C++ test executable)
 # [NB THIS IS WHAT IS USED IN THE GITHUB CI!]
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 check: runTest cmpFcheck cmpFGcheck
 else
 check: runTest cmpFcheck
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/cudacpp_src.mk b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/cudacpp_src.mk
index 25b6f8f7c8..49ccf0c4e3 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/cudacpp_src.mk
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/cudacpp_src.mk
@@ -19,7 +19,7 @@ SHELL := /bin/bash
 #=== Configure common compiler flags for CUDA and C++
 
 INCFLAGS = -I.
-OPTFLAGS = -O3 # this ends up in CUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
+OPTFLAGS = -O3 # this ends up in GPUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
 
 #-------------------------------------------------------------------------------
 
@@ -45,13 +45,13 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure the CUDA compiler (note: NVCC is already exported including ccache)
+#=== Configure the CUDA compiler (note: GPUCC is already exported including ccache)
 
-###$(info NVCC=$(NVCC))
+###$(info GPUCC=$(GPUCC))
 
 #-------------------------------------------------------------------------------
 
-#=== Configure ccache for C++ builds (note: NVCC is already exported including ccache)
+#=== Configure ccache for C++ builds (note: GPUCC is already exported including ccache)
 
 # Enable ccache if USECCACHE=1
 ifeq ($(USECCACHE)$(shell echo $(CXX) | grep ccache),1)
@@ -92,6 +92,13 @@ endif
 ###$(info OMPFLAGS=$(OMPFLAGS))
 CXXFLAGS += $(OMPFLAGS)
 
+# Add correct flags for nvcc (-x cu) and hipcc (-x hip) for GPU code (see #810)
+ifeq ($(findstring nvcc,$(GPUCC)),nvcc)
+  GPUFLAGS += -Xcompiler -fPIC -c -x cu
+else ifeq ($(findstring hipcc,$(GPUCC)),hipcc)
+  GPUFLAGS += -fPIC -c -x hip
+endif
+
 # Set the build flags appropriate to each AVX choice (example: "make AVX=none")
 # [NB MGONGPU_PVW512 is needed because "-mprefer-vector-width=256" is not exposed in a macro]
 # [See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96476]
@@ -253,20 +260,20 @@ $(BUILDDIR)/%%.o : %%.cc *.h $(BUILDDIR)/.build.$(TAG)
 # Generic target and build rules: objects from CUDA compilation
 $(BUILDDIR)/%%_cu.o : %%.cc *.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c -x cu $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $< -o $@
 
 #-------------------------------------------------------------------------------
 
 cxx_objects=$(addprefix $(BUILDDIR)/, Parameters_%(model)s.o read_slha.o)
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 cu_objects=$(addprefix $(BUILDDIR)/, Parameters_%(model)s_cu.o)
 endif
 
 # Target (and build rules): common (src) library
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so : $(cxx_objects) $(cu_objects)
 	@if [ ! -d $(LIBDIR) ]; then echo "mkdir -p $(LIBDIR)"; mkdir -p $(LIBDIR); fi
-	$(NVCC) -shared -o $@ $(cxx_objects) $(cu_objects) $(LDFLAGS)
+	$(GPUCC) -shared -o $@ $(cxx_objects) $(cu_objects) $(LDFLAGS)
 else
 $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so : $(cxx_objects)
 	@if [ ! -d $(LIBDIR) ]; then echo "mkdir -p $(LIBDIR)"; mkdir -p $(LIBDIR); fi
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/fbridge.cc b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/fbridge.cc
index 2d2b36d560..22ce3f5115 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/fbridge.cc
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/fbridge.cc
@@ -1,11 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Roiser (Oct 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "Bridge.h"
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 
 extern "C"
 {
@@ -22,7 +22,7 @@ extern "C"
    * Using the same Fortran MadEvent code, linking to the hetrerogeneous library would allow access to both CPU and GPU implementations.
    * The specific heterogeneous configuration (how many GPUs, how many threads on each CPU, etc) could be loaded in CUDA/C++ from a data file.
    */
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -46,8 +46,8 @@ extern "C"
    */
   void fbridgecreate_( CppObjectInFortran** ppbridge, const int* pnevtF, const int* pnparF, const int* pnp4F )
   {
-#ifdef __CUDACC__
-    CudaRuntime::setUp();
+#ifdef MGONGPUCPP_GPUIMPL
+    GpuRuntime::setUp();
 #endif
     // (NB: CPPProcess::initProc no longer needs to be executed here because it is called in the Bridge constructor)
     // FIXME: disable OMP in Bridge when called from Fortran
@@ -65,8 +65,8 @@ extern "C"
     Bridge<FORTRANFPTYPE>* pbridge = dynamic_cast<Bridge<FORTRANFPTYPE>*>( *ppbridge );
     if( pbridge == 0 ) throw std::runtime_error( "fbridgedelete_: invalid Bridge address" );
     delete pbridge;
-#ifdef __CUDACC__
-    CudaRuntime::tearDown();
+#ifdef MGONGPUCPP_GPUIMPL
+    GpuRuntime::tearDown();
 #endif
   }
 
@@ -96,7 +96,7 @@ extern "C"
   {
     Bridge<FORTRANFPTYPE>* pbridge = dynamic_cast<Bridge<FORTRANFPTYPE>*>( *ppbridge );
     if( pbridge == 0 ) throw std::runtime_error( "fbridgesequence_: invalid Bridge address" );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Use the device/GPU implementation in the CUDA library
     // (there is also a host implementation in this library)
     pbridge->gpu_sequence( momenta, gs, rndhel, rndcol, *pchannelId, mes, selhel, selcol );
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/fsampler.cc b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/fsampler.cc
index 2fb445372d..3743934f41 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/fsampler.cc
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/fsampler.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Feb 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "mgOnGpuConfig.h"
 
@@ -13,7 +13,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -40,7 +40,7 @@ namespace mg5amcCpu
   private:
     const int m_nevt; // The number of events in each iteration
     int m_iiter;      // The iteration counter (for random number seeding)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     HostBufferRndNumMomenta m_hstRndmom; // Memory buffers for random numbers
     HostBufferMomenta m_hstMomenta;      // Memory buffers for momenta
     HostBufferWeights m_hstWeights;      // Memory buffers for sampling weights
@@ -105,7 +105,7 @@ namespace mg5amcCpu
 
 extern "C"
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/mgOnGpuConfig.h b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/mgOnGpuConfig.h
index 1175622ff4..05013cf981 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/mgOnGpuConfig.h
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/mgOnGpuConfig.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jul 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUCONFIG_H
 #define MGONGPUCONFIG_H 1
@@ -10,12 +10,26 @@
 // There are two different code bases for standalone_cudacpp (without multichannel) and madevent+cudacpp (with multichannel)
 %(mgongpu_supports_multichannel)s
 
+// Is this a GPU (CUDA, HIP) or CPU implementation?
+#ifdef __CUDACC__
+#define MGONGPUCPP_GPUIMPL cuda
+#elif defined __HIPCC__
+#define MGONGPUCPP_GPUIMPL hip
+#include "hip/hip_runtime.h" // needed for blockDim, blockIdx, threadIdx: better in mgOnGpuConfig.h than in GpuAbstraction.h
+#else
+#undef MGONGPUCPP_GPUIMPL
+#endif
+
 // ** NB1 Throughputs (e.g. 6.8E8) are events/sec for "./gcheck.exe -p 65536 128 12"
 // ** NB2 Baseline on b7g47n0004 fluctuates (probably depends on load on other VMs)
 
 // Choose if curand is supported for generating random numbers
+// For HIP, by default, do not use curand (common random numbers will be used instead)
 // For both CUDA and C++, by default, do not inline, but allow this macro to be set from outside with e.g. -DMGONGPU_HAS_NO_CURAND
-// (there exist CUDA installations, e.g. using the HPC package, which do not include curand - see PR #784)
+// (there exist CUDA installations, e.g. using the HPC package, which do not include curand - see PR #784 and #785)
+#if defined __HIPCC__
+#define MGONGPU_HAS_NO_CURAND 1
+#else
 //#ifdef __CUDACC__
 //#undef MGONGPU_HAS_NO_CURAND // default
 ////#define MGONGPU_HAS_NO_CURAND 1
@@ -23,6 +37,7 @@
 //#undef MGONGPU_HAS_NO_CURAND // default
 ////#define MGONGPU_HAS_NO_CURAND 1
 //#endif
+#endif
 
 // Choose floating point precision (for everything but color algebra #537)
 // If one of these macros has been set from outside with e.g. -DMGONGPU_FPTYPE_FLOAT, nothing happens (issue #167)
@@ -54,23 +69,31 @@
 //#undef MGONGPU_HARDCODE_PARAM // default
 ////#define MGONGPU_HARDCODE_PARAM 1
 
-// Complex type in c++: std::complex or cxsmpl (CHOOSE ONLY ONE)
-#ifndef __CUDACC__
-//#define MGONGPU_CPPCXTYPE_STDCOMPLEX 1 // ~8 percent slower on float, same on double (5.1E6/double, 9.4E6/float)
-#define MGONGPU_CPPCXTYPE_CXSMPL 1 // new default (5.1E6/double, 10.2E6/float)
-#endif
-
-// Complex type in cuda: thrust or cucomplex or cxsmpl (CHOOSE ONLY ONE)
+// Complex type in CUDA: thrust or cucomplex or cxsmpl (CHOOSE ONLY ONE)
+// (NB THIS IS MGONGPU_*CU*CXTYPE_xxx)
 #ifdef __CUDACC__
 #define MGONGPU_CUCXTYPE_THRUST 1 // default (~1.15E9/double, ~3.2E9/float)
 //#define MGONGPU_CUCXTYPE_CUCOMPLEX 1 // ~10 percent slower (1.03E9/double, ~2.8E9/float)
 //#define MGONGPU_CUCXTYPE_CXSMPL 1 // ~10 percent slower (1.00E9/double, ~2.9E9/float)
+
+// Complex type in HIP: cxsmpl (ONLY ONE OPTION POSSIBLE? #810)
+// (NB THIS IS MGONGPU_*HIP*CXTYPE_xxx)
+#elif defined __HIPCC__
+#define MGONGPU_HIPCXTYPE_CXSMPL 1 // default for HIP
+
+// Complex type in C++: std::complex or cxsmpl (CHOOSE ONLY ONE)
+// (NB THIS IS MGONGPU_*CPP*CXTYPE_xxx)
+#else
+//#define MGONGPU_CPPCXTYPE_STDCOMPLEX 1 // ~8 percent slower on float, same on double (5.1E6/double, 9.4E6/float)
+#define MGONGPU_CPPCXTYPE_CXSMPL 1 // new default (5.1E6/double, 10.2E6/float)
 #endif
 
-// Cuda nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
+// CUDA nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
 #ifdef __CUDACC__
-#undef MGONGPU_NSIGHT_DEBUG // default
+#undef MGONGPU_NSIGHT_DEBUG // default in CUDA
 //#define MGONGPU_NSIGHT_DEBUG 1
+#else
+#undef MGONGPU_NSIGHT_DEBUG // only option in HIP or C++
 #endif
 
 // SANITY CHECKS (floating point precision for everything but color algebra #537)
@@ -86,17 +109,21 @@
 #error You cannot use double precision for color algebra and single precision elsewhere
 #endif
 
-// SANITY CHECKS (c++ complex number implementation)
-#ifndef __CUDACC__
-#if defined MGONGPU_CPPCXTYPE_STDCOMPLEX and defined MGONGPU_CPPCXTYPE_CXSMPL
-#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CPPCXTYPE_STDCOMPLEX or MGONGPU_CPPCXTYPE_CXSMPL
+// SANITY CHECKS (CUDA complex number implementation)
+#ifdef __CUDACC__
+#if defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CUCOMPLEX
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX for CUDA
+#elif defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CXSMPL for CUDA
+#elif defined MGONGPU_CUCXTYPE_CUCOMPLEX and defined MGONGPU_CUCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE OF MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL for CUDA
 #endif
 #endif
 
-// SANITY CHECKS (cuda complex number implementation)
-#ifdef __CUDACC__
-#if defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CUCOMPLEX and defined MGONGPU_CUCXTYPE_CXSMPL
-#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL
+// SANITY CHECKS (C++ complex number implementation)
+#ifndef MGONGPUCPP_GPUIMPL
+#if defined MGONGPU_CPPCXTYPE_STDCOMPLEX and defined MGONGPU_CPPCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CPPCXTYPE_STDCOMPLEX or MGONGPU_CPPCXTYPE_CXSMPL for C++
 #endif
 #endif
 
@@ -134,7 +161,7 @@ namespace mgOnGpu
   // Alignment requirement for using reinterpret_cast with SIMD vectorized code
   // (using reinterpret_cast with non aligned memory may lead to segmentation faults!)
   // Only needed for C++ code but can be enforced also in NVCC builds of C++ code using CUDA>=11.2 and C++17 (#318, #319, #333)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   constexpr int cppAlign = 64; // alignment requirement for SIMD vectorization (64-byte i.e. 512-bit)
 #endif
 
@@ -145,7 +172,7 @@ using mgOnGpu::fptype;
 using mgOnGpu::fptype2;
 
 // C++ SIMD vectorization width (this will be used to set neppV)
-#ifdef __CUDACC__ // CUDA implementation has no SIMD
+#ifdef MGONGPUCPP_GPUIMPL // CUDA and HIP implementations have no SIMD
 #undef MGONGPU_CPPSIMD
 #elif defined __AVX512VL__ && defined MGONGPU_PVW512 // C++ "512z" AVX512 with 512 width (512-bit ie 64-byte): 8 (DOUBLE) or 16 (FLOAT)
 #ifdef MGONGPU_FPTYPE_DOUBLE
@@ -175,9 +202,9 @@ using mgOnGpu::fptype2;
 #undef MGONGPU_CPPSIMD
 #endif
 
-// Cuda nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
+// CUDA nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
 // Arguments (not used so far): text is __FUNCTION__, code is 0 (start) or 1 (end)
-#if defined __CUDACC__ && defined MGONGPU_NSIGHT_DEBUG /* clang-format off */
+#if defined __CUDA__ && defined MGONGPU_NSIGHT_DEBUG /* clang-format off */
 #define mgDebugDeclare() __shared__ float mgDebugCounter[mgOnGpu::ntpbMAX];
 #define mgDebugInitialise() { mgDebugCounter[threadIdx.x] = 0; }
 #define mgDebug( code, text ) { mgDebugCounter[threadIdx.x] += 1; }
@@ -189,8 +216,8 @@ using mgOnGpu::fptype2;
 #define mgDebugFinalise() { /*noop*/ }
 #endif /* clang-format on */
 
-// Define empty CUDA declaration specifiers for C++
-#ifndef __CUDACC__
+// Define empty CUDA/HIP declaration specifiers for C++
+#ifndef MGONGPUCPP_GPUIMPL
 #define __global__
 #define __host__
 #define __device__
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/mgOnGpuCxtypes.h b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/mgOnGpuCxtypes.h
index ca9a9f00c0..7ede1dbfae 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/mgOnGpuCxtypes.h
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/mgOnGpuCxtypes.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022, based on earlier work by D. Smith) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUCXTYPES_H
 #define MGONGPUCXTYPES_H 1
@@ -19,7 +19,7 @@
 #include <complex>
 
 // Complex type in cuda: thrust or cucomplex or cxsmpl
-#ifdef __CUDACC__
+#ifdef __CUDACC__ // this must be __CUDAC__ (not MGONGPUCPP_GPUIMPL)
 #if defined MGONGPU_CUCXTYPE_THRUST
 #pragma clang diagnostic push
 #pragma clang diagnostic ignored "-Wtautological-compare" // for icpx2021/clang13 (https://stackoverflow.com/a/15864661)
@@ -30,8 +30,13 @@
 #elif not defined MGONGPU_CUCXTYPE_CXSMPL
 #error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL
 #endif
+// Complex type in HIP: cxsmpl
+#elif defined __HIPCC__
+#if not defined MGONGPU_HIPCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_HIPCXTYPE_CXSMPL
+#endif
 #else
-// Complex type in c++: std::complex or cxsmpl
+// Complex type in c++ or HIP: std::complex or cxsmpl
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
 #include <cmath>
 #elif not defined MGONGPU_CPPCXTYPE_CXSMPL
@@ -82,7 +87,7 @@ namespace mgOnGpu /* clang-format off */
 using mgOnGpu::cxsmpl;
 
 // Printout to stream for user defined types
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -92,7 +97,7 @@ namespace mg5amcCpu
   inline __host__ std::ostream&
   operator<<( std::ostream& out, const cxsmpl<FP>& c )
   {
-    out << std::complex( c.real(), c.imag() );
+    out << std::complex<FP>( c.real(), c.imag() );
     return out;
   }
 
@@ -215,14 +220,14 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
   // --- Type definitions (complex type: cxtype)
-#ifdef __CUDACC__ // cuda
+#ifdef __CUDACC__ // this must be __CUDAC__ (not MGONGPUCPP_GPUIMPL)
 #if defined MGONGPU_CUCXTYPE_THRUST
   typedef thrust::complex<fptype> cxtype;
 #elif defined MGONGPU_CUCXTYPE_CUCOMPLEX
@@ -255,13 +260,13 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
-#if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+#if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
 
   //------------------------------
   // CUDA or C++ - using cxsmpl
@@ -303,11 +308,11 @@ namespace mg5amcCpu
     return cxmake( c.real(), c.imag() );
   }
 
-#endif // #if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+#endif // #if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
 
   //==========================================================================
 
-#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_THRUST // cuda + thrust
+#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_THRUST // cuda + thrust (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // CUDA - using thrust::complex
@@ -347,7 +352,7 @@ namespace mg5amcCpu
 
   //==========================================================================
 
-#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_CUCOMPLEX // cuda + cucomplex
+#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_CUCOMPLEX // cuda + cucomplex (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // CUDA - using cuComplex
@@ -566,7 +571,7 @@ namespace mg5amcCpu
 
   //==========================================================================
 
-#if not defined __CUDACC__ and defined MGONGPU_CPPCXTYPE_STDCOMPLEX // c++ + stdcomplex
+#if not defined __CUDACC__ and defined MGONGPU_CPPCXTYPE_STDCOMPLEX // c++/hip + stdcomplex (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // C++ - using std::complex
@@ -633,7 +638,7 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/mgOnGpuFptypes.h b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/mgOnGpuFptypes.h
index 905c97d700..fa3a02664b 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/mgOnGpuFptypes.h
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/mgOnGpuFptypes.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUFPTYPES_H
 #define MGONGPUFPTYPES_H 1
@@ -12,7 +12,7 @@
 #include <cmath>
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL // cuda
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -20,7 +20,7 @@ namespace mg5amcCpu
 {
   //==========================================================================
 
-#ifdef __CUDACC__ // cuda
+#ifdef MGONGPUCPP_GPUIMPL // cuda
 
   //------------------------------
   // Floating point types - Cuda
@@ -64,11 +64,11 @@ namespace mg5amcCpu
 #endif
   }
 
-#endif // #ifdef __CUDACC__
+#endif // #ifdef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
   //------------------------------
   // Floating point types - C++
@@ -92,7 +92,7 @@ namespace mg5amcCpu
     return std::sqrt( f );
   }
 
-#endif // #ifndef __CUDACC__
+#endif // #ifndef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/mgOnGpuVectors.h b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/mgOnGpuVectors.h
index e1299ba81e..cdae04326b 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/mgOnGpuVectors.h
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/mgOnGpuVectors.h
@@ -32,7 +32,7 @@
 #endif
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -131,7 +131,7 @@ namespace mg5amcCpu
 #endif
 #endif
 
-#else // i.e #ifndef MGONGPU_CPPSIMD (this includes #ifdef __CUDACC__)
+#else // i.e #ifndef MGONGPU_CPPSIMD (this includes #ifdef MGONGPUCPP_GPUIMPL)
 
   const int neppV = 1;
 
@@ -153,13 +153,13 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
   // Printout to stream for user defined types
 
@@ -805,11 +805,11 @@ namespace mg5amcCpu
 
 #endif // #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
 
-#endif // #ifndef __CUDACC__
+#endif // #ifndef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
   //------------------------------
   // Vector types - CUDA
@@ -853,12 +853,12 @@ namespace mg5amcCpu
     return mask;
   }
 
-#endif // #ifdef __CUDACC__
+#endif // #ifdef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
   // Scalar-or-vector types: scalar in CUDA, vector or scalar in C++
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   typedef bool bool_sv;
   typedef fptype fptype_sv;
   typedef fptype2 fptype2_sv;
@@ -879,7 +879,7 @@ namespace mg5amcCpu
 #endif
 
   // Scalar-or-vector zeros: scalar in CUDA, vector or scalar in C++
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   inline __host__ __device__ cxtype cxzero_sv(){ return cxtype( 0, 0 ); }
 #elif defined MGONGPU_CPPSIMD
   inline cxtype_v cxzero_sv() { return cxtype_v(); } // RRRR=0000 IIII=0000
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/process_cc.inc b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/process_cc.inc
index 778e210468..815fd8d5b7 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/process_cc.inc
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/process_cc.inc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 %(info_lines)s
@@ -14,7 +14,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 %(hel_amps_h)s
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/process_function_definitions.inc b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/process_function_definitions.inc
index 5f1ea36b9e..9d024183db 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/process_function_definitions.inc
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/process_function_definitions.inc
@@ -4,13 +4,13 @@
 ! Copyright (C) 2020-2023 CERN and UCLouvain.
 ! Licensed under the GNU Lesser General Public License (version 3 or later).
 ! Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-! Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+! Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 !==========================================================================
 //==========================================================================
 // Class member functions for calculating the matrix elements for
 %(process_lines)s
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -44,7 +44,7 @@ namespace mg5amcCpu
   %(cipdhrdcod)s
   %(cipchrdcod)s
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   %(cipddevice)s
   %(cipcdevice)s
 #else
@@ -54,7 +54,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -80,8 +80,8 @@ namespace mg5amcCpu
     // Helicities for the process [NB do keep 'static' for this constexpr array, see issue #283]
     // *** NB There is no automatic check yet that these are in the same order as Fortran! #569 ***
 %(all_helicities)s
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -117,7 +117,7 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     %(cipdassign)s
     %(cipcassign)s
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     %(cipd2tipdSym)s
     %(cipc2tipcSym)s
 #else
@@ -150,7 +150,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -179,6 +179,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -192,6 +196,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -215,12 +221,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -241,7 +247,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -367,9 +373,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -393,7 +399,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -413,7 +419,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { %(den_factors)s }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/process_h.inc b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/process_h.inc
index 893f7f3215..2c3adf57e2 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/process_h.inc
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/process_h.inc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 %(info_lines)s
@@ -23,7 +23,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -32,7 +32,7 @@ namespace mg5amcCpu
 %(process_class_definitions)s
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -45,7 +45,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -75,7 +75,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/process_matrix.inc b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/process_matrix.inc
index 1e473edcf8..960f029d8d 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/process_matrix.inc
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/process_matrix.inc
@@ -4,7 +4,7 @@
 ! Copyright (C) 2020-2023 CERN and UCLouvain.
 ! Licensed under the GNU Lesser General Public License (version 3 or later).
 ! Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-! Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+! Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 !==========================================================================
 
       // *** COLOR CHOICE BELOW ***
@@ -17,7 +17,7 @@
       // (This method used to be called %(process_class_name)s::matrix_%(proc_name)s(%(matrix_args)s)?)
 %(color_matrix_lines)s
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -74,7 +74,7 @@
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -133,7 +133,7 @@
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%%6d ihel=%%2d me_running=%%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/process_sigmaKin_function.inc b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/process_sigmaKin_function.inc
index 159e3d8d5d..66450ae367 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/process_sigmaKin_function.inc
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/process_sigmaKin_function.inc
@@ -4,11 +4,14 @@
 ! Copyright (C) 2020-2023 CERN and UCLouvain.
 ! Licensed under the GNU Lesser General Public License (version 3 or later).
 ! Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-! Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+! Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 !==========================================================================
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -36,7 +39,7 @@
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -246,7 +249,7 @@
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/rambo.h b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/rambo.h
index e02ea52496..cd7e1008ea 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/rambo.h
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/rambo.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -18,7 +18,7 @@
 #include <iostream>
 
 // Simplified rambo version for 2 to N (with N>=2) processes with massless particles
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +83,7 @@ namespace mg5amcCpu
       static bool first = true;
       if( first )
       {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
         if constexpr( M_ACCESS::isOnDevice() ) // avoid
         {
           const int ievt0 = 0;
@@ -166,7 +166,7 @@ namespace mg5amcCpu
     wt = po2log;
     if( nparf != 2 ) wt = ( 2. * nparf - 4. ) * log( energy ) + z[nparf - 1];
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     // issue warnings if weight is too small or too large
     static int iwarn[5] = { 0, 0, 0, 0, 0 };
     if( wt < -180. )
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/runTest.cc b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/runTest.cc
index d4a760a71b..de327f2321 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/runTest.cc
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/runTest.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Hageboeck (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*xxx to run only testxxx.cc tests
 //----------------------------------------------------------------------------
@@ -18,7 +18,7 @@
 #include "RandomNumberKernels.h"
 #include "epoch_process_id.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using namespace mg5amcGpu;
 #else
 using namespace mg5amcCpu;
@@ -35,7 +35,7 @@ struct CUDA_CPU_TestBase : public TestDriverBase
     : TestDriverBase( npar, refFileName ) {}
 };
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 struct CPUTest : public CUDA_CPU_TestBase
 {
   // Struct data members (process, and memory structures for random numbers, momenta, matrix elements and weights on host and device)
@@ -119,7 +119,7 @@ struct CPUTest : public CUDA_CPU_TestBase
 };
 #endif
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 struct CUDATest : public CUDA_CPU_TestBase
 {
   // Reset the device when our test goes out of scope. Note that this should happen after
@@ -128,7 +128,7 @@ struct CUDATest : public CUDA_CPU_TestBase
   {
     ~DeviceReset()
     {
-      checkCuda( cudaDeviceReset() ); // this is needed by cuda-memcheck --leak-check full
+      checkGpu( gpuDeviceReset() ); // this is needed by cuda-memcheck --leak-check full
     }
   } deviceResetter;
 
@@ -256,7 +256,7 @@ INSTANTIATE_TEST_SUITE_P( prefix, \
                           test_suite_name, \
                           testing::Values( new CUDATest( MG_EPOCH_REFERENCE_FILE_NAME ) ) );
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 MG_INSTANTIATE_TEST_SUITE_GPU( XTESTID_GPU( MG_EPOCH_PROCESS_ID ), MadgraphTest );
 #else
 MG_INSTANTIATE_TEST_SUITE_CPU( XTESTID_CPU( MG_EPOCH_PROCESS_ID ), MadgraphTest );
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/testmisc.cc b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/testmisc.cc
index 895d6eeb56..ba9e59a8a3 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/testmisc.cc
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/testmisc.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*misc to run only testmisc.cc tests
 //----------------------------------------------------------------------------
@@ -17,7 +17,7 @@
 #include <sstream>
 #include <typeinfo>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #define TESTID( s ) s##_GPU_MISC
 #else
 #define TESTID( s ) s##_CPU_MISC
@@ -26,7 +26,7 @@
 #define XTESTID( s ) TESTID( s )
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -59,7 +59,7 @@ namespace mg5amcCpu
 
 TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testmisc )
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/testxxx.cc b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/testxxx.cc
index 2d1578cb43..dbe151e990 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/testxxx.cc
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/testxxx.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Apr 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*xxx to run only testxxx.cc tests
 //----------------------------------------------------------------------------
@@ -24,7 +24,7 @@
 #include <iomanip>
 #include <iostream>
 #include <vector>
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #define TESTID( s ) s##_GPU_XXX
 #else
 #define TESTID( s ) s##_CPU_XXX
@@ -32,7 +32,7 @@
 
 #define XTESTID( s ) TESTID( s )
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -42,7 +42,7 @@ namespace mg5amcCpu
   int FPEhandlerIevt = -1;
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU): '" << FPEhandlerMessage << "' ievt=" << FPEhandlerIevt << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU neppV=" << neppV << "): '" << FPEhandlerMessage << "' ievt=" << FPEhandlerIevt << std::endl;
@@ -53,7 +53,7 @@ namespace mg5amcCpu
 
 TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -77,7 +77,7 @@ TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
   assert( nevt %% neppM == 0 ); // nevt must be a multiple of neppM
   assert( nevt %% neppV == 0 ); // nevt must be a multiple of neppV
   // Fill in the input momenta
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   mg5amcGpu::PinnedHostBufferMomenta hstMomenta( nevt ); // AOSOA[npagM][npar=4][np4=4][neppM]
 #else
   mg5amcCpu::HostBufferMomenta hstMomenta( nevt ); // AOSOA[npagM][npar=4][np4=4][neppM]
@@ -322,7 +322,7 @@ TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
   {
     for( int ievt = 0; ievt < nevt; ievt++ )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       using namespace mg5amcGpu;
 #else
       using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/read_slha.cc b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/read_slha.cc
index 055b19a779..f8e46f2e66 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/read_slha.cc
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/read_slha.cc
@@ -11,7 +11,11 @@
 
 #include <algorithm>
 #include <cstdlib>
-#include <filesystem>
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
 #include <fstream>
 #include <iostream>
 
@@ -60,7 +64,14 @@ SLHAReader::read_slha_file( std::string file_name, bool verbose )
     {
       std::cout << "WARNING! Card file '" << file_name << "' does not exist:"
                 << " look for the file in directory $" << envpath << "='" << getenv( envpath ) << "'" << std::endl;
+      /*
+#ifdef __HIPCC__
+      const std::string file_name2 = std::experimental::filesystem::path( getenv( envpath ) ) / std::experimental::filesystem::path( file_name ).filename();
+#else
       const std::string file_name2 = std::filesystem::path( getenv( envpath ) ) / std::filesystem::path( file_name ).filename();
+#endif
+      */
+      const std::string file_name2 = std::string( getenv( envpath ) ) + "/" + file_name; // bypass std::filesystem #803
       param_card.open( file_name2.c_str(), std::ifstream::in );
       if( param_card.good() )
       {
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/model_handling.py b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/model_handling.py
index 83b61a9565..3e0ebe545f 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/model_handling.py
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/model_handling.py
@@ -1,7 +1,7 @@
 # Copyright (C) 2020-2023 CERN and UCLouvain.
 # Licensed under the GNU Lesser General Public License (version 3 or later).
 # Created by: O. Mattelaer (Sep 2021) for the MG5aMC CUDACPP plugin.
-# Further modified by: O. Mattelaer, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+# Further modified by: O. Mattelaer, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 import os
 
@@ -1110,7 +1110,7 @@ def get_process_function_definitions(self, write=True):
                                          %(len(coupling_indep), ' ), cxmake( m_pars->'.join(coupling_indep)) # AV only indep!
             replace_dict['cipcdevice'] = '__device__ __constant__ fptype cIPC[%i];'%(2*len(coupling_indep))
             replace_dict['cipcstatic'] = 'static fptype cIPC[%i];'%(2*len(coupling_indep))
-            replace_dict['cipc2tipcSym'] = 'checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, %i * sizeof( cxtype ) ) );'%len(coupling_indep)
+            replace_dict['cipc2tipcSym'] = 'gpuMemcpyToSymbol( cIPC, tIPC, %i * sizeof( cxtype ) );'%len(coupling_indep)
             replace_dict['cipc2tipc'] = 'memcpy( cIPC, tIPC, %i * sizeof( cxtype ) );'%len(coupling_indep)
             replace_dict['cipcdump'] = '\n    //for ( i=0; i<%i; i++ ) std::cout << std::setprecision(17) << "tIPC[i] = " << tIPC[i] << std::endl;'%len(coupling_indep)
             coup_str_hrd = '__device__ const fptype cIPC[%s] = { ' % (len(coupling_indep)*2)
@@ -1121,7 +1121,7 @@ def get_process_function_definitions(self, write=True):
             replace_dict['cipcassign'] = '//const cxtype tIPC[0] = { ... }; // nicoup=0'
             replace_dict['cipcdevice'] = '__device__ __constant__ fptype* cIPC = nullptr; // unused as nicoup=0'
             replace_dict['cipcstatic'] = 'static fptype* cIPC = nullptr; // unused as nicoup=0'
-            replace_dict['cipc2tipcSym'] = '//checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, %i * sizeof( cxtype ) ) ); // nicoup=0'%len(coupling_indep)
+            replace_dict['cipc2tipcSym'] = '//gpuMemcpyToSymbol( cIPC, tIPC, %i * sizeof( cxtype ) ); // nicoup=0'%len(coupling_indep)
             replace_dict['cipc2tipc'] = '//memcpy( cIPC, tIPC, %i * sizeof( cxtype ) ); // nicoup=0'%len(coupling_indep)
             replace_dict['cipcdump'] = ''
             replace_dict['cipchrdcod'] = '__device__ const fptype* cIPC = nullptr; // unused as nicoup=0'
@@ -1130,7 +1130,7 @@ def get_process_function_definitions(self, write=True):
                                          %(len(params), ', (fptype)m_pars->'.join(params))
             replace_dict['cipddevice'] = '__device__ __constant__ fptype cIPD[%i];'%(len(params))
             replace_dict['cipdstatic'] = 'static fptype cIPD[%i];'%(len(params))
-            replace_dict['cipd2tipdSym'] = 'checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, %i * sizeof( fptype ) ) );'%len(params)
+            replace_dict['cipd2tipdSym'] = 'gpuMemcpyToSymbol( cIPD, tIPD, %i * sizeof( fptype ) );'%len(params)
             replace_dict['cipd2tipd'] = 'memcpy( cIPD, tIPD, %i * sizeof( fptype ) );'%len(params)
             replace_dict['cipddump'] = '\n    //for ( i=0; i<%i; i++ ) std::cout << std::setprecision(17) << "tIPD[i] = " << tIPD[i] << std::endl;'%len(params)
             param_str_hrd = '__device__ const fptype cIPD[%s] = { ' % len(params)
@@ -1141,7 +1141,7 @@ def get_process_function_definitions(self, write=True):
             replace_dict['cipdassign'] = '//const fptype tIPD[0] = { ... }; // nparam=0'
             replace_dict['cipddevice'] = '//__device__ __constant__ fptype* cIPD = nullptr; // unused as nparam=0'
             replace_dict['cipdstatic'] = '//static fptype* cIPD = nullptr; // unused as nparam=0'
-            replace_dict['cipd2tipdSym'] = '//checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, %i * sizeof( fptype ) ) ); // nparam=0'%len(params)
+            replace_dict['cipd2tipdSym'] = '//gpuMemcpyToSymbol( cIPD, tIPD, %i * sizeof( fptype ) ); // nparam=0'%len(params)
             replace_dict['cipd2tipd'] = '//memcpy( cIPD, tIPD, %i * sizeof( fptype ) ); // nparam=0'%len(params)
             replace_dict['cipddump'] = ''
             replace_dict['cipdhrdcod'] = '//__device__ const fptype* cIPD = nullptr; // unused as nparam=0'
@@ -1219,13 +1219,13 @@ def get_all_sigmaKin_lines(self, color_amplitudes, class_name):
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -1252,7 +1252,7 @@ def get_all_sigmaKin_lines(self, color_amplitudes, class_name):
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( \"calculate_wavefunctions: ihel=%2d\\n\", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( \"calculate_wavefunctions: ievt00=%d\\n\", ievt00 );
 #endif""")
             nwavefuncs = self.matrix_elements[0].get_number_of_wavefunctions()
@@ -1289,7 +1289,7 @@ def get_all_sigmaKin_lines(self, color_amplitudes, class_name):
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif""")
             ret_lines += helas_calls
@@ -1718,8 +1718,10 @@ def super_get_matrix_element_calls(self, matrix_element, color_amplitudes, multi
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -1835,7 +1837,7 @@ def get_external(self, wf, argument):
             split_line2 = [ str.lstrip(' ').rstrip(' ') for str in split_line2] # AV
             split_line2.insert(2, '0') # add parameter fmass=0
             line2 = ', '.join(split_line2)
-            text = '#if not( defined __CUDACC__ and defined MGONGPU_TEST_DIVERGENCE )\n      %s\n#else\n      if( ( blockDim.x * blockIdx.x + threadIdx.x ) %% 2 == 0 )\n        %s\n      else\n        %s\n#endif\n' # AV
+            text = '#if not( defined MGONGPUCPP_GPUIMPL and defined MGONGPU_TEST_DIVERGENCE )\n      %s\n#else\n      if( ( blockDim.x * blockIdx.x + threadIdx.x ) %% 2 == 0 )\n        %s\n      else\n        %s\n#endif\n' # AV
             return text % (line, line, line2)
         text = '%s\n' # AV
         return text % line
diff --git a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/output.py b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/output.py
index 5267141530..c89295c01f 100644
--- a/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/output.py
+++ b/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/output.py
@@ -1,7 +1,7 @@
 # Copyright (C) 2020-2023 CERN and UCLouvain.
 # Licensed under the GNU Lesser General Public License (version 3 or later).
 # Created by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-# Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2021-2023) for the MG5aMC CUDACPP plugin.
+# Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2021-2023) for the MG5aMC CUDACPP plugin.
 
 import os
 import subprocess
@@ -88,9 +88,9 @@ class PLUGIN_ProcessExporter(PLUGIN_export_cpp.ProcessExporterGPU):
                      'CMake': [s+'CMake/Compilers.txt', s+'CMake/Platforms.txt', s+'CMake/Macros.txt'],
                      'src': [s+'gpu/rambo.h', s+'read_slha.h', s+'read_slha.cc',
                              s+'gpu/mgOnGpuFptypes.h', s+'gpu/mgOnGpuCxtypes.h', s+'gpu/mgOnGpuVectors.h',
-                             s+'CMake/src/CMakeLists.txt'],
+                             s+'CMake/src/CMakeLists.txt' ],
                      'SubProcesses': [s+'gpu/nvtx.h', s+'gpu/timer.h', s+'gpu/timermap.h',
-                                      s+'gpu/ompnumthreads.h', s+'gpu/CudaRuntime.h',
+                                      s+'gpu/ompnumthreads.h', s+'gpu/GpuRuntime.h', s+'gpu/GpuAbstraction.h',
                                       s+'gpu/MemoryAccessHelpers.h', s+'gpu/MemoryAccessVectors.h',
                                       s+'gpu/MemoryAccessMatrixElements.h', s+'gpu/MemoryAccessMomenta.h',
                                       s+'gpu/MemoryAccessRandomNumbers.h', s+'gpu/MemoryAccessWeights.h',
@@ -111,7 +111,7 @@ class PLUGIN_ProcessExporter(PLUGIN_export_cpp.ProcessExporterGPU):
                                       s+'CMake/SubProcesses/CMakeLists.txt'],
                      'test': [s+'gpu/cudacpp_test.mk']}
     to_link_in_P = ['nvtx.h', 'timer.h', 'timermap.h',
-                    'ompnumthreads.h', 'CudaRuntime.h',
+                    'ompnumthreads.h', 'GpuRuntime.h', 'GpuAbstraction.h',
                     'MemoryAccessHelpers.h', 'MemoryAccessVectors.h',
                     'MemoryAccessMatrixElements.h', 'MemoryAccessMomenta.h',
                     'MemoryAccessRandomNumbers.h', 'MemoryAccessWeights.h',
diff --git a/epochX/cudacpp/ee_mumu.mad/CODEGEN_mad_ee_mumu_log.txt b/epochX/cudacpp/ee_mumu.mad/CODEGEN_mad_ee_mumu_log.txt
index 36b42987c5..a484a3ce73 100644
--- a/epochX/cudacpp/ee_mumu.mad/CODEGEN_mad_ee_mumu_log.txt
+++ b/epochX/cudacpp/ee_mumu.mad/CODEGEN_mad_ee_mumu_log.txt
@@ -52,7 +52,7 @@ Note that you can still compile and run aMC@NLO with the built-in PDFs
 
 Using default text editor "vi". Set another one in ./input/mg5_configuration.txt
 Using default eps viewer "evince". Set another one in ./input/mg5_configuration.txt
-No valid web browser found. Please set in ./input/mg5_configuration.txt
+Using default web browser "firefox". Set another one in ./input/mg5_configuration.txt
 import /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_mad_ee_mumu.mg
 The import format was not given, so we guess it as command
 set stdout_level DEBUG
@@ -62,7 +62,7 @@ generate e+ e- > mu+ mu-
 No model currently active, so we import the Standard Model
 INFO: load particles 
 INFO: load vertices 
-DEBUG: model prefixing  takes 0.005498409271240234 
+DEBUG: model prefixing  takes 0.0055043697357177734 
 INFO: Restrict model sm with file models/sm/restrict_default.dat . 
 DEBUG: Simplifying conditional expressions 
 DEBUG: remove interactions: u s w+ at order: QED=1 
@@ -174,7 +174,7 @@ INFO: Generating Helas calls for process: e+ e- > mu+ mu- WEIGHTED<=4 @1
 INFO: Processing color information for process: e+ e- > mu+ mu- @1 
 INFO: Creating files in directory P1_epem_mupmum 
 DEBUG:  kwargs[prefix] = 0 [model_handling.py at line 1058] 
-DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7f5f7935eb80> [export_v4.py at line 6262] 
+DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7fe0c4707b20> [export_v4.py at line 6262] 
 INFO: Creating files in directory . 
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.h
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.cc
@@ -191,19 +191,19 @@ INFO: Created files CPPProcess.h and CPPProcess.cc in directory ./.
 INFO: Generating Feynman diagrams for Process: e+ e- > mu+ mu- WEIGHTED<=4 @1 
 INFO: Finding symmetric diagrams for subprocess group epem_mupmum 
 Generated helas calls for 1 subprocesses (2 diagrams) in 0.004 s
-Wrote files for 8 helas calls in 0.102 s
+Wrote files for 8 helas calls in 0.098 s
 ALOHA: aloha starts to compute helicity amplitudes
 ALOHA: aloha creates FFV1 routines
 ALOHA: aloha creates FFV2 routines
 ALOHA: aloha creates FFV4 routines
-ALOHA: aloha creates 3 routines in  0.203 s
+ALOHA: aloha creates 3 routines in  0.198 s
 DEBUG:  Entering PLUGIN_ProcessExporter.convert_model (create the model) [output.py at line 202] 
 ALOHA: aloha starts to compute helicity amplitudes
 ALOHA: aloha creates FFV1 routines
 ALOHA: aloha creates FFV2 routines
 ALOHA: aloha creates FFV4 routines
 ALOHA: aloha creates FFV2_4 routines
-ALOHA: aloha creates 7 routines in  0.260 s
+ALOHA: aloha creates 7 routines in  0.252 s
 <class 'aloha.create_aloha.AbstractRoutine'> FFV1
 <class 'aloha.create_aloha.AbstractRoutine'> FFV1
 <class 'aloha.create_aloha.AbstractRoutine'> FFV2
@@ -248,9 +248,9 @@ Type "launch" to generate events from this process, or see
 Run "open index.html" to see more information about this process.
 quit
 
-real	0m1.900s
-user	0m1.697s
-sys	0m0.195s
+real	0m1.882s
+user	0m1.658s
+sys	0m0.191s
 Code generation completed in 2 seconds
 ************************************************************
 *                                                          *
@@ -277,7 +277,7 @@ INFO: load configuration from /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/mg5amc
 INFO: load configuration from /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_mad_ee_mumu/Cards/me5_configuration.txt  
 Using default text editor "vi". Set another one in ./input/mg5_configuration.txt
 Using default eps viewer "evince". Set another one in ./input/mg5_configuration.txt
-No valid web browser found. Please set in ./input/mg5_configuration.txt
+Using default web browser "firefox". Set another one in ./input/mg5_configuration.txt
 treatcards run
 quit
 INFO:  
@@ -307,7 +307,7 @@ INFO: load configuration from /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/mg5amc
 INFO: load configuration from /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_mad_ee_mumu/Cards/me5_configuration.txt  
 Using default text editor "vi". Set another one in ./input/mg5_configuration.txt
 Using default eps viewer "evince". Set another one in ./input/mg5_configuration.txt
-No valid web browser found. Please set in ./input/mg5_configuration.txt
+Using default web browser "firefox". Set another one in ./input/mg5_configuration.txt
 treatcards param
 quit
 INFO:  
diff --git a/epochX/cudacpp/ee_mumu.mad/COPYRIGHT b/epochX/cudacpp/ee_mumu.mad/COPYRIGHT
index a134b5fef9..9036d9260a 100644
--- a/epochX/cudacpp/ee_mumu.mad/COPYRIGHT
+++ b/epochX/cudacpp/ee_mumu.mad/COPYRIGHT
@@ -15,6 +15,7 @@ The full development team currently includes the following authors :
   Stephan Hageboeck (CERN)
   Olivier Mattelaer (Universite Catholique de Louvain, original author)
   Stefan Roiser (CERN, original author)
+  Jorgen Teig (CERN)
   Andrea Valassi (CERN, original author)
   Zenny Wettersten (CERN)
 See https://github.com/madgraph5/madgraph4gpu for more details. For the full
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/Bridge.h b/epochX/cudacpp/ee_mumu.mad/SubProcesses/Bridge.h
index bf8b5e024d..f9ed70dfde 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/Bridge.h
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/Bridge.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Roiser (Nov 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef BRIDGE_H
 #define BRIDGE_H 1
@@ -14,16 +14,23 @@
 #include "MemoryAccessMomenta.h"  // for MemoryAccessMomenta::neppM
 #include "MemoryBuffers.h"        // for HostBufferMomenta, DeviceBufferMomenta etc
 
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
+
+#include <sys/stat.h> // bypass std::filesystem #803
+
 #include <algorithm>
 #include <cassert>
 #include <cmath>
 #include <cstring>
-#include <filesystem>
 #include <iostream>
 #include <memory>
 #include <type_traits>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +90,7 @@ namespace mg5amcCpu
     Bridge& operator=( const Bridge& ) = delete;
     Bridge& operator=( Bridge&& ) = delete;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     /**
      * Set the gpublocks and gputhreads for the gpusequence - throws if evnt != gpublocks*gputhreads
      * (this is needed for BridgeKernel tests rather than for actual production use in Fortran)
@@ -150,7 +157,7 @@ namespace mg5amcCpu
     unsigned int m_nevt; // number of events
     int m_nGoodHel;      // the number of good helicities (-1 initially when they have not yet been calculated)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     int m_gputhreads; // number of gpu threads (default set from number of events, can be modified)
     int m_gpublocks;  // number of gpu blocks (default set from number of events, can be modified)
     DeviceBuffer<FORTRANFPTYPE, sizePerEventMomenta> m_devMomentaF;
@@ -187,12 +194,12 @@ namespace mg5amcCpu
   // Forward declare transposition methods
   //
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
   template<typename Tin, typename Tout>
   __global__ void dev_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt );
 
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
 
   template<typename Tin, typename Tout>
   void hst_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt );
@@ -209,7 +216,7 @@ namespace mg5amcCpu
   Bridge<FORTRANFPTYPE>::Bridge( unsigned int nevtF, unsigned int nparF, unsigned int np4F )
     : m_nevt( nevtF )
     , m_nGoodHel( -1 )
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     , m_gputhreads( 256 )                  // default number of gpu threads
     , m_gpublocks( m_nevt / m_gputhreads ) // this ensures m_nevt <= m_gpublocks*m_gputhreads
     , m_devMomentaF( m_nevt )
@@ -233,7 +240,7 @@ namespace mg5amcCpu
   {
     if( nparF != CPPProcess::npar ) throw std::runtime_error( "Bridge constructor: npar mismatch" );
     if( np4F != CPPProcess::np4 ) throw std::runtime_error( "Bridge constructor: np4 mismatch" );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( ( m_nevt < s_gputhreadsmin ) || ( m_nevt % s_gputhreadsmin != 0 ) )
       throw std::runtime_error( "Bridge constructor: nevt should be a multiple of " + std::to_string( s_gputhreadsmin ) );
     while( m_nevt != m_gpublocks * m_gputhreads )
@@ -249,20 +256,28 @@ namespace mg5amcCpu
 #else
     std::cout << "WARNING! Instantiate host Bridge (nevt=" << m_nevt << ")" << std::endl;
     m_pmek.reset( new MatrixElementKernelHost( m_hstMomentaC, m_hstGs, m_hstRndHel, m_hstRndCol, m_hstMEs, m_hstSelHel, m_hstSelCol, m_nevt ) );
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
     // Create a process object, read param card and set parameters
     // FIXME: the process instance can happily go out of scope because it is only needed to read parameters?
     // FIXME: the CPPProcess should really be a singleton? what if fbridgecreate is called from several Fortran threads?
     CPPProcess process( /*verbose=*/false );
     std::string paramCard = "../../Cards/param_card.dat";
-    if( !std::filesystem::exists( paramCard ) )
-    {
-      paramCard = "../" + paramCard;
-    }
+    /*
+#ifdef __HIPCC__
+    if( !std::experimental::filesystem::exists( paramCard ) ) paramCard = "../" + paramCard;
+#else
+    if( !std::filesystem::exists( paramCard ) ) paramCard = "../" + paramCard;
+#endif
+    */
+    //struct stat dummybuffer; // bypass std::filesystem #803
+    //if( !( stat( paramCard.c_str(), &dummyBuffer ) == 0 ) ) paramCard = "../" + paramCard; //
+    auto fileExists = []( std::string& fileName )
+    { struct stat buffer; return stat( fileName.c_str(), &buffer ) == 0; };
+    if( !fileExists( paramCard ) ) paramCard = "../" + paramCard; // bypass std::filesystem #803
     process.initProc( paramCard );
   }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::set_gpugrid( const int gpublocks, const int gputhreads )
   {
@@ -276,7 +291,7 @@ namespace mg5amcCpu
   }
 #endif
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::gpu_sequence( const FORTRANFPTYPE* momenta,
                                             const FORTRANFPTYPE* gs,
@@ -291,14 +306,14 @@ namespace mg5amcCpu
     constexpr int neppM = MemoryAccessMomenta::neppM;
     if constexpr( neppM == 1 && std::is_same_v<FORTRANFPTYPE, fptype> )
     {
-      checkCuda( cudaMemcpy( m_devMomentaC.data(), momenta, m_devMomentaC.bytes(), cudaMemcpyHostToDevice ) );
+      gpuMemcpy( m_devMomentaC.data(), momenta, m_devMomentaC.bytes(), gpuMemcpyHostToDevice );
     }
     else
     {
-      checkCuda( cudaMemcpy( m_devMomentaF.data(), momenta, m_devMomentaF.bytes(), cudaMemcpyHostToDevice ) );
+      gpuMemcpy( m_devMomentaF.data(), momenta, m_devMomentaF.bytes(), gpuMemcpyHostToDevice );
       const int thrPerEvt = CPPProcess::npar * CPPProcess::np4; // AV: transpose alg does 1 element per thread (NOT 1 event per thread)
       //const int thrPerEvt = 1; // AV: try new alg with 1 event per thread... this seems slower
-      dev_transposeMomentaF2C<<<m_gpublocks * thrPerEvt, m_gputhreads>>>( m_devMomentaF.data(), m_devMomentaC.data(), m_nevt );
+      gpuLaunchKernel( dev_transposeMomentaF2C, m_gpublocks * thrPerEvt, m_gputhreads, m_devMomentaF.data(), m_devMomentaC.data(), m_nevt );
     }
     if constexpr( std::is_same_v<FORTRANFPTYPE, fptype> )
     {
@@ -341,7 +356,7 @@ namespace mg5amcCpu
   }
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::cpu_sequence( const FORTRANFPTYPE* momenta,
                                             const FORTRANFPTYPE* gs,
@@ -396,7 +411,7 @@ namespace mg5amcCpu
   // - C++ array: momenta[npagM][npar][np4][neppM] with nevt=npagM*neppM (AOSOA)
   //
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename Tin, typename Tout>
   __global__ void dev_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt )
   {
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/BridgeKernels.cc b/epochX/cudacpp/ee_mumu.mad/SubProcesses/BridgeKernels.cc
index d58066c9c1..eaf4037a24 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/BridgeKernels.cc
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/BridgeKernels.cc
@@ -1,17 +1,18 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "BridgeKernels.h"
 
+#include "GpuAbstraction.h"
 #include "MemoryAccessMomenta.h"
 
 #include <sstream>
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -45,7 +46,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu
 {
 
@@ -96,7 +97,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/BridgeKernels.h b/epochX/cudacpp/ee_mumu.mad/SubProcesses/BridgeKernels.h
index 15eb4bff4d..3efef8ce97 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/BridgeKernels.h
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/BridgeKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef BRIDGEKERNELS_H
 #define BRIDGEKERNELS_H 1
@@ -12,7 +12,7 @@
 #include "MatrixElementKernels.h"
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -49,7 +49,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A Bridge wrapper class encapsulating matrix element calculations on a CPU host
   class BridgeKernelHost final : public BridgeKernelBase
   {
@@ -89,7 +89,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A Bridge wrapper class encapsulating matrix element calculations on a GPU device
   class BridgeKernelDevice : public BridgeKernelBase
   {
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/CommonRandomNumberKernel.cc b/epochX/cudacpp/ee_mumu.mad/SubProcesses/CommonRandomNumberKernel.cc
index 985b39f576..010bc4cbd0 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/CommonRandomNumberKernel.cc
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/CommonRandomNumberKernel.cc
@@ -1,15 +1,16 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "CommonRandomNumbers.h"
+#include "GpuAbstraction.h"
 #include "MemoryBuffers.h"
 #include "RandomNumberKernels.h"
 
 #include <cassert>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/CrossSectionKernels.cc b/epochX/cudacpp/ee_mumu.mad/SubProcesses/CrossSectionKernels.cc
index 0b355a3c8d..c15b39844d 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/CrossSectionKernels.cc
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/CrossSectionKernels.cc
@@ -1,10 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "CrossSectionKernels.h"
 
+#include "GpuAbstraction.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessWeights.h"
 #include "MemoryBuffers.h"
@@ -77,7 +78,7 @@ debug_me_is_abnormal( const fptype& me, size_t ievtALL )
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -185,7 +186,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/CrossSectionKernels.h b/epochX/cudacpp/ee_mumu.mad/SubProcesses/CrossSectionKernels.h
index 7933ca4bbf..4d9659e04e 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/CrossSectionKernels.h
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/CrossSectionKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef CROSSSECTIONKERNELS_H
 #define CROSSSECTIONKERNELS_H 1
@@ -13,7 +13,7 @@
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -96,7 +96,7 @@ namespace mg5amcCpu
   //--------------------------------------------------------------------------
 
   /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating the calculation of event statistics on a GPU device
   class CrossSectionKernelDevice : public CrossSectionKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/CurandRandomNumberKernel.cc b/epochX/cudacpp/ee_mumu.mad/SubProcesses/CurandRandomNumberKernel.cc
index eb56333b03..08a16f6f2c 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/CurandRandomNumberKernel.cc
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/CurandRandomNumberKernel.cc
@@ -1,9 +1,9 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "MemoryBuffers.h"
 #include "RandomNumberKernels.h"
 
@@ -22,7 +22,7 @@ inline void assertCurand( curandStatus_t code, const char *file, int line, bool
 }
 #endif /* clang-format on */
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -36,7 +36,7 @@ namespace mg5amcCpu
   {
     if( m_isOnDevice )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if( !m_rnarray.isOnDevice() )
         throw std::runtime_error( "CurandRandomNumberKernel on device with a host random number array" );
 #else
@@ -114,7 +114,7 @@ namespace mg5amcCpu
     /*
     printf( "\nCurandRandomNumberKernel::generateRnarray size = %d\n", (int)m_rnarray.size() );
     fptype* data = m_rnarray.data();
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( m_rnarray.isOnDevice() )
     {
       data = new fptype[m_rnarray.size()]();
@@ -123,7 +123,7 @@ namespace mg5amcCpu
 #endif
     for( int i = 0; i < ( (int)m_rnarray.size() / 4 ); i++ )
       printf( "[%4d] %f %f %f %f\n", i * 4, data[i * 4], data[i * 4 + 2], data[i * 4 + 2], data[i * 4 + 3] );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( m_rnarray.isOnDevice() ) delete[] data;
 #endif
     */
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/EventStatistics.h b/epochX/cudacpp/ee_mumu.mad/SubProcesses/EventStatistics.h
index 48b51e0a49..b425a5bade 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/EventStatistics.h
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/EventStatistics.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef EventStatistics_H
 #define EventStatistics_H 1
@@ -16,7 +16,7 @@
 #include <limits>
 #include <string>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/GpuAbstraction.h b/epochX/cudacpp/ee_mumu.mad/SubProcesses/GpuAbstraction.h
new file mode 100644
index 0000000000..9c467b1e04
--- /dev/null
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/GpuAbstraction.h
@@ -0,0 +1,69 @@
+// Copyright (C) 2020-2023 CERN and UCLouvain.
+// Licensed under the GNU Lesser General Public License (version 3 or later).
+// Created by: J. Teig (Jul 2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+
+#ifndef MG5AMC_GPUABSTRACTION_H
+#define MG5AMC_GPUABSTRACTION_H 1
+
+#include <cassert>
+
+//--------------------------------------------------------------------------
+
+#ifdef __CUDACC__
+
+#define gpuError_t cudaError_t
+#define gpuPeekAtLastError cudaPeekAtLastError
+#define gpuGetErrorString cudaGetErrorString
+#define gpuSuccess cudaSuccess
+
+#define gpuMallocHost( ptr, size ) checkGpu( cudaMallocHost( ptr, size ) )
+#define gpuMalloc( ptr, size ) checkGpu( cudaMalloc( ptr, size ) )
+
+#define gpuMemcpy( dstData, srcData, srcBytes, func ) checkGpu( cudaMemcpy( dstData, srcData, srcBytes, func ) )
+#define gpuMemcpyHostToDevice cudaMemcpyHostToDevice
+#define gpuMemcpyDeviceToHost cudaMemcpyDeviceToHost
+#define gpuMemcpyToSymbol( type1, type2, size ) checkGpu( cudaMemcpyToSymbol( type1, type2, size ) )
+
+#define gpuFree( ptr ) checkGpu( cudaFree( ptr ) )
+#define gpuFreeHost( ptr ) checkGpu( cudaFreeHost( ptr ) )
+
+#define gpuSetDevice cudaSetDevice
+#define gpuDeviceSynchronize cudaDeviceSynchronize
+#define gpuDeviceReset cudaDeviceReset
+
+#define gpuLaunchKernel( kernel, blocks, threads, ... ) kernel<<<blocks, threads>>>( __VA_ARGS__ )
+#define gpuLaunchKernelSharedMem( kernel, blocks, threads, sharedMem, ... ) kernel<<<blocks, threads, sharedMem>>>( __VA_ARGS__ )
+
+//--------------------------------------------------------------------------
+
+#elif defined __HIPCC__
+
+#define gpuError_t hipError_t
+#define gpuPeekAtLastError hipPeekAtLastError
+#define gpuGetErrorString hipGetErrorString
+#define gpuSuccess hipSuccess
+
+#define gpuMallocHost( ptr, size ) checkGpu( hipHostMalloc( ptr, size ) ) // HostMalloc better
+#define gpuMalloc( ptr, size ) checkGpu( hipMalloc( ptr, size ) )
+
+#define gpuMemcpy( dstData, srcData, srcBytes, func ) checkGpu( hipMemcpy( dstData, srcData, srcBytes, func ) )
+#define gpuMemcpyHostToDevice hipMemcpyHostToDevice
+#define gpuMemcpyDeviceToHost hipMemcpyDeviceToHost
+#define gpuMemcpyToSymbol( type1, type2, size ) checkGpu( hipMemcpyToSymbol( type1, type2, size ) )
+
+#define gpuFree( ptr ) checkGpu( hipFree( ptr ) )
+#define gpuFreeHost( ptr ) checkGpu( hipHostFree( ptr ) )
+
+#define gpuSetDevice hipSetDevice
+#define gpuDeviceSynchronize hipDeviceSynchronize
+#define gpuDeviceReset hipDeviceReset
+
+#define gpuLaunchKernel( kernel, blocks, threads, ... ) kernel<<<blocks, threads>>>( __VA_ARGS__ )
+#define gpuLaunchKernelSharedMem( kernel, blocks, threads, sharedMem, ... ) kernel<<<blocks, threads, sharedMem>>>( __VA_ARGS__ )
+
+//--------------------------------------------------------------------------
+
+#endif
+
+#endif // MG5AMC_GPUABSTRACTION_H
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/CudaRuntime.h b/epochX/cudacpp/ee_mumu.mad/SubProcesses/GpuRuntime.h
similarity index 62%
rename from epochX/cudacpp/ee_mumu.mad/SubProcesses/CudaRuntime.h
rename to epochX/cudacpp/ee_mumu.mad/SubProcesses/GpuRuntime.h
index 64ce52f4b3..93579ef08b 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/CudaRuntime.h
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/GpuRuntime.h
@@ -1,49 +1,50 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
-// Created by: S. Roiser (Jul 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Created by: J. Teig (Jun 2023, based on earlier work by S. Roiser) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
-#ifndef MG5AMC_CUDARUNTIME_H
-#define MG5AMC_CUDARUNTIME_H 1
+#ifndef MG5AMC_GPURUNTIME_H
+#define MG5AMC_GPURUNTIME_H 1
 
 // MG5AMC on GPU uses the CUDA runtime API, not the lower level CUDA driver API
 // See https://docs.nvidia.com/cuda/cuda-runtime-api/driver-vs-runtime-api.html#driver-vs-runtime-api
 
-#include <cassert>
+#include "GpuAbstraction.h"
+
 #include <iostream>
 
 //--------------------------------------------------------------------------
 
 // See https://stackoverflow.com/a/14038590
-#ifdef __CUDACC__ /* clang-format off */
-#define checkCuda( code ) { assertCuda( code, __FILE__, __LINE__ ); }
-inline void assertCuda( cudaError_t code, const char* file, int line, bool abort = true )
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
+#define checkGpu( code ) { assertGpu( code, __FILE__, __LINE__ ); }
+inline void assertGpu( gpuError_t code, const char* file, int line, bool abort = true )
 {
-  if( code != cudaSuccess )
+  if( code != gpuSuccess )
   {
-    printf( "ERROR! assertCuda: '%s' (%d) in %s:%d\n", cudaGetErrorString( code ), code, file, line );
-    if( abort ) assert( code == cudaSuccess );
+    printf( "ERROR! assertGpu: '%s' (%d) in %s:%d\n", gpuGetErrorString( code ), code, file, line );
+    if( abort ) assert( code == gpuSuccess );
   }
 }
 #endif /* clang-format on */
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
+  // Instantiate a GpuRuntime at the beginnining of the application's main to
+  // invoke gpuSetDevice(0) in the constructor and book a gpuDeviceReset() call in the destructor
   // *** FIXME! This will all need to be designed differently when going to multi-GPU nodes! ***
-  struct CudaRuntime final
+  struct GpuRuntime final
   {
-    CudaRuntime( const bool debug = true )
+    GpuRuntime( const bool debug = true )
       : m_debug( debug ) { setUp( m_debug ); }
-    ~CudaRuntime() { tearDown( m_debug ); }
-    CudaRuntime( const CudaRuntime& ) = delete;
-    CudaRuntime( CudaRuntime&& ) = delete;
-    CudaRuntime& operator=( const CudaRuntime& ) = delete;
-    CudaRuntime& operator=( CudaRuntime&& ) = delete;
+    ~GpuRuntime() { tearDown( m_debug ); }
+    GpuRuntime( const GpuRuntime& ) = delete;
+    GpuRuntime( GpuRuntime&& ) = delete;
+    GpuRuntime& operator=( const GpuRuntime& ) = delete;
+    GpuRuntime& operator=( GpuRuntime&& ) = delete;
     bool m_debug;
 
     // Set up CUDA application
@@ -62,8 +63,8 @@ namespace mg5amcGpu
       */
       // Replace cudaFree(0) by cudaSetDevice(0), even if it is not really needed either
       // (but see https://developer.nvidia.com/blog/cuda-pro-tip-always-set-current-device-avoid-multithreading-bugs)
-      if( debug ) std::cout << "__CudaRuntime: calling cudaSetDevice(0)" << std::endl;
-      checkCuda( cudaSetDevice( 0 ) ); // SLOW!
+      if( debug ) std::cout << "__GpuRuntime: calling GpuSetDevice(0)" << std::endl;
+      checkGpu( gpuSetDevice( 0 ) ); // SLOW!
     }
 
     // Tear down CUDA application (call cudaDeviceReset)
@@ -72,14 +73,13 @@ namespace mg5amcGpu
     // See https://docs.nvidia.com/cuda/cuda-memcheck/index.html#leak-checking
     static void tearDown( const bool debug = true )
     {
-      if( debug ) std::cout << "__CudaRuntime: calling cudaDeviceReset()" << std::endl;
-      checkCuda( cudaDeviceReset() );
+      if( debug ) std::cout << "__GpuRuntime: calling GpuDeviceReset()" << std::endl;
+      checkGpu( gpuDeviceReset() );
     }
   };
-
 }
 #endif
 
 //--------------------------------------------------------------------------
 
-#endif // MG5AMC_CUDARUNTIME_H
+#endif // MG5AMC_GPURUNTIME_H
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/MadgraphTest.h b/epochX/cudacpp/ee_mumu.mad/SubProcesses/MadgraphTest.h
index ef40624c88..6054185300 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/MadgraphTest.h
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/MadgraphTest.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Hageboeck (Dec 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MADGRAPHTEST_H_
 #define MADGRAPHTEST_H_ 1
@@ -14,7 +14,11 @@
 
 #include <array>
 #include <cmath>
-#include <filesystem>
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
 #include <fstream>
 #include <iomanip>
 #include <memory>
@@ -22,7 +26,7 @@
 #include <string>
 #include <vector>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using mg5amcGpu::CPPProcess;
 #else
 using mg5amcCpu::CPPProcess;
@@ -201,7 +205,7 @@ class MadgraphTest : public testing::TestWithParam<TestDriverBase*>
 
 // Since we link both the CPU-only and GPU tests into the same executable, we prevent
 // a multiply defined symbol by only compiling this in the non-CUDA phase:
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
 /// Compare momenta and matrix elements.
 /// This uses an implementation of TestDriverBase to run a madgraph workflow,
@@ -219,7 +223,14 @@ TEST_P( MadgraphTest, CompareMomentaAndME )
   const char* dumpEventsC = getenv( "CUDACPP_RUNTEST_DUMPEVENTS" );
   const bool dumpEvents = ( dumpEventsC != 0 ) && ( std::string( dumpEventsC ) != "" );
   const std::string refFileName = testDriver->getRefFileName();
+  /*
+#ifdef __HIPCC__
+  const std::string dumpFileName = std::experimental::filesystem::path( refFileName ).filename();
+#else
   const std::string dumpFileName = std::filesystem::path( refFileName ).filename();
+#endif
+  */
+  const std::string dumpFileName = refFileName; // bypass std::filesystem #803
   std::ofstream dumpFile;
   if( dumpEvents )
   {
@@ -307,6 +318,6 @@ TEST_P( MadgraphTest, CompareMomentaAndME )
   }
 }
 
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
 
 #endif /* MADGRAPHTEST_H_ */
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/MatrixElementKernels.cc b/epochX/cudacpp/ee_mumu.mad/SubProcesses/MatrixElementKernels.cc
index 74b5239ebf..81699dfea9 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/MatrixElementKernels.cc
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/MatrixElementKernels.cc
@@ -1,12 +1,12 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "MatrixElementKernels.h"
 
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h" // Includes the abstraction for Nvidia/AMD compilation
 #include "MemoryAccessMomenta.h"
 #include "MemoryBuffers.h"
 
@@ -14,7 +14,7 @@
 
 //============================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu
 {
 
@@ -150,7 +150,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
@@ -209,13 +209,13 @@ namespace mg5amcGpu
     PinnedHostBufferHelicityMask hstIsGoodHel( ncomb );
     DeviceBufferHelicityMask devIsGoodHel( ncomb );
     // ... 0d1. Compute good helicity mask on the device
-    computeDependentCouplings<<<m_gpublocks, m_gputhreads>>>( m_gs.data(), m_couplings.data() );
+    gpuLaunchKernel( computeDependentCouplings, m_gpublocks, m_gputhreads, m_gs.data(), m_couplings.data() );
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-    sigmaKin_getGoodHel<<<m_gpublocks, m_gputhreads>>>( m_momenta.data(), m_couplings.data(), m_matrixElements.data(), m_numerators.data(), m_denominators.data(), devIsGoodHel.data() );
+    gpuLaunchKernel( sigmaKin_getGoodHel, m_gpublocks, m_gputhreads, m_momenta.data(), m_couplings.data(), m_matrixElements.data(), m_numerators.data(), m_denominators.data(), devIsGoodHel.data() );
 #else
-    sigmaKin_getGoodHel<<<m_gpublocks, m_gputhreads>>>( m_momenta.data(), m_couplings.data(), m_matrixElements.data(), devIsGoodHel.data() );
+    gpuLaunchKernel( sigmaKin_getGoodHel, m_gpublocks, m_gputhreads, m_momenta.data(), m_couplings.data(), m_matrixElements.data(), devIsGoodHel.data() );
 #endif
-    checkCuda( cudaPeekAtLastError() );
+    checkGpu( gpuPeekAtLastError() );
     // ... 0d2. Copy back good helicity mask to the host
     copyHostFromDevice( hstIsGoodHel, devIsGoodHel );
     // ... 0d3. Copy back good helicity list to constant memory on the device
@@ -226,19 +226,19 @@ namespace mg5amcGpu
 
   void MatrixElementKernelDevice::computeMatrixElements( const unsigned int channelId )
   {
-    computeDependentCouplings<<<m_gpublocks, m_gputhreads>>>( m_gs.data(), m_couplings.data() );
+    gpuLaunchKernel( computeDependentCouplings, m_gpublocks, m_gputhreads, m_gs.data(), m_couplings.data() );
 #ifndef MGONGPU_NSIGHT_DEBUG
     constexpr unsigned int sharedMemSize = 0;
 #else
     constexpr unsigned int sharedMemSize = ntpbMAX * sizeof( float );
 #endif
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-    sigmaKin<<<m_gpublocks, m_gputhreads, sharedMemSize>>>( m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), channelId, m_numerators.data(), m_denominators.data(), m_selhel.data(), m_selcol.data() );
+    gpuLaunchKernelSharedMem( sigmaKin, m_gpublocks, m_gputhreads, sharedMemSize, m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), channelId, m_numerators.data(), m_denominators.data(), m_selhel.data(), m_selcol.data() );
 #else
-    sigmaKin<<<m_gpublocks, m_gputhreads, sharedMemSize>>>( m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), m_selhel.data(), m_selcol.data() );
+    gpuLaunchKernelSharedMem( sigmaKin, m_gpublocks, m_gputhreads, sharedMemSize, m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), m_selhel.data(), m_selcol.data() );
 #endif
-    checkCuda( cudaPeekAtLastError() );
-    checkCuda( cudaDeviceSynchronize() );
+    checkGpu( gpuPeekAtLastError() );
+    checkGpu( gpuDeviceSynchronize() );
   }
 
   //--------------------------------------------------------------------------
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/MatrixElementKernels.h b/epochX/cudacpp/ee_mumu.mad/SubProcesses/MatrixElementKernels.h
index 23e84757a2..72bd8f195b 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/MatrixElementKernels.h
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/MatrixElementKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MATRIXELEMENTKERNELS_H
 #define MATRIXELEMENTKERNELS_H 1
@@ -10,7 +10,7 @@
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -81,7 +81,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating matrix element calculations on a CPU host
   class MatrixElementKernelHost final : public MatrixElementKernelBase, public NumberOfEvents
   {
@@ -130,7 +130,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating matrix element calculations on a GPU device
   class MatrixElementKernelDevice : public MatrixElementKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessAmplitudes.h b/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessAmplitudes.h
index 573b3bbbc9..ffb76e93de 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessAmplitudes.h
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessAmplitudes.h
@@ -15,7 +15,7 @@
 #define MGONGPU_TRIVIAL_AMPLITUDES 1
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessCouplings.h b/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessCouplings.h
index 35a3af42e0..3afdf3e554 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessCouplings.h
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessCouplings.h
@@ -15,7 +15,7 @@
 #include "MemoryBuffers.h"       // for HostBufferCouplings::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessCouplingsFixed.h b/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessCouplingsFixed.h
index dc0d93afff..ffcdf4dbef 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessCouplingsFixed.h
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessCouplingsFixed.h
@@ -14,7 +14,7 @@
 //#include "MemoryAccessHelpers.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessDenominators.h b/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessDenominators.h
index 3bce635718..66f2d32a6b 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessDenominators.h
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessDenominators.h
@@ -10,7 +10,7 @@
 #include "MemoryAccessGs.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessGs.h b/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessGs.h
index 31311aa375..4c726b30f3 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessGs.h
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessGs.h
@@ -13,7 +13,7 @@
 #include "MemoryBuffers.h" // for HostBufferMatrixElements::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessHelpers.h b/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessHelpers.h
index c82a6c7635..db73e4e064 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessHelpers.h
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessHelpers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessHelpers_H
 #define MemoryAccessHelpers_H 1
@@ -105,7 +105,7 @@ class KernelAccessHelper : public MemoryAccessHelper<T>
     }
     else
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
       //printf( "kernelAccessRecord: ievt=%d threadId=%d\n", ievt, threadIdx.x );
       return T::ieventAccessRecord( buffer, ievt ); // NB fptype and fptype_sv coincide for CUDA
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessMatrixElements.h b/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessMatrixElements.h
index f32e6fea5b..3741011971 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessMatrixElements.h
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessMatrixElements.h
@@ -13,7 +13,7 @@
 #include "MemoryBuffers.h" // for HostBufferMatrixElements::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessMomenta.h b/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessMomenta.h
index 29266de32c..3be229d392 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessMomenta.h
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessMomenta.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessMomenta_H
 #define MemoryAccessMomenta_H 1
@@ -13,7 +13,7 @@
 #include "MemoryAccessVectors.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -30,7 +30,7 @@ namespace mg5amcCpu
 
     // Number of Events Per Page in the momenta AOSOA memory buffer layout
     // (these are all best kept as a compile-time constants: see issue #23)
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
     // -----------------------------------------------------------------------------------------------
     // --- GPUs: neppM is best set to a power of 2 times the number of fptype's in a 32-byte cacheline
     // --- This is relevant to ensure coalesced access to momenta in global memory
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessNumerators.h b/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessNumerators.h
index b152183b28..18991f4fa6 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessNumerators.h
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessNumerators.h
@@ -10,7 +10,7 @@
 #include "MemoryAccessGs.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessRandomNumbers.h b/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessRandomNumbers.h
index e2988d39f3..40cb089135 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessRandomNumbers.h
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessRandomNumbers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessRandomNumbers_H
 #define MemoryAccessRandomNumbers_H 1
@@ -11,7 +11,7 @@
 #include "CPPProcess.h"
 #include "MemoryAccessHelpers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using mg5amcGpu::CPPProcess;
 #else
 using mg5amcCpu::CPPProcess;
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessVectors.h b/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessVectors.h
index e9b197368e..08faccff0f 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessVectors.h
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessVectors.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessVectors_H
 #define MemoryAccessVectors_H 1
@@ -10,7 +10,7 @@
 
 #include "mgOnGpuVectors.h"
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu // this is only needed for CPU SIMD vectorization
 {
 
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessWavefunctions.h b/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessWavefunctions.h
index 5428aaf933..33bef4559e 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessWavefunctions.h
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryAccessWavefunctions.h
@@ -15,7 +15,7 @@
 #define MGONGPU_TRIVIAL_WAVEFUNCTIONS 1
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryBuffers.h b/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryBuffers.h
index 3093e6ed18..7756a71621 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryBuffers.h
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/MemoryBuffers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021, based on earlier work by S. Hageboeck) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryBuffers_H
 #define MemoryBuffers_H 1
@@ -11,12 +11,12 @@
 #include "mgOnGpuCxtypes.h"
 
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "Parameters_sm.h"
 
 #include <sstream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -87,7 +87,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   constexpr bool HostBufferALIGNED = false;   // ismisaligned=false
   constexpr bool HostBufferMISALIGNED = true; // ismisaligned=true
 
@@ -119,7 +119,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA pinned host buffer
   template<typename T>
   class PinnedHostBufferBase : public BufferBase<T>
@@ -128,18 +128,18 @@ namespace mg5amcCpu
     PinnedHostBufferBase( const size_t size )
       : BufferBase<T>( size, false )
     {
-      checkCuda( cudaMallocHost( &( this->m_data ), this->bytes() ) );
+      gpuMallocHost( &( this->m_data ), this->bytes() );
     }
     virtual ~PinnedHostBufferBase()
     {
-      checkCuda( cudaFreeHost( this->m_data ) );
+      gpuFreeHost( this->m_data );
     }
   };
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA device buffer
   template<typename T>
   class DeviceBufferBase : public BufferBase<T>
@@ -148,18 +148,18 @@ namespace mg5amcCpu
     DeviceBufferBase( const size_t size )
       : BufferBase<T>( size, true )
     {
-      checkCuda( cudaMalloc( &( this->m_data ), this->bytes() ) );
+      gpuMalloc( &( this->m_data ), this->bytes() );
     }
     virtual ~DeviceBufferBase()
     {
-      checkCuda( cudaFree( this->m_data ) );
+      gpuFree( this->m_data );
     }
   };
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for a given number of events
   template<typename T, size_t sizePerEvent, bool ismisaligned>
   class HostBuffer : public HostBufferBase<T, ismisaligned>, virtual private NumberOfEvents
@@ -175,7 +175,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA pinned host buffer for a given number of events
   template<typename T, size_t sizePerEvent>
   class PinnedHostBuffer : public PinnedHostBufferBase<T>, virtual private NumberOfEvents
@@ -191,7 +191,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA device buffer for a given number of events
   template<typename T, size_t sizePerEvent>
   class DeviceBuffer : public DeviceBufferBase<T>, virtual private NumberOfEvents
@@ -213,7 +213,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for momenta random numbers
   constexpr size_t sizePerEventRndNumMomenta = MemoryBuffers::np4 * MemoryBuffers::nparf;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for momenta random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumMomenta, HostBufferALIGNED> HostBufferRndNumMomenta;
 #else
@@ -232,7 +232,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer with ONE fptype per event
   constexpr size_t sizePerEventOneFp = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer with ONE fptype per event
   typedef HostBuffer<fptype, sizePerEventOneFp, HostBufferALIGNED> HostBufferOneFp;
 #else
@@ -257,7 +257,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for Gs
   constexpr size_t sizePerEventGs = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventGs, HostBufferALIGNED> HostBufferGs;
 #else
@@ -276,7 +276,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for numerators
   constexpr size_t sizePerEventNumerators = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventNumerators, HostBufferALIGNED> HostBufferNumerators;
 #else
@@ -296,7 +296,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for denominators
   constexpr size_t sizePerEventDenominators = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventDenominators, HostBufferALIGNED> HostBufferDenominators;
 #else
@@ -315,7 +315,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for random numbers
   constexpr size_t sizePerEventCouplings = MemoryBuffers::ndcoup * MemoryBuffers::nx2;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventCouplings, HostBufferALIGNED> HostBufferCouplings;
 #else
@@ -333,7 +333,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for momenta
   constexpr size_t sizePerEventMomenta = MemoryBuffers::np4 * MemoryBuffers::npar;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for momenta
   typedef HostBuffer<fptype, sizePerEventMomenta, HostBufferALIGNED> HostBufferMomenta;
   //typedef HostBuffer<fptype, sizePerEventMomenta, HostBufferMISALIGNED> HostBufferMomenta; // TEST MISALIGNMENT!
@@ -352,7 +352,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for sampling weights
   constexpr size_t sizePerEventWeights = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for sampling weights
   typedef HostBuffer<fptype, sizePerEventWeights, HostBufferALIGNED> HostBufferWeights;
 #else
@@ -370,7 +370,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for matrix elements
   constexpr size_t sizePerEventMatrixElements = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for matrix elements
   typedef HostBuffer<fptype, sizePerEventMatrixElements, HostBufferALIGNED> HostBufferMatrixElements;
 #else
@@ -385,7 +385,7 @@ namespace mg5amcCpu
   // A base class encapsulating a memory buffer for the helicity mask
   typedef BufferBase<bool> BufferHelicityMask;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for the helicity mask
   typedef HostBufferBase<bool, HostBufferALIGNED> HostBufferHelicityMask;
 #else
@@ -403,7 +403,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for wavefunctions
   constexpr size_t sizePerEventWavefunctions = MemoryBuffers::nw6 * MemoryBuffers::nx2;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for wavefunctions
   typedef HostBuffer<fptype, sizePerEventWavefunctions, HostBufferALIGNED> HostBufferWavefunctions;
 #else
@@ -421,7 +421,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for helicity random numbers
   constexpr size_t sizePerEventRndNumHelicity = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for helicity random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumHelicity, HostBufferALIGNED> HostBufferRndNumHelicity;
 #else
@@ -439,7 +439,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for color random numbers
   constexpr size_t sizePerEventRndNumColor = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for color random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumColor, HostBufferALIGNED> HostBufferRndNumColor;
 #else
@@ -457,7 +457,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for helicity selection
   constexpr size_t sizePerEventSelectedHelicity = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for helicity selection
   typedef HostBuffer<int, sizePerEventSelectedHelicity, HostBufferALIGNED> HostBufferSelectedHelicity;
 #else
@@ -475,7 +475,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for color selection
   constexpr size_t sizePerEventSelectedColor = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for color selection
   typedef HostBuffer<int, sizePerEventSelectedColor, HostBufferALIGNED> HostBufferSelectedColor;
 #else
@@ -487,7 +487,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<class Tdst, class Tsrc>
   void copyDeviceFromHost( Tdst& dst, const Tsrc& src ) // keep the same order of arguments as in memcpy
   {
@@ -504,13 +504,13 @@ namespace mg5amcCpu
       throw std::runtime_error( sstr.str() );
     }
     // NB (PR #45): cudaMemcpy involves an intermediate memcpy to pinned memory if host array is a not a pinned host array
-    checkCuda( cudaMemcpy( dst.data(), src.data(), src.bytes(), cudaMemcpyHostToDevice ) );
+    gpuMemcpy( dst.data(), src.data(), src.bytes(), gpuMemcpyHostToDevice );
   }
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<class Tdst, class Tsrc>
   void copyHostFromDevice( Tdst& dst, const Tsrc& src ) // keep the same order of arguments as in memcpy
   {
@@ -527,7 +527,7 @@ namespace mg5amcCpu
       throw std::runtime_error( sstr.str() );
     }
     // NB (PR #45): cudaMemcpy involves an intermediate memcpy to pinned memory if host array is a not a pinned host array
-    checkCuda( cudaMemcpy( dst.data(), src.data(), src.bytes(), cudaMemcpyDeviceToHost ) );
+    gpuMemcpy( dst.data(), src.data(), src.bytes(), gpuMemcpyDeviceToHost );
   }
 #endif
 
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/CPPProcess.cc b/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/CPPProcess.cc
index 9193aa2382..c57ff8d2b0 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/CPPProcess.cc
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_sm.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -46,7 +45,7 @@
 // Class member functions for calculating the matrix elements for
 // Process: e+ e- > mu+ mu- WEIGHTED<=4 @1
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -80,7 +79,7 @@ namespace mg5amcCpu
   __device__ const fptype cIPD[2] = { (fptype)Parameters_sm::mdl_MZ, (fptype)Parameters_sm::mdl_WZ };
   __device__ const fptype cIPC[6] = { (fptype)Parameters_sm::GC_3.real(), (fptype)Parameters_sm::GC_3.imag(), (fptype)Parameters_sm::GC_50.real(), (fptype)Parameters_sm::GC_50.imag(), (fptype)Parameters_sm::GC_59.real(), (fptype)Parameters_sm::GC_59.imag() };
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ fptype cIPD[2];
   __device__ __constant__ fptype cIPC[6];
 #else
@@ -90,7 +89,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -118,13 +117,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -151,7 +150,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -187,7 +186,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -200,8 +199,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -286,7 +287,7 @@ namespace mg5amcCpu
       // [NB do keep 'static' for these constexpr arrays, see issue #283]
       static constexpr fptype2 cf[ncolor][ncolor] = { { 1 } }; // 2-D array[1][1]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -343,7 +344,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -402,7 +403,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -449,8 +450,8 @@ namespace mg5amcCpu
       { 1, -1, 1, 1 },
       { 1, -1, -1, -1 },
       { 1, -1, -1, 1 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -490,9 +491,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     const fptype tIPD[2] = { (fptype)m_pars->mdl_MZ, (fptype)m_pars->mdl_WZ };
     const cxtype tIPC[3] = { cxmake( m_pars->GC_3 ), cxmake( m_pars->GC_50 ), cxmake( m_pars->GC_59 ) };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) ) );
-    checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 3 * sizeof( cxtype ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) );
+    gpuMemcpyToSymbol( cIPC, tIPC, 3 * sizeof( cxtype ) );
 #else
     memcpy( cIPD, tIPD, 2 * sizeof( fptype ) );
     memcpy( cIPC, tIPC, 3 * sizeof( cxtype ) );
@@ -529,7 +530,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -558,6 +559,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -571,6 +576,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -594,12 +601,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -620,7 +627,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -746,9 +753,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -772,7 +779,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -792,7 +799,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 4 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -806,9 +813,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -836,7 +846,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -1046,7 +1056,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/CPPProcess.h b/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/CPPProcess.h
index 77b610753c..0b29ffb3ff 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/CPPProcess.h
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -107,7 +107,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -120,7 +120,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -150,7 +150,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/CudaRuntime.h b/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/GpuAbstraction.h b/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/GpuRuntime.h b/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/check_sa.cc b/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/check_sa.cc
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/RamboSamplingKernels.cc b/epochX/cudacpp/ee_mumu.mad/SubProcesses/RamboSamplingKernels.cc
index da68aa9255..79abbcc4f8 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/RamboSamplingKernels.cc
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/RamboSamplingKernels.cc
@@ -1,11 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "RamboSamplingKernels.h"
 
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "MemoryAccessMomenta.h"
 #include "MemoryAccessRandomNumbers.h"
 #include "MemoryAccessWeights.h"
@@ -14,7 +14,7 @@
 
 #include <sstream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -92,7 +92,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingKernelDevice::RamboSamplingKernelDevice( const fptype energy,               // input: energy
                                                         const BufferRndNumMomenta& rndmom, // input: random numbers in [0,1]
                                                         BufferMomenta& momenta,            // output: momenta
@@ -135,7 +135,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   getMomentaInitialDevice( const fptype energy,
                            fptype* momenta )
@@ -147,17 +147,17 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   void
   RamboSamplingKernelDevice::getMomentaInitial()
   {
-    getMomentaInitialDevice<<<m_gpublocks, m_gputhreads>>>( m_energy, m_momenta.data() );
+    gpuLaunchKernel( getMomentaInitialDevice, m_gpublocks, m_gputhreads, m_energy, m_momenta.data() );
   }
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   getMomentaFinalDevice( const fptype energy,
                          const fptype* rndmom,
@@ -171,11 +171,11 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   void
   RamboSamplingKernelDevice::getMomentaFinal()
   {
-    getMomentaFinalDevice<<<m_gpublocks, m_gputhreads>>>( m_energy, m_rndmom.data(), m_momenta.data(), m_weights.data() );
+    gpuLaunchKernel( getMomentaFinalDevice, m_gpublocks, m_gputhreads, m_energy, m_rndmom.data(), m_momenta.data(), m_weights.data() );
   }
 #endif
 
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/RamboSamplingKernels.h b/epochX/cudacpp/ee_mumu.mad/SubProcesses/RamboSamplingKernels.h
index 184089efd7..7c214cd74b 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/RamboSamplingKernels.h
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/RamboSamplingKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef RAMBOSAMPLINGKERNELS_H
 #define RAMBOSAMPLINGKERNELS_H 1
@@ -10,7 +10,7 @@
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -93,7 +93,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating RAMBO phase space sampling on a GPU device
   class RamboSamplingKernelDevice final : public SamplingKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/RandomNumberKernels.h b/epochX/cudacpp/ee_mumu.mad/SubProcesses/RandomNumberKernels.h
index 188a72c2c9..21d63beeac 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/RandomNumberKernels.h
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/RandomNumberKernels.h
@@ -1,14 +1,14 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef RANDOMNUMBERKERNELS_H
 #define RANDOMNUMBERKERNELS_H 1
 
 #include "mgOnGpuConfig.h"
 
-// NB This must come AFTER mgOnGpuConfig.h which contains our definition of __global__ when __CUDACC__ is not defined
+// NB This must come AFTER mgOnGpuConfig.h which contains our definition of __global__ when MGONGPUCPP_GPUIMPL is not defined
 #ifndef MGONGPU_HAS_NO_CURAND
 //#include "curand.h"
 struct curandGenerator_st; // forward definition from curand.h
@@ -16,7 +16,7 @@ struct curandGenerator_st; // forward definition from curand.h
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/cudacpp.mk b/epochX/cudacpp/ee_mumu.mad/SubProcesses/cudacpp.mk
index 509307506b..df74dfc284 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/cudacpp.mk
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/cudacpp.mk
@@ -1,7 +1,7 @@
 # Copyright (C) 2020-2023 CERN and UCLouvain.
 # Licensed under the GNU Lesser General Public License (version 3 or later).
 # Created by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-# Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+# Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #=== Determine the name of this makefile (https://ftp.gnu.org/old-gnu/Manuals/make-3.80/html_node/make_17.html)
 #=== NB: use ':=' to ensure that the value of CUDACPP_MAKEFILE is not modified further down after including make_opts
@@ -42,10 +42,10 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure common compiler flags for C++ and CUDA
+#=== Configure common compiler flags for C++ and CUDA/HIP
 
 INCFLAGS = -I.
-OPTFLAGS = -O3 # this ends up in CUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
+OPTFLAGS = -O3 # this ends up in GPUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
 
 # Dependency on src directory
 MG5AMC_COMMONLIB = mg5amc_common
@@ -121,24 +121,45 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure the CUDA compiler
+#=== Configure the GPU compiler (CUDA or HIP)
+
+# FIXME! (AV 24.01.2024)
+# In the current implementation (without separate builds for C++ and CUDA/HIP), we first check for cudacc and hipcc in CUDA_HOME and HIP_HOME.
+# If CUDA_HOME or HIP_HOME are not set, try to determine them from the path to cudacc and hipcc.
+# While convoluted, this is currently necessary to allow disabling CUDA/HIP builds by setting CUDA_HOME or HIP_HOME to invalid paths.
+# This will (probably?) be fixed when separate C++ and CUDA/HIP builds are implemented (PR #775).
 
 # If CXX is not a single word (example "clang++ --gcc-toolchain...") then disable CUDA builds (issue #505)
-# This is because it is impossible to pass this to "CUFLAGS += -ccbin <host-compiler>" below
+# This is because it is impossible to pass this to "GPUFLAGS += -ccbin <host-compiler>" below
 ifneq ($(words $(subst ccache ,,$(CXX))),1) # allow at most "CXX=ccache <host-compiler>" from outside
   $(warning CUDA builds are not supported for multi-word CXX "$(CXX)")
   override CUDA_HOME=disabled
 endif
 
-# If CUDA_HOME is not set, try to set it from the location of nvcc
+# If CUDA_HOME is not set, try to set it from the path to nvcc
 ifndef CUDA_HOME
-  CUDA_HOME = $(patsubst %bin/nvcc,%,$(shell which nvcc 2>/dev/null))
+  CUDA_HOME = $(patsubst %/bin/nvcc,%,$(shell which nvcc 2>/dev/null))
   $(warning CUDA_HOME was not set: using "$(CUDA_HOME)")
 endif
 
-# Set NVCC as $(CUDA_HOME)/bin/nvcc if it exists
+# If HIP_HOME is not set, try to set it from the path to hipcc
+ifndef HIP_HOME
+  HIP_HOME = $(patsubst %/bin/hipcc,%,$(shell which hipcc 2>/dev/null))
+  $(warning HIP_HOME was not set: using "$(HIP_HOME)")
+endif
+
+# FIXME! (AV 24.01.2024)
+# In the current implementation (without separate builds for C++ and CUDA/HIP),
+# builds are performed for HIP only if CUDA is not found in the path.
+# If both CUDA and HIP are installed, HIP builds can be triggered by unsetting CUDA_HOME.
+# This will be fixed when separate C++ and CUDA/HIP builds are implemented (PR #775).
+
+#--- Option 1: CUDA exists -> use CUDA
+
+# Set GPUCC as $(CUDA_HOME)/bin/nvcc if it exists
 ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
-  NVCC = $(CUDA_HOME)/bin/nvcc
+
+  GPUCC = $(CUDA_HOME)/bin/nvcc
   USE_NVTX ?=-DUSE_NVTX
   # See https://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/index.html
   # See https://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/
@@ -148,7 +169,7 @@ ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
   # Examples: use 60 for P100 (Piz Daint), 80 for A100 (Juwels Booster, NVidia raplab/Curiosity).
   MADGRAPH_CUDA_ARCHITECTURE ?= 70
   ###CUARCHFLAGS = -gencode arch=compute_$(MADGRAPH_CUDA_ARCHITECTURE),code=compute_$(MADGRAPH_CUDA_ARCHITECTURE) -gencode arch=compute_$(MADGRAPH_CUDA_ARCHITECTURE),code=sm_$(MADGRAPH_CUDA_ARCHITECTURE) # Older implementation (AV): go back to this one for multi-GPU support #533
-  ###CUARCHFLAGS = --gpu-architecture=compute_$(MADGRAPH_CUDA_ARCHITECTURE) --gpu-code=sm_$(MADGRAPH_CUDA_ARCHITECTURE),compute_$(MADGRAPH_CUDA_ARCHITECTURE) # Newer implementation (SH): cannot use this as-is for multi-GPU support #533
+  ###CUARCHFLAGS = --gpu-architecture=compute_$(MADGRAPH_CUDA_ARCHITECTURE) --gpu-code=sm_$(MADGRAPH_CUDA_ARCHITECTURE),compute_$(MADGRAPH_CUDA_ARCHITECTURE)  # Newer implementation (SH): cannot use this as-is for multi-GPU support #533
   comma:=,
   CUARCHFLAGS = $(foreach arch,$(subst $(comma), ,$(MADGRAPH_CUDA_ARCHITECTURE)),-gencode arch=compute_$(arch),code=compute_$(arch) -gencode arch=compute_$(arch),code=sm_$(arch))
   CUINC = -I$(CUDA_HOME)/include/
@@ -158,41 +179,79 @@ ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
     CURANDLIBFLAGS = -L$(CUDA_HOME)/lib64/ -lcurand # NB: -lcuda is not needed here!
   endif
   CUOPTFLAGS = -lineinfo
-  CUFLAGS = $(foreach opt, $(OPTFLAGS), -Xcompiler $(opt)) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
-  ###CUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
-  ###NVCC_VERSION = $(shell $(NVCC) --version | grep 'Cuda compilation tools' | cut -d' ' -f5 | cut -d, -f1)
-  CUFLAGS += -std=c++17 # need CUDA >= 11.2 (see #333): this is enforced in mgOnGpuConfig.h
+  ###GPUFLAGS = $(OPTFLAGS) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
+  GPUFLAGS = $(foreach opt, $(OPTFLAGS), -Xcompiler $(opt)) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
+  ###GPUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
+  ###GPUCC_VERSION = $(shell $(GPUCC) --version | grep 'Cuda compilation tools' | cut -d' ' -f5 | cut -d, -f1)
+  GPUFLAGS += -std=c++17 # need CUDA >= 11.2 (see #333): this is enforced in mgOnGpuConfig.h
   # Without -maxrregcount: baseline throughput: 6.5E8 (16384 32 12) up to 7.3E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 160 # improves throughput: 6.9E8 (16384 32 12) up to 7.7E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 128 # improves throughput: 7.3E8 (16384 32 12) up to 7.6E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 96 # degrades throughput: 4.1E8 (16384 32 12) up to 4.5E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 64 # degrades throughput: 1.7E8 (16384 32 12) flat at 1.7E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 160 # improves throughput: 6.9E8 (16384 32 12) up to 7.7E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 128 # improves throughput: 7.3E8 (16384 32 12) up to 7.6E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 96 # degrades throughput: 4.1E8 (16384 32 12) up to 4.5E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 64 # degrades throughput: 1.7E8 (16384 32 12) flat at 1.7E8 (65536 128 12)
+  CUBUILDRULEFLAGS = -Xcompiler -fPIC -c
+  CCBUILDRULEFLAGS = -Xcompiler -fPIC -c -x cu
+  CUDATESTFLAGS = -lcuda
+
+  # Set the host C++ compiler for GPUCC via "-ccbin <host-compiler>"
+  # (NB issue #505: this must be a single word, "clang++ --gcc-toolchain..." is not supported)
+  GPUFLAGS += -ccbin $(shell which $(subst ccache ,,$(CXX)))
+
+  # Allow newer (unsupported) C++ compilers with older versions of CUDA if ALLOW_UNSUPPORTED_COMPILER_IN_CUDA is set (#504)
+  ifneq ($(origin ALLOW_UNSUPPORTED_COMPILER_IN_CUDA),undefined)
+  GPUFLAGS += -allow-unsupported-compiler
+  endif
+
 else ifneq ($(origin REQUIRE_CUDA),undefined)
+
   # If REQUIRE_CUDA is set but no cuda is found, stop here (e.g. for CI tests on GPU #443)
-  $(error No cuda installation found (set CUDA_HOME or make nvcc visible in PATH))
+  $(error No cuda installation found (set CUDA_HOME or make GPUCC visible in PATH))
+
+#--- Option 2: CUDA does not exist, HIP exists -> use HIP
+
+# Set GPUCC as $(HIP_HOME)/bin/hipcc if it exists
+else ifneq ($(wildcard $(HIP_HOME)/bin/hipcc),)
+
+  GPUCC = $(HIP_HOME)/bin/hipcc
+  #USE_NVTX ?=-DUSE_NVTX # should maybe find something equivalent to this in HIP?
+  HIPARCHFLAGS = -target x86_64-linux-gnu --offload-arch=gfx90a
+  HIPINC = -I$(HIP_HOME)/include/
+  # Note: -DHIP_FAST_MATH is equivalent to -use_fast_math in HIP 
+  # (but only for single precision line 208: https://rocm-developer-tools.github.io/HIP/hcc__detail_2math__functions_8h_source.html)
+  # Note: CUOPTFLAGS should not be used for HIP, it had been added here but was then removed (#808)
+  GPUFLAGS = $(OPTFLAGS) $(INCFLAGS) $(HIPINC) $(HIPARCHFLAGS) -DHIP_FAST_MATH -DHIP_PLATFORM=amd -fPIC
+  ###GPUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
+  GPUFLAGS += -std=c++17
+  ###GPUFLAGS+= --maxrregcount 255 # (AV: is this option valid on HIP and meaningful on AMD GPUs?)
+  CUBUILDRULEFLAGS = -fPIC -c
+  CCBUILDRULEFLAGS = -fPIC -c -x hip
+
+else ifneq ($(origin REQUIRE_HIP),undefined)
+
+  # If REQUIRE_HIP is set but no HIP is found, stop here (e.g. for CI tests on GPU #443)
+  $(error No hip installation found (set HIP_HOME or make GPUCC visible in PATH))
+
+#--- Option 3: CUDA does not exist, HIP does not exist -> switch off both CUDA and HIP
+
 else
-  # No cuda. Switch cuda compilation off and go to common random numbers in C++
+
+  # No cudacc and no hipcc: switch CUDA and HIP compilation off and go to common random numbers in C++
   $(warning CUDA_HOME is not set or is invalid: export CUDA_HOME to compile with cuda)
-  override NVCC=
+  $(warning HIP_HOME is not set or is invalid: export HIP_HOME to compile with hip)
+  override GPUCC=
   override USE_NVTX=
   override CUINC=
   override CURANDLIBFLAGS=
-endif
-export NVCC
-export CUFLAGS
 
-# Set the host C++ compiler for nvcc via "-ccbin <host-compiler>"
-# (NB issue #505: this must be a single word, "clang++ --gcc-toolchain..." is not supported)
-CUFLAGS += -ccbin $(shell which $(subst ccache ,,$(CXX)))
-
-# Allow newer (unsupported) C++ compilers with older versions of CUDA if ALLOW_UNSUPPORTED_COMPILER_IN_CUDA is set (#504)
-ifneq ($(origin ALLOW_UNSUPPORTED_COMPILER_IN_CUDA),undefined)
-CUFLAGS += -allow-unsupported-compiler
 endif
 
+# Export GPUCC (so that it can also be used in cudacpp_src.mk?)
+export GPUCC
+export GPUFLAGS
+
 #-------------------------------------------------------------------------------
 
-#=== Configure ccache for C++ and CUDA builds
+#=== Configure ccache for C++ and CUDA/HIP builds
 
 # Enable ccache if USECCACHE=1
 ifeq ($(USECCACHE)$(shell echo $(CXX) | grep ccache),1)
@@ -201,15 +260,15 @@ endif
 #ifeq ($(USECCACHE)$(shell echo $(AR) | grep ccache),1)
 #  override AR:=ccache $(AR)
 #endif
-ifneq ($(NVCC),)
-  ifeq ($(USECCACHE)$(shell echo $(NVCC) | grep ccache),1)
-    override NVCC:=ccache $(NVCC)
+ifneq ($(GPUCC),)
+  ifeq ($(USECCACHE)$(shell echo $(GPUCC) | grep ccache),1)
+    override GPUCC:=ccache $(GPUCC)
   endif
 endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure PowerPC-specific compiler flags for C++ and CUDA
+#=== Configure PowerPC-specific compiler flags for C++ and CUDA/HIP
 
 # PowerPC-specific CXX compiler flags (being reviewed)
 ifeq ($(UNAME_P),ppc64le)
@@ -225,9 +284,9 @@ else
   ######CXXFLAGS+= -fno-semantic-interposition # no benefit (neither alone, nor combined with -flto)
 endif
 
-# PowerPC-specific CUDA compiler flags (to be reviewed!)
+# PowerPC-specific CUDA/HIP compiler flags (to be reviewed!)
 ifeq ($(UNAME_P),ppc64le)
-  CUFLAGS+= -Xcompiler -mno-float128
+  GPUFLAGS+= -Xcompiler -mno-float128
 endif
 
 #-------------------------------------------------------------------------------
@@ -235,9 +294,11 @@ endif
 #=== Configure defaults and check if user-defined choices exist for OMPFLAGS, AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
 
 # Set the default OMPFLAGS choice
-ifneq ($(shell $(CXX) --version | egrep '^Intel'),)
+ifneq ($(findstring hipcc,$(GPUCC)),)
+override OMPFLAGS = # disable OpenMP MT when using hipcc #802
+else ifneq ($(shell $(CXX) --version | egrep '^Intel'),)
 override OMPFLAGS = -fopenmp
-###override OMPFLAGS = # disable OpenMP MT on Intel (was ok without nvcc but not ok with nvcc before #578)
+###override OMPFLAGS = # disable OpenMP MT on Intel (was ok without GPUCC but not ok with GPUCC before #578)
 else ifneq ($(shell $(CXX) --version | egrep '^(clang)'),)
 override OMPFLAGS = -fopenmp
 ###override OMPFLAGS = # disable OpenMP MT on clang (was not ok without or with nvcc before #578)
@@ -293,7 +354,10 @@ endif
 
 # Set the default RNDGEN (random number generator) choice
 ifeq ($(RNDGEN),)
-  ifeq ($(NVCC),)
+  ifeq ($(GPUCC),)
+    override RNDGEN = hasNoCurand
+  # Edgecase for HIP compilation
+  else ifeq ($(findstring hipcc,$(GPUCC)),hipcc)
     override RNDGEN = hasNoCurand
   else ifeq ($(RNDGEN),)
     override RNDGEN = hasCurand
@@ -310,7 +374,7 @@ export OMPFLAGS
 
 #-------------------------------------------------------------------------------
 
-#=== Set the CUDA/C++ compiler flags appropriate to user-defined choices of AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
+#=== Set the CUDA/HIP/C++ compiler flags appropriate to user-defined choices of AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
 
 # Set the build flags appropriate to OMPFLAGS
 $(info OMPFLAGS=$(OMPFLAGS))
@@ -368,13 +432,13 @@ CXXFLAGS+= $(AVXFLAGS)
 $(info FPTYPE=$(FPTYPE))
 ifeq ($(FPTYPE),d)
   CXXFLAGS += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
-  CUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
+  GPUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
 else ifeq ($(FPTYPE),f)
   CXXFLAGS += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
-  CUFLAGS  += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
+  GPUFLAGS  += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
 else ifeq ($(FPTYPE),m)
   CXXFLAGS += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
-  CUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
+  GPUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
 else
   $(error Unknown FPTYPE='$(FPTYPE)': only 'd', 'f' and 'm' are supported)
 endif
@@ -383,7 +447,7 @@ endif
 $(info HELINL=$(HELINL))
 ifeq ($(HELINL),1)
   CXXFLAGS += -DMGONGPU_INLINE_HELAMPS
-  CUFLAGS  += -DMGONGPU_INLINE_HELAMPS
+  GPUFLAGS  += -DMGONGPU_INLINE_HELAMPS
 else ifneq ($(HELINL),0)
   $(error Unknown HELINL='$(HELINL)': only '0' and '1' are supported)
 endif
@@ -392,7 +456,7 @@ endif
 $(info HRDCOD=$(HRDCOD))
 ifeq ($(HRDCOD),1)
   CXXFLAGS += -DMGONGPU_HARDCODE_PARAM
-  CUFLAGS  += -DMGONGPU_HARDCODE_PARAM
+  GPUFLAGS  += -DMGONGPU_HARDCODE_PARAM
 else ifneq ($(HRDCOD),0)
   $(error Unknown HRDCOD='$(HRDCOD)': only '0' and '1' are supported)
 endif
@@ -444,11 +508,11 @@ ifeq ($(UNAME_S),Darwin)
   override CULIBFLAGSRPATH2 =
 else
   # RPATH to cuda/cpp libs when linking executables
-  override CXXLIBFLAGSRPATH = -Wl,-rpath,$(LIBDIRRPATH)
-  override CULIBFLAGSRPATH = -Xlinker -rpath,$(LIBDIRRPATH)
+  override CXXLIBFLAGSRPATH = -Wl,-rpath=$(LIBDIRRPATH)
+  override CULIBFLAGSRPATH = -Xlinker -rpath=$(LIBDIRRPATH)
   # RPATH to common lib when linking cuda/cpp libs
-  override CXXLIBFLAGSRPATH2 = -Wl,-rpath,'$$ORIGIN'
-  override CULIBFLAGSRPATH2 = -Xlinker -rpath,'$$ORIGIN'
+  override CXXLIBFLAGSRPATH2 = -Wl,-rpath='$$ORIGIN'
+  override CULIBFLAGSRPATH2 = -Xlinker -rpath='$$ORIGIN'
 endif
 
 # Setting LD_LIBRARY_PATH or DYLD_LIBRARY_PATH in the RUNTIME is no longer necessary (neither on Linux nor on Mac)
@@ -461,7 +525,7 @@ override RUNTIME =
 cxx_main=$(BUILDDIR)/check.exe
 fcxx_main=$(BUILDDIR)/fcheck.exe
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 cu_main=$(BUILDDIR)/gcheck.exe
 fcu_main=$(BUILDDIR)/fgcheck.exe
 else
@@ -492,14 +556,15 @@ $(BUILDDIR)/.build.$(TAG):
 	@touch $(BUILDDIR)/.build.$(TAG)
 
 # Generic target and build rules: objects from CUDA compilation
-ifneq ($(NVCC),)
+# NB: CCBUILDRULEFLAGS includes "-x cu" for nvcc and "-x hip" for hipcc (#810)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/%.o : %.cu *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $(CUBUILDRULEFLAGS) $< -o $@
 
 $(BUILDDIR)/%_cu.o : %.cc *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c -x cu $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $(CCBUILDRULEFLAGS) $< -o $@
 endif
 
 # Generic target and build rules: objects from C++ compilation
@@ -509,11 +574,14 @@ $(BUILDDIR)/%.o : %.cc *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	$(CXX) $(CPPFLAGS) $(CXXFLAGS) -fPIC -c $< -o $@
 
 # Apply special build flags only to CrossSectionKernel.cc and gCrossSectionKernel.cu (no fast math, see #117 and #516)
+# Added edgecase for HIP compilation
 ifeq ($(shell $(CXX) --version | grep ^nvc++),)
 $(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS := $(filter-out -ffast-math,$(CXXFLAGS))
 $(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS += -fno-fast-math
-ifneq ($(NVCC),)
-$(BUILDDIR)/gCrossSectionKernels.o: CUFLAGS += -Xcompiler -fno-fast-math
+ifeq ($(findstring nvcc,$(GPUCC)),nvcc)
+  $(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -Xcompiler -fno-fast-math
+else
+  $(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -fno-fast-math
 endif
 endif
 
@@ -530,10 +598,10 @@ ifeq ($(RNDGEN),hasCurand)
 $(BUILDDIR)/CurandRandomNumberKernel.o: CXXFLAGS += $(CUINC)
 endif
 
-# Avoid "warning: builtin __has_trivial_... is deprecated; use __is_trivially_... instead" in nvcc with icx2023 (#592)
+# Avoid "warning: builtin __has_trivial_... is deprecated; use __is_trivially_... instead" in GPUCC with icx2023 (#592)
 ifneq ($(shell $(CXX) --version | egrep '^(Intel)'),)
-ifneq ($(NVCC),)
-CUFLAGS += -Xcompiler -Wno-deprecated-builtins
+ifneq ($(GPUCC),)
+GPUFLAGS += -Wno-deprecated-builtins
 endif
 endif
 
@@ -541,8 +609,8 @@ endif
 # This patch does remove the warning, but I prefer to keep it disabled for the moment...
 ###ifneq ($(shell $(CXX) --version | egrep '^(clang|Apple clang|Intel)'),)
 ###$(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS += -Wno-overriding-t-option
-###ifneq ($(NVCC),)
-###$(BUILDDIR)/gCrossSectionKernels.o: CUFLAGS += -Xcompiler -Wno-overriding-t-option
+###ifneq ($(GPUCC),)
+###$(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -Xcompiler -Wno-overriding-t-option
 ###endif
 ###endif
 
@@ -569,7 +637,7 @@ MG5AMC_CXXLIB = mg5amc_$(processid_short)_cpp
 cxx_objects_lib=$(BUILDDIR)/CPPProcess.o $(BUILDDIR)/MatrixElementKernels.o $(BUILDDIR)/BridgeKernels.o $(BUILDDIR)/CrossSectionKernels.o
 cxx_objects_exe=$(BUILDDIR)/CommonRandomNumberKernel.o $(BUILDDIR)/RamboSamplingKernels.o
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 MG5AMC_CULIB = mg5amc_$(processid_short)_cuda
 cu_objects_lib=$(BUILDDIR)/gCPPProcess.o $(BUILDDIR)/gMatrixElementKernels.o $(BUILDDIR)/gBridgeKernels.o $(BUILDDIR)/gCrossSectionKernels.o
 cu_objects_exe=$(BUILDDIR)/gCommonRandomNumberKernel.o $(BUILDDIR)/gRamboSamplingKernels.o
@@ -581,11 +649,17 @@ $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: cxx_objects_lib += $(BUILDDIR)/fbridge.o
 $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib)
 	$(CXX) -shared -o $@ $(cxx_objects_lib) $(CXXLIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: $(BUILDDIR)/fbridge_cu.o
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: cu_objects_lib += $(BUILDDIR)/fbridge_cu.o
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cu_objects_lib)
-	$(NVCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+# Bypass std::filesystem completely to ease portability on LUMI #803
+#ifneq ($(findstring hipcc,$(GPUCC)),)
+#	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB) -lstdc++fs
+#else
+#	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+#endif
 endif
 
 #-------------------------------------------------------------------------------
@@ -602,16 +676,16 @@ $(cxx_main): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PAT
 $(cxx_main): $(BUILDDIR)/check_sa.o $(LIBDIR)/lib$(MG5AMC_CXXLIB).so $(cxx_objects_exe) $(BUILDDIR)/CurandRandomNumberKernel.o
 	$(CXX) -o $@ $(BUILDDIR)/check_sa.o $(OMPFLAGS) -ldl -pthread $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe) $(BUILDDIR)/CurandRandomNumberKernel.o $(CURANDLIBFLAGS)
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(cu_main): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(cu_main): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(cu_main): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(cu_main): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 else ifneq ($(shell $(CXX) --version | grep ^nvc++),) # support nvc++ #531
 $(cu_main): LIBFLAGS += -L$(patsubst %bin/nvc++,%lib,$(subst ccache ,,$(CXX))) -lnvhpcatm -lnvcpumath -lnvc
 endif
 $(cu_main): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(cu_main): $(BUILDDIR)/gcheck_sa.o $(LIBDIR)/lib$(MG5AMC_CULIB).so $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o
-	$(NVCC) -o $@ $(BUILDDIR)/gcheck_sa.o $(CUARCHFLAGS) $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o $(CURANDLIBFLAGS)
+	$(GPUCC) -o $@ $(BUILDDIR)/gcheck_sa.o $(CUARCHFLAGS) $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o $(CURANDLIBFLAGS)
 endif
 
 #-------------------------------------------------------------------------------
@@ -635,19 +709,27 @@ $(fcxx_main): LIBFLAGS += -L$(shell dirname $(shell $(FC) --print-file-name libg
 endif
 $(fcxx_main): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(fcxx_main): $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler.o $(LIBDIR)/lib$(MG5AMC_CXXLIB).so $(cxx_objects_exe)
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(BUILDDIR)/fcheck_sa.o $(OMPFLAGS) $(BUILDDIR)/fsampler.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe) -lstdc++
+else
 	$(CXX) -o $@ $(BUILDDIR)/fcheck_sa.o $(OMPFLAGS) $(BUILDDIR)/fsampler.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe)
+endif
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(fcu_main): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(fcu_main): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(fcu_main): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(fcu_main): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 endif
 ifeq ($(UNAME_S),Darwin)
 $(fcu_main): LIBFLAGS += -L$(shell dirname $(shell $(FC) --print-file-name libgfortran.dylib)) # add path to libgfortran on Mac #375
 endif
 $(fcu_main): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(fcu_main): $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBDIR)/lib$(MG5AMC_CULIB).so $(cu_objects_exe)
-	$(NVCC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe)
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) -lstdc++ -L$(shell dirname $(shell $(GPUCC) -print-prog-name=clang))/../../lib -lamdhip64
+else
+	$(GPUCC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe)
+endif
 endif
 
 #-------------------------------------------------------------------------------
@@ -659,7 +741,7 @@ $(BUILDDIR)/testxxx.o: testxxx_cc_ref.txt
 $(testmain): $(BUILDDIR)/testxxx.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/testxxx.o # Comment out this line to skip the C++ test of xxx functions
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/testxxx_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/testxxx_cu.o: INCFLAGS += $(GTESTINC)
 $(BUILDDIR)/testxxx_cu.o: testxxx_cc_ref.txt
@@ -672,7 +754,7 @@ $(BUILDDIR)/testmisc.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/testmisc.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/testmisc.o # Comment out this line to skip the C++ miscellaneous tests
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/testmisc_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/testmisc_cu.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/testmisc_cu.o
@@ -684,12 +766,12 @@ $(BUILDDIR)/runTest.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/runTest.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/runTest.o
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/runTest_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/runTest_cu.o: INCFLAGS += $(GTESTINC)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(testmain): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(testmain): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(testmain): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(testmain): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 else ifneq ($(shell $(CXX) --version | grep ^nvc++),) # support nvc++ #531
 $(testmain): LIBFLAGS += -L$(patsubst %bin/nvc++,%lib,$(subst ccache ,,$(CXX))) -lnvhpcatm -lnvcpumath -lnvc
 endif
@@ -713,14 +795,23 @@ $(testmain): LIBFLAGS += -lgomp
 endif
 endif
 
-ifeq ($(NVCC),) # link only runTest.o
+# Bypass std::filesystem completely to ease portability on LUMI #803
+#ifneq ($(findstring hipcc,$(GPUCC)),)
+#$(testmain): LIBFLAGS += -lstdc++fs
+#endif
+
+ifeq ($(GPUCC),) # link only runTest.o
 $(testmain): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(testmain): $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib) $(cxx_objects_exe) $(GTESTLIBS)
 	$(CXX) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) -ldl -pthread $(LIBFLAGS)
 else # link both runTest.o and runTest_cu.o
 $(testmain): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(testmain): $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) $(GTESTLIBS)
-	$(NVCC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) -lcuda
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) $(CUDATESTFLAGS) -lstdc++ -lpthread  -L$(shell dirname $(shell $(GPUCC) -print-prog-name=clang))/../../lib -lamdhip64
+else
+	$(GPUCC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) $(CUDATESTFLAGS)
+endif
 endif
 
 # Use target gtestlibs to build only googletest
@@ -829,9 +920,9 @@ ifeq ($(USECCACHE),1)
 	ccache --version | head -1
 endif
 	@echo ""
-	@echo NVCC=$(NVCC)
-ifneq ($(NVCC),)
-	$(NVCC) --version
+	@echo GPUCC=$(GPUCC)
+ifneq ($(GPUCC),)
+	$(GPUCC) --version
 endif
 	@echo ""
 	@echo CXX=$(CXX)
@@ -850,7 +941,7 @@ endif
 
 # Target: check (run the C++ test executable)
 # [NB THIS IS WHAT IS USED IN THE GITHUB CI!]
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 check: runTest cmpFcheck cmpFGcheck
 else
 check: runTest cmpFcheck
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/fbridge.cc b/epochX/cudacpp/ee_mumu.mad/SubProcesses/fbridge.cc
index 2d2b36d560..22ce3f5115 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/fbridge.cc
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/fbridge.cc
@@ -1,11 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Roiser (Oct 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "Bridge.h"
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 
 extern "C"
 {
@@ -22,7 +22,7 @@ extern "C"
    * Using the same Fortran MadEvent code, linking to the hetrerogeneous library would allow access to both CPU and GPU implementations.
    * The specific heterogeneous configuration (how many GPUs, how many threads on each CPU, etc) could be loaded in CUDA/C++ from a data file.
    */
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -46,8 +46,8 @@ extern "C"
    */
   void fbridgecreate_( CppObjectInFortran** ppbridge, const int* pnevtF, const int* pnparF, const int* pnp4F )
   {
-#ifdef __CUDACC__
-    CudaRuntime::setUp();
+#ifdef MGONGPUCPP_GPUIMPL
+    GpuRuntime::setUp();
 #endif
     // (NB: CPPProcess::initProc no longer needs to be executed here because it is called in the Bridge constructor)
     // FIXME: disable OMP in Bridge when called from Fortran
@@ -65,8 +65,8 @@ extern "C"
     Bridge<FORTRANFPTYPE>* pbridge = dynamic_cast<Bridge<FORTRANFPTYPE>*>( *ppbridge );
     if( pbridge == 0 ) throw std::runtime_error( "fbridgedelete_: invalid Bridge address" );
     delete pbridge;
-#ifdef __CUDACC__
-    CudaRuntime::tearDown();
+#ifdef MGONGPUCPP_GPUIMPL
+    GpuRuntime::tearDown();
 #endif
   }
 
@@ -96,7 +96,7 @@ extern "C"
   {
     Bridge<FORTRANFPTYPE>* pbridge = dynamic_cast<Bridge<FORTRANFPTYPE>*>( *ppbridge );
     if( pbridge == 0 ) throw std::runtime_error( "fbridgesequence_: invalid Bridge address" );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Use the device/GPU implementation in the CUDA library
     // (there is also a host implementation in this library)
     pbridge->gpu_sequence( momenta, gs, rndhel, rndcol, *pchannelId, mes, selhel, selcol );
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/fsampler.cc b/epochX/cudacpp/ee_mumu.mad/SubProcesses/fsampler.cc
index 2fb445372d..3743934f41 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/fsampler.cc
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/fsampler.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Feb 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "mgOnGpuConfig.h"
 
@@ -13,7 +13,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -40,7 +40,7 @@ namespace mg5amcCpu
   private:
     const int m_nevt; // The number of events in each iteration
     int m_iiter;      // The iteration counter (for random number seeding)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     HostBufferRndNumMomenta m_hstRndmom; // Memory buffers for random numbers
     HostBufferMomenta m_hstMomenta;      // Memory buffers for momenta
     HostBufferWeights m_hstWeights;      // Memory buffers for sampling weights
@@ -105,7 +105,7 @@ namespace mg5amcCpu
 
 extern "C"
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/runTest.cc b/epochX/cudacpp/ee_mumu.mad/SubProcesses/runTest.cc
index d4a760a71b..de327f2321 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/runTest.cc
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/runTest.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Hageboeck (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*xxx to run only testxxx.cc tests
 //----------------------------------------------------------------------------
@@ -18,7 +18,7 @@
 #include "RandomNumberKernels.h"
 #include "epoch_process_id.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using namespace mg5amcGpu;
 #else
 using namespace mg5amcCpu;
@@ -35,7 +35,7 @@ struct CUDA_CPU_TestBase : public TestDriverBase
     : TestDriverBase( npar, refFileName ) {}
 };
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 struct CPUTest : public CUDA_CPU_TestBase
 {
   // Struct data members (process, and memory structures for random numbers, momenta, matrix elements and weights on host and device)
@@ -119,7 +119,7 @@ struct CPUTest : public CUDA_CPU_TestBase
 };
 #endif
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 struct CUDATest : public CUDA_CPU_TestBase
 {
   // Reset the device when our test goes out of scope. Note that this should happen after
@@ -128,7 +128,7 @@ struct CUDATest : public CUDA_CPU_TestBase
   {
     ~DeviceReset()
     {
-      checkCuda( cudaDeviceReset() ); // this is needed by cuda-memcheck --leak-check full
+      checkGpu( gpuDeviceReset() ); // this is needed by cuda-memcheck --leak-check full
     }
   } deviceResetter;
 
@@ -256,7 +256,7 @@ INSTANTIATE_TEST_SUITE_P( prefix, \
                           test_suite_name, \
                           testing::Values( new CUDATest( MG_EPOCH_REFERENCE_FILE_NAME ) ) );
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 MG_INSTANTIATE_TEST_SUITE_GPU( XTESTID_GPU( MG_EPOCH_PROCESS_ID ), MadgraphTest );
 #else
 MG_INSTANTIATE_TEST_SUITE_CPU( XTESTID_CPU( MG_EPOCH_PROCESS_ID ), MadgraphTest );
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/testmisc.cc b/epochX/cudacpp/ee_mumu.mad/SubProcesses/testmisc.cc
index 895d6eeb56..ba9e59a8a3 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/testmisc.cc
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/testmisc.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*misc to run only testmisc.cc tests
 //----------------------------------------------------------------------------
@@ -17,7 +17,7 @@
 #include <sstream>
 #include <typeinfo>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #define TESTID( s ) s##_GPU_MISC
 #else
 #define TESTID( s ) s##_CPU_MISC
@@ -26,7 +26,7 @@
 #define XTESTID( s ) TESTID( s )
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -59,7 +59,7 @@ namespace mg5amcCpu
 
 TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testmisc )
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/ee_mumu.mad/SubProcesses/testxxx.cc b/epochX/cudacpp/ee_mumu.mad/SubProcesses/testxxx.cc
index 3361fe5aa9..e5167de00c 100644
--- a/epochX/cudacpp/ee_mumu.mad/SubProcesses/testxxx.cc
+++ b/epochX/cudacpp/ee_mumu.mad/SubProcesses/testxxx.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Apr 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*xxx to run only testxxx.cc tests
 //----------------------------------------------------------------------------
@@ -24,7 +24,7 @@
 #include <iomanip>
 #include <iostream>
 #include <vector>
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #define TESTID( s ) s##_GPU_XXX
 #else
 #define TESTID( s ) s##_CPU_XXX
@@ -32,7 +32,7 @@
 
 #define XTESTID( s ) TESTID( s )
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -42,7 +42,7 @@ namespace mg5amcCpu
   int FPEhandlerIevt = -1;
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU): '" << FPEhandlerMessage << "' ievt=" << FPEhandlerIevt << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU neppV=" << neppV << "): '" << FPEhandlerMessage << "' ievt=" << FPEhandlerIevt << std::endl;
@@ -53,7 +53,7 @@ namespace mg5amcCpu
 
 TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -77,7 +77,7 @@ TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
   assert( nevt % neppM == 0 ); // nevt must be a multiple of neppM
   assert( nevt % neppV == 0 ); // nevt must be a multiple of neppV
   // Fill in the input momenta
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   mg5amcGpu::PinnedHostBufferMomenta hstMomenta( nevt ); // AOSOA[npagM][npar=4][np4=4][neppM]
 #else
   mg5amcCpu::HostBufferMomenta hstMomenta( nevt ); // AOSOA[npagM][npar=4][np4=4][neppM]
@@ -322,7 +322,7 @@ TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
   {
     for( int ievt = 0; ievt < nevt; ievt++ )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       using namespace mg5amcGpu;
 #else
       using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/ee_mumu.mad/bin/internal/launch_plugin.py b/epochX/cudacpp/ee_mumu.mad/bin/internal/launch_plugin.py
index 3b09713e12..90af6b7053 100644
--- a/epochX/cudacpp/ee_mumu.mad/bin/internal/launch_plugin.py
+++ b/epochX/cudacpp/ee_mumu.mad/bin/internal/launch_plugin.py
@@ -103,7 +103,7 @@ def default_setup(self):
                        fct_mod=(self.reset_makeopts,(),{}),
                        allowed=['auto', 'none', 'sse4', 'avx2','512y','512z'])
         self.add_param('cudacpp_backend', 'CPP', include=False, hidden=False,
-                       allowed=['Fortan', 'CPP', 'CUDA'])
+                       allowed=['Fortran', 'CPP', 'CUDA'])
         self['vector_size'] = 16 # already setup in default class (just change value)
         self['aloha_flag'] = '--fast-math'
         self['matrix_flag'] = '-O3'
diff --git a/epochX/cudacpp/ee_mumu.mad/src/HelAmps_sm.h b/epochX/cudacpp/ee_mumu.mad/src/HelAmps_sm.h
index 9fa30cfd7f..e878fcd28e 100644
--- a/epochX/cudacpp/ee_mumu.mad/src/HelAmps_sm.h
+++ b/epochX/cudacpp/ee_mumu.mad/src/HelAmps_sm.h
@@ -5,7 +5,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -28,7 +28,7 @@
 //#include <iomanip>
 //#include <iostream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/ee_mumu.mad/src/Parameters_sm.cc b/epochX/cudacpp/ee_mumu.mad/src/Parameters_sm.cc
index 0b4be4d5ed..cffc5d3bff 100644
--- a/epochX/cudacpp/ee_mumu.mad/src/Parameters_sm.cc
+++ b/epochX/cudacpp/ee_mumu.mad/src/Parameters_sm.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -17,7 +17,7 @@
 #include <iomanip>
 #include <iostream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using namespace mg5amcGpu;
 #else
 using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/ee_mumu.mad/src/Parameters_sm.h b/epochX/cudacpp/ee_mumu.mad/src/Parameters_sm.h
index 64d0b8e761..2a6d960581 100644
--- a/epochX/cudacpp/ee_mumu.mad/src/Parameters_sm.h
+++ b/epochX/cudacpp/ee_mumu.mad/src/Parameters_sm.h
@@ -27,7 +27,7 @@
 #include "read_slha.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -93,7 +93,7 @@ namespace mg5amcCpu
 #include <limits>
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -218,7 +218,7 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -235,7 +235,7 @@ namespace mg5amcCpu
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wunused-variable"  // e.g. <<warning: unused variable ‘mdl_G__exp__2’ [-Wunused-variable]>>
 #pragma GCC diagnostic ignored "-Wunused-parameter" // e.g. <<warning: unused parameter ‘G’ [-Wunused-parameter]>>
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #pragma nv_diagnostic push
 #pragma nv_diag_suppress 177 // e.g. <<warning #177-D: variable "mdl_G__exp__2" was declared but never referenced>>
 #endif
@@ -259,7 +259,7 @@ namespace mg5amcCpu
       // End SM implementation - no special handling of vectors of floats as in EFT (#439)
       return out;
     }
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #pragma GCC diagnostic pop
 #pragma nv_diagnostic pop
 #endif
diff --git a/epochX/cudacpp/ee_mumu.mad/src/cudacpp_src.mk b/epochX/cudacpp/ee_mumu.mad/src/cudacpp_src.mk
index d4cc628aec..b2b9da5288 100644
--- a/epochX/cudacpp/ee_mumu.mad/src/cudacpp_src.mk
+++ b/epochX/cudacpp/ee_mumu.mad/src/cudacpp_src.mk
@@ -19,7 +19,7 @@ SHELL := /bin/bash
 #=== Configure common compiler flags for CUDA and C++
 
 INCFLAGS = -I.
-OPTFLAGS = -O3 # this ends up in CUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
+OPTFLAGS = -O3 # this ends up in GPUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
 
 #-------------------------------------------------------------------------------
 
@@ -45,13 +45,13 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure the CUDA compiler (note: NVCC is already exported including ccache)
+#=== Configure the CUDA compiler (note: GPUCC is already exported including ccache)
 
-###$(info NVCC=$(NVCC))
+###$(info GPUCC=$(GPUCC))
 
 #-------------------------------------------------------------------------------
 
-#=== Configure ccache for C++ builds (note: NVCC is already exported including ccache)
+#=== Configure ccache for C++ builds (note: GPUCC is already exported including ccache)
 
 # Enable ccache if USECCACHE=1
 ifeq ($(USECCACHE)$(shell echo $(CXX) | grep ccache),1)
@@ -92,6 +92,13 @@ endif
 ###$(info OMPFLAGS=$(OMPFLAGS))
 CXXFLAGS += $(OMPFLAGS)
 
+# Add correct flags for nvcc (-x cu) and hipcc (-x hip) for GPU code (see #810)
+ifeq ($(findstring nvcc,$(GPUCC)),nvcc)
+  GPUFLAGS += -Xcompiler -fPIC -c -x cu
+else ifeq ($(findstring hipcc,$(GPUCC)),hipcc)
+  GPUFLAGS += -fPIC -c -x hip
+endif
+
 # Set the build flags appropriate to each AVX choice (example: "make AVX=none")
 # [NB MGONGPU_PVW512 is needed because "-mprefer-vector-width=256" is not exposed in a macro]
 # [See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96476]
@@ -253,20 +260,20 @@ $(BUILDDIR)/%.o : %.cc *.h $(BUILDDIR)/.build.$(TAG)
 # Generic target and build rules: objects from CUDA compilation
 $(BUILDDIR)/%_cu.o : %.cc *.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c -x cu $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $< -o $@
 
 #-------------------------------------------------------------------------------
 
 cxx_objects=$(addprefix $(BUILDDIR)/, Parameters_sm.o read_slha.o)
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 cu_objects=$(addprefix $(BUILDDIR)/, Parameters_sm_cu.o)
 endif
 
 # Target (and build rules): common (src) library
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so : $(cxx_objects) $(cu_objects)
 	@if [ ! -d $(LIBDIR) ]; then echo "mkdir -p $(LIBDIR)"; mkdir -p $(LIBDIR); fi
-	$(NVCC) -shared -o $@ $(cxx_objects) $(cu_objects) $(LDFLAGS)
+	$(GPUCC) -shared -o $@ $(cxx_objects) $(cu_objects) $(LDFLAGS)
 else
 $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so : $(cxx_objects)
 	@if [ ! -d $(LIBDIR) ]; then echo "mkdir -p $(LIBDIR)"; mkdir -p $(LIBDIR); fi
diff --git a/epochX/cudacpp/ee_mumu.mad/src/mgOnGpuConfig.h b/epochX/cudacpp/ee_mumu.mad/src/mgOnGpuConfig.h
index 80032e528b..6bde4466d0 100644
--- a/epochX/cudacpp/ee_mumu.mad/src/mgOnGpuConfig.h
+++ b/epochX/cudacpp/ee_mumu.mad/src/mgOnGpuConfig.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jul 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUCONFIG_H
 #define MGONGPUCONFIG_H 1
@@ -10,12 +10,26 @@
 // There are two different code bases for standalone_cudacpp (without multichannel) and madevent+cudacpp (with multichannel)
 #define MGONGPU_SUPPORTS_MULTICHANNEL 1
 
+// Is this a GPU (CUDA, HIP) or CPU implementation?
+#ifdef __CUDACC__
+#define MGONGPUCPP_GPUIMPL cuda
+#elif defined __HIPCC__
+#define MGONGPUCPP_GPUIMPL hip
+#include "hip/hip_runtime.h" // needed for blockDim, blockIdx, threadIdx: better in mgOnGpuConfig.h than in GpuAbstraction.h
+#else
+#undef MGONGPUCPP_GPUIMPL
+#endif
+
 // ** NB1 Throughputs (e.g. 6.8E8) are events/sec for "./gcheck.exe -p 65536 128 12"
 // ** NB2 Baseline on b7g47n0004 fluctuates (probably depends on load on other VMs)
 
 // Choose if curand is supported for generating random numbers
+// For HIP, by default, do not use curand (common random numbers will be used instead)
 // For both CUDA and C++, by default, do not inline, but allow this macro to be set from outside with e.g. -DMGONGPU_HAS_NO_CURAND
-// (there exist CUDA installations, e.g. using the HPC package, which do not include curand - see PR #784)
+// (there exist CUDA installations, e.g. using the HPC package, which do not include curand - see PR #784 and #785)
+#if defined __HIPCC__
+#define MGONGPU_HAS_NO_CURAND 1
+#else
 //#ifdef __CUDACC__
 //#undef MGONGPU_HAS_NO_CURAND // default
 ////#define MGONGPU_HAS_NO_CURAND 1
@@ -23,6 +37,7 @@
 //#undef MGONGPU_HAS_NO_CURAND // default
 ////#define MGONGPU_HAS_NO_CURAND 1
 //#endif
+#endif
 
 // Choose floating point precision (for everything but color algebra #537)
 // If one of these macros has been set from outside with e.g. -DMGONGPU_FPTYPE_FLOAT, nothing happens (issue #167)
@@ -54,23 +69,31 @@
 //#undef MGONGPU_HARDCODE_PARAM // default
 ////#define MGONGPU_HARDCODE_PARAM 1
 
-// Complex type in c++: std::complex or cxsmpl (CHOOSE ONLY ONE)
-#ifndef __CUDACC__
-//#define MGONGPU_CPPCXTYPE_STDCOMPLEX 1 // ~8 percent slower on float, same on double (5.1E6/double, 9.4E6/float)
-#define MGONGPU_CPPCXTYPE_CXSMPL 1 // new default (5.1E6/double, 10.2E6/float)
-#endif
-
-// Complex type in cuda: thrust or cucomplex or cxsmpl (CHOOSE ONLY ONE)
+// Complex type in CUDA: thrust or cucomplex or cxsmpl (CHOOSE ONLY ONE)
+// (NB THIS IS MGONGPU_*CU*CXTYPE_xxx)
 #ifdef __CUDACC__
 #define MGONGPU_CUCXTYPE_THRUST 1 // default (~1.15E9/double, ~3.2E9/float)
 //#define MGONGPU_CUCXTYPE_CUCOMPLEX 1 // ~10 percent slower (1.03E9/double, ~2.8E9/float)
 //#define MGONGPU_CUCXTYPE_CXSMPL 1 // ~10 percent slower (1.00E9/double, ~2.9E9/float)
+
+// Complex type in HIP: cxsmpl (ONLY ONE OPTION POSSIBLE? #810)
+// (NB THIS IS MGONGPU_*HIP*CXTYPE_xxx)
+#elif defined __HIPCC__
+#define MGONGPU_HIPCXTYPE_CXSMPL 1 // default for HIP
+
+// Complex type in C++: std::complex or cxsmpl (CHOOSE ONLY ONE)
+// (NB THIS IS MGONGPU_*CPP*CXTYPE_xxx)
+#else
+//#define MGONGPU_CPPCXTYPE_STDCOMPLEX 1 // ~8 percent slower on float, same on double (5.1E6/double, 9.4E6/float)
+#define MGONGPU_CPPCXTYPE_CXSMPL 1 // new default (5.1E6/double, 10.2E6/float)
 #endif
 
-// Cuda nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
+// CUDA nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
 #ifdef __CUDACC__
-#undef MGONGPU_NSIGHT_DEBUG // default
+#undef MGONGPU_NSIGHT_DEBUG // default in CUDA
 //#define MGONGPU_NSIGHT_DEBUG 1
+#else
+#undef MGONGPU_NSIGHT_DEBUG // only option in HIP or C++
 #endif
 
 // SANITY CHECKS (floating point precision for everything but color algebra #537)
@@ -86,17 +109,21 @@
 #error You cannot use double precision for color algebra and single precision elsewhere
 #endif
 
-// SANITY CHECKS (c++ complex number implementation)
-#ifndef __CUDACC__
-#if defined MGONGPU_CPPCXTYPE_STDCOMPLEX and defined MGONGPU_CPPCXTYPE_CXSMPL
-#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CPPCXTYPE_STDCOMPLEX or MGONGPU_CPPCXTYPE_CXSMPL
+// SANITY CHECKS (CUDA complex number implementation)
+#ifdef __CUDACC__
+#if defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CUCOMPLEX
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX for CUDA
+#elif defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CXSMPL for CUDA
+#elif defined MGONGPU_CUCXTYPE_CUCOMPLEX and defined MGONGPU_CUCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE OF MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL for CUDA
 #endif
 #endif
 
-// SANITY CHECKS (cuda complex number implementation)
-#ifdef __CUDACC__
-#if defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CUCOMPLEX and defined MGONGPU_CUCXTYPE_CXSMPL
-#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL
+// SANITY CHECKS (C++ complex number implementation)
+#ifndef MGONGPUCPP_GPUIMPL
+#if defined MGONGPU_CPPCXTYPE_STDCOMPLEX and defined MGONGPU_CPPCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CPPCXTYPE_STDCOMPLEX or MGONGPU_CPPCXTYPE_CXSMPL for C++
 #endif
 #endif
 
@@ -134,7 +161,7 @@ namespace mgOnGpu
   // Alignment requirement for using reinterpret_cast with SIMD vectorized code
   // (using reinterpret_cast with non aligned memory may lead to segmentation faults!)
   // Only needed for C++ code but can be enforced also in NVCC builds of C++ code using CUDA>=11.2 and C++17 (#318, #319, #333)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   constexpr int cppAlign = 64; // alignment requirement for SIMD vectorization (64-byte i.e. 512-bit)
 #endif
 
@@ -145,7 +172,7 @@ using mgOnGpu::fptype;
 using mgOnGpu::fptype2;
 
 // C++ SIMD vectorization width (this will be used to set neppV)
-#ifdef __CUDACC__ // CUDA implementation has no SIMD
+#ifdef MGONGPUCPP_GPUIMPL // CUDA and HIP implementations have no SIMD
 #undef MGONGPU_CPPSIMD
 #elif defined __AVX512VL__ && defined MGONGPU_PVW512 // C++ "512z" AVX512 with 512 width (512-bit ie 64-byte): 8 (DOUBLE) or 16 (FLOAT)
 #ifdef MGONGPU_FPTYPE_DOUBLE
@@ -175,9 +202,9 @@ using mgOnGpu::fptype2;
 #undef MGONGPU_CPPSIMD
 #endif
 
-// Cuda nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
+// CUDA nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
 // Arguments (not used so far): text is __FUNCTION__, code is 0 (start) or 1 (end)
-#if defined __CUDACC__ && defined MGONGPU_NSIGHT_DEBUG /* clang-format off */
+#if defined __CUDA__ && defined MGONGPU_NSIGHT_DEBUG /* clang-format off */
 #define mgDebugDeclare() __shared__ float mgDebugCounter[mgOnGpu::ntpbMAX];
 #define mgDebugInitialise() { mgDebugCounter[threadIdx.x] = 0; }
 #define mgDebug( code, text ) { mgDebugCounter[threadIdx.x] += 1; }
@@ -189,8 +216,8 @@ using mgOnGpu::fptype2;
 #define mgDebugFinalise() { /*noop*/ }
 #endif /* clang-format on */
 
-// Define empty CUDA declaration specifiers for C++
-#ifndef __CUDACC__
+// Define empty CUDA/HIP declaration specifiers for C++
+#ifndef MGONGPUCPP_GPUIMPL
 #define __global__
 #define __host__
 #define __device__
diff --git a/epochX/cudacpp/ee_mumu.mad/src/mgOnGpuCxtypes.h b/epochX/cudacpp/ee_mumu.mad/src/mgOnGpuCxtypes.h
index ca9a9f00c0..7ede1dbfae 100644
--- a/epochX/cudacpp/ee_mumu.mad/src/mgOnGpuCxtypes.h
+++ b/epochX/cudacpp/ee_mumu.mad/src/mgOnGpuCxtypes.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022, based on earlier work by D. Smith) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUCXTYPES_H
 #define MGONGPUCXTYPES_H 1
@@ -19,7 +19,7 @@
 #include <complex>
 
 // Complex type in cuda: thrust or cucomplex or cxsmpl
-#ifdef __CUDACC__
+#ifdef __CUDACC__ // this must be __CUDAC__ (not MGONGPUCPP_GPUIMPL)
 #if defined MGONGPU_CUCXTYPE_THRUST
 #pragma clang diagnostic push
 #pragma clang diagnostic ignored "-Wtautological-compare" // for icpx2021/clang13 (https://stackoverflow.com/a/15864661)
@@ -30,8 +30,13 @@
 #elif not defined MGONGPU_CUCXTYPE_CXSMPL
 #error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL
 #endif
+// Complex type in HIP: cxsmpl
+#elif defined __HIPCC__
+#if not defined MGONGPU_HIPCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_HIPCXTYPE_CXSMPL
+#endif
 #else
-// Complex type in c++: std::complex or cxsmpl
+// Complex type in c++ or HIP: std::complex or cxsmpl
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
 #include <cmath>
 #elif not defined MGONGPU_CPPCXTYPE_CXSMPL
@@ -82,7 +87,7 @@ namespace mgOnGpu /* clang-format off */
 using mgOnGpu::cxsmpl;
 
 // Printout to stream for user defined types
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -92,7 +97,7 @@ namespace mg5amcCpu
   inline __host__ std::ostream&
   operator<<( std::ostream& out, const cxsmpl<FP>& c )
   {
-    out << std::complex( c.real(), c.imag() );
+    out << std::complex<FP>( c.real(), c.imag() );
     return out;
   }
 
@@ -215,14 +220,14 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
   // --- Type definitions (complex type: cxtype)
-#ifdef __CUDACC__ // cuda
+#ifdef __CUDACC__ // this must be __CUDAC__ (not MGONGPUCPP_GPUIMPL)
 #if defined MGONGPU_CUCXTYPE_THRUST
   typedef thrust::complex<fptype> cxtype;
 #elif defined MGONGPU_CUCXTYPE_CUCOMPLEX
@@ -255,13 +260,13 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
-#if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+#if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
 
   //------------------------------
   // CUDA or C++ - using cxsmpl
@@ -303,11 +308,11 @@ namespace mg5amcCpu
     return cxmake( c.real(), c.imag() );
   }
 
-#endif // #if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+#endif // #if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
 
   //==========================================================================
 
-#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_THRUST // cuda + thrust
+#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_THRUST // cuda + thrust (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // CUDA - using thrust::complex
@@ -347,7 +352,7 @@ namespace mg5amcCpu
 
   //==========================================================================
 
-#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_CUCOMPLEX // cuda + cucomplex
+#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_CUCOMPLEX // cuda + cucomplex (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // CUDA - using cuComplex
@@ -566,7 +571,7 @@ namespace mg5amcCpu
 
   //==========================================================================
 
-#if not defined __CUDACC__ and defined MGONGPU_CPPCXTYPE_STDCOMPLEX // c++ + stdcomplex
+#if not defined __CUDACC__ and defined MGONGPU_CPPCXTYPE_STDCOMPLEX // c++/hip + stdcomplex (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // C++ - using std::complex
@@ -633,7 +638,7 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/ee_mumu.mad/src/mgOnGpuFptypes.h b/epochX/cudacpp/ee_mumu.mad/src/mgOnGpuFptypes.h
index 905c97d700..fa3a02664b 100644
--- a/epochX/cudacpp/ee_mumu.mad/src/mgOnGpuFptypes.h
+++ b/epochX/cudacpp/ee_mumu.mad/src/mgOnGpuFptypes.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUFPTYPES_H
 #define MGONGPUFPTYPES_H 1
@@ -12,7 +12,7 @@
 #include <cmath>
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL // cuda
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -20,7 +20,7 @@ namespace mg5amcCpu
 {
   //==========================================================================
 
-#ifdef __CUDACC__ // cuda
+#ifdef MGONGPUCPP_GPUIMPL // cuda
 
   //------------------------------
   // Floating point types - Cuda
@@ -64,11 +64,11 @@ namespace mg5amcCpu
 #endif
   }
 
-#endif // #ifdef __CUDACC__
+#endif // #ifdef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
   //------------------------------
   // Floating point types - C++
@@ -92,7 +92,7 @@ namespace mg5amcCpu
     return std::sqrt( f );
   }
 
-#endif // #ifndef __CUDACC__
+#endif // #ifndef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
diff --git a/epochX/cudacpp/ee_mumu.mad/src/mgOnGpuVectors.h b/epochX/cudacpp/ee_mumu.mad/src/mgOnGpuVectors.h
index e1299ba81e..cdae04326b 100644
--- a/epochX/cudacpp/ee_mumu.mad/src/mgOnGpuVectors.h
+++ b/epochX/cudacpp/ee_mumu.mad/src/mgOnGpuVectors.h
@@ -32,7 +32,7 @@
 #endif
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -131,7 +131,7 @@ namespace mg5amcCpu
 #endif
 #endif
 
-#else // i.e #ifndef MGONGPU_CPPSIMD (this includes #ifdef __CUDACC__)
+#else // i.e #ifndef MGONGPU_CPPSIMD (this includes #ifdef MGONGPUCPP_GPUIMPL)
 
   const int neppV = 1;
 
@@ -153,13 +153,13 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
   // Printout to stream for user defined types
 
@@ -805,11 +805,11 @@ namespace mg5amcCpu
 
 #endif // #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
 
-#endif // #ifndef __CUDACC__
+#endif // #ifndef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
   //------------------------------
   // Vector types - CUDA
@@ -853,12 +853,12 @@ namespace mg5amcCpu
     return mask;
   }
 
-#endif // #ifdef __CUDACC__
+#endif // #ifdef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
   // Scalar-or-vector types: scalar in CUDA, vector or scalar in C++
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   typedef bool bool_sv;
   typedef fptype fptype_sv;
   typedef fptype2 fptype2_sv;
@@ -879,7 +879,7 @@ namespace mg5amcCpu
 #endif
 
   // Scalar-or-vector zeros: scalar in CUDA, vector or scalar in C++
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   inline __host__ __device__ cxtype cxzero_sv(){ return cxtype( 0, 0 ); }
 #elif defined MGONGPU_CPPSIMD
   inline cxtype_v cxzero_sv() { return cxtype_v(); } // RRRR=0000 IIII=0000
diff --git a/epochX/cudacpp/ee_mumu.mad/src/rambo.h b/epochX/cudacpp/ee_mumu.mad/src/rambo.h
index e02ea52496..cd7e1008ea 100644
--- a/epochX/cudacpp/ee_mumu.mad/src/rambo.h
+++ b/epochX/cudacpp/ee_mumu.mad/src/rambo.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -18,7 +18,7 @@
 #include <iostream>
 
 // Simplified rambo version for 2 to N (with N>=2) processes with massless particles
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +83,7 @@ namespace mg5amcCpu
       static bool first = true;
       if( first )
       {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
         if constexpr( M_ACCESS::isOnDevice() ) // avoid
         {
           const int ievt0 = 0;
@@ -166,7 +166,7 @@ namespace mg5amcCpu
     wt = po2log;
     if( nparf != 2 ) wt = ( 2. * nparf - 4. ) * log( energy ) + z[nparf - 1];
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     // issue warnings if weight is too small or too large
     static int iwarn[5] = { 0, 0, 0, 0, 0 };
     if( wt < -180. )
diff --git a/epochX/cudacpp/ee_mumu.mad/src/read_slha.cc b/epochX/cudacpp/ee_mumu.mad/src/read_slha.cc
index 055b19a779..f8e46f2e66 100644
--- a/epochX/cudacpp/ee_mumu.mad/src/read_slha.cc
+++ b/epochX/cudacpp/ee_mumu.mad/src/read_slha.cc
@@ -11,7 +11,11 @@
 
 #include <algorithm>
 #include <cstdlib>
-#include <filesystem>
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
 #include <fstream>
 #include <iostream>
 
@@ -60,7 +64,14 @@ SLHAReader::read_slha_file( std::string file_name, bool verbose )
     {
       std::cout << "WARNING! Card file '" << file_name << "' does not exist:"
                 << " look for the file in directory $" << envpath << "='" << getenv( envpath ) << "'" << std::endl;
+      /*
+#ifdef __HIPCC__
+      const std::string file_name2 = std::experimental::filesystem::path( getenv( envpath ) ) / std::experimental::filesystem::path( file_name ).filename();
+#else
       const std::string file_name2 = std::filesystem::path( getenv( envpath ) ) / std::filesystem::path( file_name ).filename();
+#endif
+      */
+      const std::string file_name2 = std::string( getenv( envpath ) ) + "/" + file_name; // bypass std::filesystem #803
       param_card.open( file_name2.c_str(), std::ifstream::in );
       if( param_card.good() )
       {
diff --git a/epochX/cudacpp/ee_mumu.sa/CODEGEN_cudacpp_ee_mumu_log.txt b/epochX/cudacpp/ee_mumu.sa/CODEGEN_cudacpp_ee_mumu_log.txt
index 636fab0372..2764fbfcfb 100644
--- a/epochX/cudacpp/ee_mumu.sa/CODEGEN_cudacpp_ee_mumu_log.txt
+++ b/epochX/cudacpp/ee_mumu.sa/CODEGEN_cudacpp_ee_mumu_log.txt
@@ -52,7 +52,7 @@ Note that you can still compile and run aMC@NLO with the built-in PDFs
 
 Using default text editor "vi". Set another one in ./input/mg5_configuration.txt
 Using default eps viewer "evince". Set another one in ./input/mg5_configuration.txt
-No valid web browser found. Please set in ./input/mg5_configuration.txt
+Using default web browser "firefox". Set another one in ./input/mg5_configuration.txt
 import /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_cudacpp_ee_mumu.mg
 The import format was not given, so we guess it as command
 set stdout_level DEBUG
@@ -62,7 +62,7 @@ generate e+ e- > mu+ mu-
 No model currently active, so we import the Standard Model
 INFO: load particles 
 INFO: load vertices 
-DEBUG: model prefixing  takes 0.00569605827331543 
+DEBUG: model prefixing  takes 0.00559234619140625 
 INFO: Restrict model sm with file models/sm/restrict_default.dat . 
 DEBUG: Simplifying conditional expressions 
 DEBUG: remove interactions: u s w+ at order: QED=1 
@@ -181,7 +181,7 @@ ALOHA: aloha creates FFV1 routines
 ALOHA: aloha creates FFV2 routines
 ALOHA: aloha creates FFV4 routines
 ALOHA: aloha creates FFV2_4 routines
-ALOHA: aloha creates 4 routines in  0.271 s
+ALOHA: aloha creates 4 routines in  0.266 s
 <class 'aloha.create_aloha.AbstractRoutine'> FFV1
 <class 'aloha.create_aloha.AbstractRoutine'> FFV1
 <class 'aloha.create_aloha.AbstractRoutine'> FFV2
@@ -200,7 +200,7 @@ INFO: Created files Parameters_sm.h and Parameters_sm.cc in directory
 INFO: /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_cudacpp_ee_mumu/src/. and /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_cudacpp_ee_mumu/src/. 
 quit
 
-real	0m0.662s
-user	0m0.604s
-sys	0m0.052s
+real	0m0.654s
+user	0m0.591s
+sys	0m0.054s
 Code generation completed in 1 seconds
diff --git a/epochX/cudacpp/ee_mumu.sa/COPYRIGHT b/epochX/cudacpp/ee_mumu.sa/COPYRIGHT
index a134b5fef9..9036d9260a 100644
--- a/epochX/cudacpp/ee_mumu.sa/COPYRIGHT
+++ b/epochX/cudacpp/ee_mumu.sa/COPYRIGHT
@@ -15,6 +15,7 @@ The full development team currently includes the following authors :
   Stephan Hageboeck (CERN)
   Olivier Mattelaer (Universite Catholique de Louvain, original author)
   Stefan Roiser (CERN, original author)
+  Jorgen Teig (CERN)
   Andrea Valassi (CERN, original author)
   Zenny Wettersten (CERN)
 See https://github.com/madgraph5/madgraph4gpu for more details. For the full
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/Bridge.h b/epochX/cudacpp/ee_mumu.sa/SubProcesses/Bridge.h
index bf8b5e024d..f9ed70dfde 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/Bridge.h
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/Bridge.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Roiser (Nov 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef BRIDGE_H
 #define BRIDGE_H 1
@@ -14,16 +14,23 @@
 #include "MemoryAccessMomenta.h"  // for MemoryAccessMomenta::neppM
 #include "MemoryBuffers.h"        // for HostBufferMomenta, DeviceBufferMomenta etc
 
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
+
+#include <sys/stat.h> // bypass std::filesystem #803
+
 #include <algorithm>
 #include <cassert>
 #include <cmath>
 #include <cstring>
-#include <filesystem>
 #include <iostream>
 #include <memory>
 #include <type_traits>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +90,7 @@ namespace mg5amcCpu
     Bridge& operator=( const Bridge& ) = delete;
     Bridge& operator=( Bridge&& ) = delete;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     /**
      * Set the gpublocks and gputhreads for the gpusequence - throws if evnt != gpublocks*gputhreads
      * (this is needed for BridgeKernel tests rather than for actual production use in Fortran)
@@ -150,7 +157,7 @@ namespace mg5amcCpu
     unsigned int m_nevt; // number of events
     int m_nGoodHel;      // the number of good helicities (-1 initially when they have not yet been calculated)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     int m_gputhreads; // number of gpu threads (default set from number of events, can be modified)
     int m_gpublocks;  // number of gpu blocks (default set from number of events, can be modified)
     DeviceBuffer<FORTRANFPTYPE, sizePerEventMomenta> m_devMomentaF;
@@ -187,12 +194,12 @@ namespace mg5amcCpu
   // Forward declare transposition methods
   //
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
   template<typename Tin, typename Tout>
   __global__ void dev_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt );
 
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
 
   template<typename Tin, typename Tout>
   void hst_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt );
@@ -209,7 +216,7 @@ namespace mg5amcCpu
   Bridge<FORTRANFPTYPE>::Bridge( unsigned int nevtF, unsigned int nparF, unsigned int np4F )
     : m_nevt( nevtF )
     , m_nGoodHel( -1 )
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     , m_gputhreads( 256 )                  // default number of gpu threads
     , m_gpublocks( m_nevt / m_gputhreads ) // this ensures m_nevt <= m_gpublocks*m_gputhreads
     , m_devMomentaF( m_nevt )
@@ -233,7 +240,7 @@ namespace mg5amcCpu
   {
     if( nparF != CPPProcess::npar ) throw std::runtime_error( "Bridge constructor: npar mismatch" );
     if( np4F != CPPProcess::np4 ) throw std::runtime_error( "Bridge constructor: np4 mismatch" );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( ( m_nevt < s_gputhreadsmin ) || ( m_nevt % s_gputhreadsmin != 0 ) )
       throw std::runtime_error( "Bridge constructor: nevt should be a multiple of " + std::to_string( s_gputhreadsmin ) );
     while( m_nevt != m_gpublocks * m_gputhreads )
@@ -249,20 +256,28 @@ namespace mg5amcCpu
 #else
     std::cout << "WARNING! Instantiate host Bridge (nevt=" << m_nevt << ")" << std::endl;
     m_pmek.reset( new MatrixElementKernelHost( m_hstMomentaC, m_hstGs, m_hstRndHel, m_hstRndCol, m_hstMEs, m_hstSelHel, m_hstSelCol, m_nevt ) );
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
     // Create a process object, read param card and set parameters
     // FIXME: the process instance can happily go out of scope because it is only needed to read parameters?
     // FIXME: the CPPProcess should really be a singleton? what if fbridgecreate is called from several Fortran threads?
     CPPProcess process( /*verbose=*/false );
     std::string paramCard = "../../Cards/param_card.dat";
-    if( !std::filesystem::exists( paramCard ) )
-    {
-      paramCard = "../" + paramCard;
-    }
+    /*
+#ifdef __HIPCC__
+    if( !std::experimental::filesystem::exists( paramCard ) ) paramCard = "../" + paramCard;
+#else
+    if( !std::filesystem::exists( paramCard ) ) paramCard = "../" + paramCard;
+#endif
+    */
+    //struct stat dummybuffer; // bypass std::filesystem #803
+    //if( !( stat( paramCard.c_str(), &dummyBuffer ) == 0 ) ) paramCard = "../" + paramCard; //
+    auto fileExists = []( std::string& fileName )
+    { struct stat buffer; return stat( fileName.c_str(), &buffer ) == 0; };
+    if( !fileExists( paramCard ) ) paramCard = "../" + paramCard; // bypass std::filesystem #803
     process.initProc( paramCard );
   }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::set_gpugrid( const int gpublocks, const int gputhreads )
   {
@@ -276,7 +291,7 @@ namespace mg5amcCpu
   }
 #endif
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::gpu_sequence( const FORTRANFPTYPE* momenta,
                                             const FORTRANFPTYPE* gs,
@@ -291,14 +306,14 @@ namespace mg5amcCpu
     constexpr int neppM = MemoryAccessMomenta::neppM;
     if constexpr( neppM == 1 && std::is_same_v<FORTRANFPTYPE, fptype> )
     {
-      checkCuda( cudaMemcpy( m_devMomentaC.data(), momenta, m_devMomentaC.bytes(), cudaMemcpyHostToDevice ) );
+      gpuMemcpy( m_devMomentaC.data(), momenta, m_devMomentaC.bytes(), gpuMemcpyHostToDevice );
     }
     else
     {
-      checkCuda( cudaMemcpy( m_devMomentaF.data(), momenta, m_devMomentaF.bytes(), cudaMemcpyHostToDevice ) );
+      gpuMemcpy( m_devMomentaF.data(), momenta, m_devMomentaF.bytes(), gpuMemcpyHostToDevice );
       const int thrPerEvt = CPPProcess::npar * CPPProcess::np4; // AV: transpose alg does 1 element per thread (NOT 1 event per thread)
       //const int thrPerEvt = 1; // AV: try new alg with 1 event per thread... this seems slower
-      dev_transposeMomentaF2C<<<m_gpublocks * thrPerEvt, m_gputhreads>>>( m_devMomentaF.data(), m_devMomentaC.data(), m_nevt );
+      gpuLaunchKernel( dev_transposeMomentaF2C, m_gpublocks * thrPerEvt, m_gputhreads, m_devMomentaF.data(), m_devMomentaC.data(), m_nevt );
     }
     if constexpr( std::is_same_v<FORTRANFPTYPE, fptype> )
     {
@@ -341,7 +356,7 @@ namespace mg5amcCpu
   }
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::cpu_sequence( const FORTRANFPTYPE* momenta,
                                             const FORTRANFPTYPE* gs,
@@ -396,7 +411,7 @@ namespace mg5amcCpu
   // - C++ array: momenta[npagM][npar][np4][neppM] with nevt=npagM*neppM (AOSOA)
   //
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename Tin, typename Tout>
   __global__ void dev_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt )
   {
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/BridgeKernels.cc b/epochX/cudacpp/ee_mumu.sa/SubProcesses/BridgeKernels.cc
index d58066c9c1..eaf4037a24 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/BridgeKernels.cc
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/BridgeKernels.cc
@@ -1,17 +1,18 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "BridgeKernels.h"
 
+#include "GpuAbstraction.h"
 #include "MemoryAccessMomenta.h"
 
 #include <sstream>
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -45,7 +46,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu
 {
 
@@ -96,7 +97,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/BridgeKernels.h b/epochX/cudacpp/ee_mumu.sa/SubProcesses/BridgeKernels.h
index 15eb4bff4d..3efef8ce97 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/BridgeKernels.h
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/BridgeKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef BRIDGEKERNELS_H
 #define BRIDGEKERNELS_H 1
@@ -12,7 +12,7 @@
 #include "MatrixElementKernels.h"
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -49,7 +49,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A Bridge wrapper class encapsulating matrix element calculations on a CPU host
   class BridgeKernelHost final : public BridgeKernelBase
   {
@@ -89,7 +89,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A Bridge wrapper class encapsulating matrix element calculations on a GPU device
   class BridgeKernelDevice : public BridgeKernelBase
   {
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/CommonRandomNumberKernel.cc b/epochX/cudacpp/ee_mumu.sa/SubProcesses/CommonRandomNumberKernel.cc
index 985b39f576..010bc4cbd0 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/CommonRandomNumberKernel.cc
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/CommonRandomNumberKernel.cc
@@ -1,15 +1,16 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "CommonRandomNumbers.h"
+#include "GpuAbstraction.h"
 #include "MemoryBuffers.h"
 #include "RandomNumberKernels.h"
 
 #include <cassert>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/CrossSectionKernels.cc b/epochX/cudacpp/ee_mumu.sa/SubProcesses/CrossSectionKernels.cc
index 0b355a3c8d..c15b39844d 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/CrossSectionKernels.cc
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/CrossSectionKernels.cc
@@ -1,10 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "CrossSectionKernels.h"
 
+#include "GpuAbstraction.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessWeights.h"
 #include "MemoryBuffers.h"
@@ -77,7 +78,7 @@ debug_me_is_abnormal( const fptype& me, size_t ievtALL )
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -185,7 +186,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/CrossSectionKernels.h b/epochX/cudacpp/ee_mumu.sa/SubProcesses/CrossSectionKernels.h
index 7933ca4bbf..4d9659e04e 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/CrossSectionKernels.h
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/CrossSectionKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef CROSSSECTIONKERNELS_H
 #define CROSSSECTIONKERNELS_H 1
@@ -13,7 +13,7 @@
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -96,7 +96,7 @@ namespace mg5amcCpu
   //--------------------------------------------------------------------------
 
   /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating the calculation of event statistics on a GPU device
   class CrossSectionKernelDevice : public CrossSectionKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/CurandRandomNumberKernel.cc b/epochX/cudacpp/ee_mumu.sa/SubProcesses/CurandRandomNumberKernel.cc
index eb56333b03..08a16f6f2c 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/CurandRandomNumberKernel.cc
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/CurandRandomNumberKernel.cc
@@ -1,9 +1,9 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "MemoryBuffers.h"
 #include "RandomNumberKernels.h"
 
@@ -22,7 +22,7 @@ inline void assertCurand( curandStatus_t code, const char *file, int line, bool
 }
 #endif /* clang-format on */
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -36,7 +36,7 @@ namespace mg5amcCpu
   {
     if( m_isOnDevice )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if( !m_rnarray.isOnDevice() )
         throw std::runtime_error( "CurandRandomNumberKernel on device with a host random number array" );
 #else
@@ -114,7 +114,7 @@ namespace mg5amcCpu
     /*
     printf( "\nCurandRandomNumberKernel::generateRnarray size = %d\n", (int)m_rnarray.size() );
     fptype* data = m_rnarray.data();
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( m_rnarray.isOnDevice() )
     {
       data = new fptype[m_rnarray.size()]();
@@ -123,7 +123,7 @@ namespace mg5amcCpu
 #endif
     for( int i = 0; i < ( (int)m_rnarray.size() / 4 ); i++ )
       printf( "[%4d] %f %f %f %f\n", i * 4, data[i * 4], data[i * 4 + 2], data[i * 4 + 2], data[i * 4 + 3] );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( m_rnarray.isOnDevice() ) delete[] data;
 #endif
     */
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/EventStatistics.h b/epochX/cudacpp/ee_mumu.sa/SubProcesses/EventStatistics.h
index 48b51e0a49..b425a5bade 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/EventStatistics.h
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/EventStatistics.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef EventStatistics_H
 #define EventStatistics_H 1
@@ -16,7 +16,7 @@
 #include <limits>
 #include <string>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/GpuAbstraction.h b/epochX/cudacpp/ee_mumu.sa/SubProcesses/GpuAbstraction.h
new file mode 100644
index 0000000000..9c467b1e04
--- /dev/null
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/GpuAbstraction.h
@@ -0,0 +1,69 @@
+// Copyright (C) 2020-2023 CERN and UCLouvain.
+// Licensed under the GNU Lesser General Public License (version 3 or later).
+// Created by: J. Teig (Jul 2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+
+#ifndef MG5AMC_GPUABSTRACTION_H
+#define MG5AMC_GPUABSTRACTION_H 1
+
+#include <cassert>
+
+//--------------------------------------------------------------------------
+
+#ifdef __CUDACC__
+
+#define gpuError_t cudaError_t
+#define gpuPeekAtLastError cudaPeekAtLastError
+#define gpuGetErrorString cudaGetErrorString
+#define gpuSuccess cudaSuccess
+
+#define gpuMallocHost( ptr, size ) checkGpu( cudaMallocHost( ptr, size ) )
+#define gpuMalloc( ptr, size ) checkGpu( cudaMalloc( ptr, size ) )
+
+#define gpuMemcpy( dstData, srcData, srcBytes, func ) checkGpu( cudaMemcpy( dstData, srcData, srcBytes, func ) )
+#define gpuMemcpyHostToDevice cudaMemcpyHostToDevice
+#define gpuMemcpyDeviceToHost cudaMemcpyDeviceToHost
+#define gpuMemcpyToSymbol( type1, type2, size ) checkGpu( cudaMemcpyToSymbol( type1, type2, size ) )
+
+#define gpuFree( ptr ) checkGpu( cudaFree( ptr ) )
+#define gpuFreeHost( ptr ) checkGpu( cudaFreeHost( ptr ) )
+
+#define gpuSetDevice cudaSetDevice
+#define gpuDeviceSynchronize cudaDeviceSynchronize
+#define gpuDeviceReset cudaDeviceReset
+
+#define gpuLaunchKernel( kernel, blocks, threads, ... ) kernel<<<blocks, threads>>>( __VA_ARGS__ )
+#define gpuLaunchKernelSharedMem( kernel, blocks, threads, sharedMem, ... ) kernel<<<blocks, threads, sharedMem>>>( __VA_ARGS__ )
+
+//--------------------------------------------------------------------------
+
+#elif defined __HIPCC__
+
+#define gpuError_t hipError_t
+#define gpuPeekAtLastError hipPeekAtLastError
+#define gpuGetErrorString hipGetErrorString
+#define gpuSuccess hipSuccess
+
+#define gpuMallocHost( ptr, size ) checkGpu( hipHostMalloc( ptr, size ) ) // HostMalloc better
+#define gpuMalloc( ptr, size ) checkGpu( hipMalloc( ptr, size ) )
+
+#define gpuMemcpy( dstData, srcData, srcBytes, func ) checkGpu( hipMemcpy( dstData, srcData, srcBytes, func ) )
+#define gpuMemcpyHostToDevice hipMemcpyHostToDevice
+#define gpuMemcpyDeviceToHost hipMemcpyDeviceToHost
+#define gpuMemcpyToSymbol( type1, type2, size ) checkGpu( hipMemcpyToSymbol( type1, type2, size ) )
+
+#define gpuFree( ptr ) checkGpu( hipFree( ptr ) )
+#define gpuFreeHost( ptr ) checkGpu( hipHostFree( ptr ) )
+
+#define gpuSetDevice hipSetDevice
+#define gpuDeviceSynchronize hipDeviceSynchronize
+#define gpuDeviceReset hipDeviceReset
+
+#define gpuLaunchKernel( kernel, blocks, threads, ... ) kernel<<<blocks, threads>>>( __VA_ARGS__ )
+#define gpuLaunchKernelSharedMem( kernel, blocks, threads, sharedMem, ... ) kernel<<<blocks, threads, sharedMem>>>( __VA_ARGS__ )
+
+//--------------------------------------------------------------------------
+
+#endif
+
+#endif // MG5AMC_GPUABSTRACTION_H
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/CudaRuntime.h b/epochX/cudacpp/ee_mumu.sa/SubProcesses/GpuRuntime.h
similarity index 62%
rename from epochX/cudacpp/gg_tt.mad/SubProcesses/CudaRuntime.h
rename to epochX/cudacpp/ee_mumu.sa/SubProcesses/GpuRuntime.h
index 64ce52f4b3..93579ef08b 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/CudaRuntime.h
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/GpuRuntime.h
@@ -1,49 +1,50 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
-// Created by: S. Roiser (Jul 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Created by: J. Teig (Jun 2023, based on earlier work by S. Roiser) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
-#ifndef MG5AMC_CUDARUNTIME_H
-#define MG5AMC_CUDARUNTIME_H 1
+#ifndef MG5AMC_GPURUNTIME_H
+#define MG5AMC_GPURUNTIME_H 1
 
 // MG5AMC on GPU uses the CUDA runtime API, not the lower level CUDA driver API
 // See https://docs.nvidia.com/cuda/cuda-runtime-api/driver-vs-runtime-api.html#driver-vs-runtime-api
 
-#include <cassert>
+#include "GpuAbstraction.h"
+
 #include <iostream>
 
 //--------------------------------------------------------------------------
 
 // See https://stackoverflow.com/a/14038590
-#ifdef __CUDACC__ /* clang-format off */
-#define checkCuda( code ) { assertCuda( code, __FILE__, __LINE__ ); }
-inline void assertCuda( cudaError_t code, const char* file, int line, bool abort = true )
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
+#define checkGpu( code ) { assertGpu( code, __FILE__, __LINE__ ); }
+inline void assertGpu( gpuError_t code, const char* file, int line, bool abort = true )
 {
-  if( code != cudaSuccess )
+  if( code != gpuSuccess )
   {
-    printf( "ERROR! assertCuda: '%s' (%d) in %s:%d\n", cudaGetErrorString( code ), code, file, line );
-    if( abort ) assert( code == cudaSuccess );
+    printf( "ERROR! assertGpu: '%s' (%d) in %s:%d\n", gpuGetErrorString( code ), code, file, line );
+    if( abort ) assert( code == gpuSuccess );
   }
 }
 #endif /* clang-format on */
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
+  // Instantiate a GpuRuntime at the beginnining of the application's main to
+  // invoke gpuSetDevice(0) in the constructor and book a gpuDeviceReset() call in the destructor
   // *** FIXME! This will all need to be designed differently when going to multi-GPU nodes! ***
-  struct CudaRuntime final
+  struct GpuRuntime final
   {
-    CudaRuntime( const bool debug = true )
+    GpuRuntime( const bool debug = true )
       : m_debug( debug ) { setUp( m_debug ); }
-    ~CudaRuntime() { tearDown( m_debug ); }
-    CudaRuntime( const CudaRuntime& ) = delete;
-    CudaRuntime( CudaRuntime&& ) = delete;
-    CudaRuntime& operator=( const CudaRuntime& ) = delete;
-    CudaRuntime& operator=( CudaRuntime&& ) = delete;
+    ~GpuRuntime() { tearDown( m_debug ); }
+    GpuRuntime( const GpuRuntime& ) = delete;
+    GpuRuntime( GpuRuntime&& ) = delete;
+    GpuRuntime& operator=( const GpuRuntime& ) = delete;
+    GpuRuntime& operator=( GpuRuntime&& ) = delete;
     bool m_debug;
 
     // Set up CUDA application
@@ -62,8 +63,8 @@ namespace mg5amcGpu
       */
       // Replace cudaFree(0) by cudaSetDevice(0), even if it is not really needed either
       // (but see https://developer.nvidia.com/blog/cuda-pro-tip-always-set-current-device-avoid-multithreading-bugs)
-      if( debug ) std::cout << "__CudaRuntime: calling cudaSetDevice(0)" << std::endl;
-      checkCuda( cudaSetDevice( 0 ) ); // SLOW!
+      if( debug ) std::cout << "__GpuRuntime: calling GpuSetDevice(0)" << std::endl;
+      checkGpu( gpuSetDevice( 0 ) ); // SLOW!
     }
 
     // Tear down CUDA application (call cudaDeviceReset)
@@ -72,14 +73,13 @@ namespace mg5amcGpu
     // See https://docs.nvidia.com/cuda/cuda-memcheck/index.html#leak-checking
     static void tearDown( const bool debug = true )
     {
-      if( debug ) std::cout << "__CudaRuntime: calling cudaDeviceReset()" << std::endl;
-      checkCuda( cudaDeviceReset() );
+      if( debug ) std::cout << "__GpuRuntime: calling GpuDeviceReset()" << std::endl;
+      checkGpu( gpuDeviceReset() );
     }
   };
-
 }
 #endif
 
 //--------------------------------------------------------------------------
 
-#endif // MG5AMC_CUDARUNTIME_H
+#endif // MG5AMC_GPURUNTIME_H
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/MadgraphTest.h b/epochX/cudacpp/ee_mumu.sa/SubProcesses/MadgraphTest.h
index ef40624c88..6054185300 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/MadgraphTest.h
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/MadgraphTest.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Hageboeck (Dec 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MADGRAPHTEST_H_
 #define MADGRAPHTEST_H_ 1
@@ -14,7 +14,11 @@
 
 #include <array>
 #include <cmath>
-#include <filesystem>
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
 #include <fstream>
 #include <iomanip>
 #include <memory>
@@ -22,7 +26,7 @@
 #include <string>
 #include <vector>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using mg5amcGpu::CPPProcess;
 #else
 using mg5amcCpu::CPPProcess;
@@ -201,7 +205,7 @@ class MadgraphTest : public testing::TestWithParam<TestDriverBase*>
 
 // Since we link both the CPU-only and GPU tests into the same executable, we prevent
 // a multiply defined symbol by only compiling this in the non-CUDA phase:
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
 /// Compare momenta and matrix elements.
 /// This uses an implementation of TestDriverBase to run a madgraph workflow,
@@ -219,7 +223,14 @@ TEST_P( MadgraphTest, CompareMomentaAndME )
   const char* dumpEventsC = getenv( "CUDACPP_RUNTEST_DUMPEVENTS" );
   const bool dumpEvents = ( dumpEventsC != 0 ) && ( std::string( dumpEventsC ) != "" );
   const std::string refFileName = testDriver->getRefFileName();
+  /*
+#ifdef __HIPCC__
+  const std::string dumpFileName = std::experimental::filesystem::path( refFileName ).filename();
+#else
   const std::string dumpFileName = std::filesystem::path( refFileName ).filename();
+#endif
+  */
+  const std::string dumpFileName = refFileName; // bypass std::filesystem #803
   std::ofstream dumpFile;
   if( dumpEvents )
   {
@@ -307,6 +318,6 @@ TEST_P( MadgraphTest, CompareMomentaAndME )
   }
 }
 
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
 
 #endif /* MADGRAPHTEST_H_ */
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/MatrixElementKernels.cc b/epochX/cudacpp/ee_mumu.sa/SubProcesses/MatrixElementKernels.cc
index 74b5239ebf..81699dfea9 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/MatrixElementKernels.cc
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/MatrixElementKernels.cc
@@ -1,12 +1,12 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "MatrixElementKernels.h"
 
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h" // Includes the abstraction for Nvidia/AMD compilation
 #include "MemoryAccessMomenta.h"
 #include "MemoryBuffers.h"
 
@@ -14,7 +14,7 @@
 
 //============================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu
 {
 
@@ -150,7 +150,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
@@ -209,13 +209,13 @@ namespace mg5amcGpu
     PinnedHostBufferHelicityMask hstIsGoodHel( ncomb );
     DeviceBufferHelicityMask devIsGoodHel( ncomb );
     // ... 0d1. Compute good helicity mask on the device
-    computeDependentCouplings<<<m_gpublocks, m_gputhreads>>>( m_gs.data(), m_couplings.data() );
+    gpuLaunchKernel( computeDependentCouplings, m_gpublocks, m_gputhreads, m_gs.data(), m_couplings.data() );
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-    sigmaKin_getGoodHel<<<m_gpublocks, m_gputhreads>>>( m_momenta.data(), m_couplings.data(), m_matrixElements.data(), m_numerators.data(), m_denominators.data(), devIsGoodHel.data() );
+    gpuLaunchKernel( sigmaKin_getGoodHel, m_gpublocks, m_gputhreads, m_momenta.data(), m_couplings.data(), m_matrixElements.data(), m_numerators.data(), m_denominators.data(), devIsGoodHel.data() );
 #else
-    sigmaKin_getGoodHel<<<m_gpublocks, m_gputhreads>>>( m_momenta.data(), m_couplings.data(), m_matrixElements.data(), devIsGoodHel.data() );
+    gpuLaunchKernel( sigmaKin_getGoodHel, m_gpublocks, m_gputhreads, m_momenta.data(), m_couplings.data(), m_matrixElements.data(), devIsGoodHel.data() );
 #endif
-    checkCuda( cudaPeekAtLastError() );
+    checkGpu( gpuPeekAtLastError() );
     // ... 0d2. Copy back good helicity mask to the host
     copyHostFromDevice( hstIsGoodHel, devIsGoodHel );
     // ... 0d3. Copy back good helicity list to constant memory on the device
@@ -226,19 +226,19 @@ namespace mg5amcGpu
 
   void MatrixElementKernelDevice::computeMatrixElements( const unsigned int channelId )
   {
-    computeDependentCouplings<<<m_gpublocks, m_gputhreads>>>( m_gs.data(), m_couplings.data() );
+    gpuLaunchKernel( computeDependentCouplings, m_gpublocks, m_gputhreads, m_gs.data(), m_couplings.data() );
 #ifndef MGONGPU_NSIGHT_DEBUG
     constexpr unsigned int sharedMemSize = 0;
 #else
     constexpr unsigned int sharedMemSize = ntpbMAX * sizeof( float );
 #endif
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-    sigmaKin<<<m_gpublocks, m_gputhreads, sharedMemSize>>>( m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), channelId, m_numerators.data(), m_denominators.data(), m_selhel.data(), m_selcol.data() );
+    gpuLaunchKernelSharedMem( sigmaKin, m_gpublocks, m_gputhreads, sharedMemSize, m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), channelId, m_numerators.data(), m_denominators.data(), m_selhel.data(), m_selcol.data() );
 #else
-    sigmaKin<<<m_gpublocks, m_gputhreads, sharedMemSize>>>( m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), m_selhel.data(), m_selcol.data() );
+    gpuLaunchKernelSharedMem( sigmaKin, m_gpublocks, m_gputhreads, sharedMemSize, m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), m_selhel.data(), m_selcol.data() );
 #endif
-    checkCuda( cudaPeekAtLastError() );
-    checkCuda( cudaDeviceSynchronize() );
+    checkGpu( gpuPeekAtLastError() );
+    checkGpu( gpuDeviceSynchronize() );
   }
 
   //--------------------------------------------------------------------------
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/MatrixElementKernels.h b/epochX/cudacpp/ee_mumu.sa/SubProcesses/MatrixElementKernels.h
index 23e84757a2..72bd8f195b 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/MatrixElementKernels.h
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/MatrixElementKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MATRIXELEMENTKERNELS_H
 #define MATRIXELEMENTKERNELS_H 1
@@ -10,7 +10,7 @@
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -81,7 +81,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating matrix element calculations on a CPU host
   class MatrixElementKernelHost final : public MatrixElementKernelBase, public NumberOfEvents
   {
@@ -130,7 +130,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating matrix element calculations on a GPU device
   class MatrixElementKernelDevice : public MatrixElementKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessAmplitudes.h b/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessAmplitudes.h
index 573b3bbbc9..ffb76e93de 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessAmplitudes.h
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessAmplitudes.h
@@ -15,7 +15,7 @@
 #define MGONGPU_TRIVIAL_AMPLITUDES 1
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessCouplings.h b/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessCouplings.h
index 35a3af42e0..3afdf3e554 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessCouplings.h
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessCouplings.h
@@ -15,7 +15,7 @@
 #include "MemoryBuffers.h"       // for HostBufferCouplings::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessCouplingsFixed.h b/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessCouplingsFixed.h
index dc0d93afff..ffcdf4dbef 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessCouplingsFixed.h
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessCouplingsFixed.h
@@ -14,7 +14,7 @@
 //#include "MemoryAccessHelpers.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessDenominators.h b/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessDenominators.h
index 3bce635718..66f2d32a6b 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessDenominators.h
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessDenominators.h
@@ -10,7 +10,7 @@
 #include "MemoryAccessGs.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessGs.h b/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessGs.h
index 31311aa375..4c726b30f3 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessGs.h
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessGs.h
@@ -13,7 +13,7 @@
 #include "MemoryBuffers.h" // for HostBufferMatrixElements::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessHelpers.h b/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessHelpers.h
index c82a6c7635..db73e4e064 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessHelpers.h
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessHelpers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessHelpers_H
 #define MemoryAccessHelpers_H 1
@@ -105,7 +105,7 @@ class KernelAccessHelper : public MemoryAccessHelper<T>
     }
     else
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
       //printf( "kernelAccessRecord: ievt=%d threadId=%d\n", ievt, threadIdx.x );
       return T::ieventAccessRecord( buffer, ievt ); // NB fptype and fptype_sv coincide for CUDA
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessMatrixElements.h b/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessMatrixElements.h
index f32e6fea5b..3741011971 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessMatrixElements.h
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessMatrixElements.h
@@ -13,7 +13,7 @@
 #include "MemoryBuffers.h" // for HostBufferMatrixElements::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessMomenta.h b/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessMomenta.h
index 29266de32c..3be229d392 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessMomenta.h
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessMomenta.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessMomenta_H
 #define MemoryAccessMomenta_H 1
@@ -13,7 +13,7 @@
 #include "MemoryAccessVectors.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -30,7 +30,7 @@ namespace mg5amcCpu
 
     // Number of Events Per Page in the momenta AOSOA memory buffer layout
     // (these are all best kept as a compile-time constants: see issue #23)
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
     // -----------------------------------------------------------------------------------------------
     // --- GPUs: neppM is best set to a power of 2 times the number of fptype's in a 32-byte cacheline
     // --- This is relevant to ensure coalesced access to momenta in global memory
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessNumerators.h b/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessNumerators.h
index b152183b28..18991f4fa6 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessNumerators.h
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessNumerators.h
@@ -10,7 +10,7 @@
 #include "MemoryAccessGs.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessRandomNumbers.h b/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessRandomNumbers.h
index e2988d39f3..40cb089135 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessRandomNumbers.h
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessRandomNumbers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessRandomNumbers_H
 #define MemoryAccessRandomNumbers_H 1
@@ -11,7 +11,7 @@
 #include "CPPProcess.h"
 #include "MemoryAccessHelpers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using mg5amcGpu::CPPProcess;
 #else
 using mg5amcCpu::CPPProcess;
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessVectors.h b/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessVectors.h
index e9b197368e..08faccff0f 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessVectors.h
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessVectors.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessVectors_H
 #define MemoryAccessVectors_H 1
@@ -10,7 +10,7 @@
 
 #include "mgOnGpuVectors.h"
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu // this is only needed for CPU SIMD vectorization
 {
 
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessWavefunctions.h b/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessWavefunctions.h
index 5428aaf933..33bef4559e 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessWavefunctions.h
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryAccessWavefunctions.h
@@ -15,7 +15,7 @@
 #define MGONGPU_TRIVIAL_WAVEFUNCTIONS 1
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryBuffers.h b/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryBuffers.h
index 3093e6ed18..7756a71621 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryBuffers.h
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/MemoryBuffers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021, based on earlier work by S. Hageboeck) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryBuffers_H
 #define MemoryBuffers_H 1
@@ -11,12 +11,12 @@
 #include "mgOnGpuCxtypes.h"
 
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "Parameters_sm.h"
 
 #include <sstream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -87,7 +87,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   constexpr bool HostBufferALIGNED = false;   // ismisaligned=false
   constexpr bool HostBufferMISALIGNED = true; // ismisaligned=true
 
@@ -119,7 +119,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA pinned host buffer
   template<typename T>
   class PinnedHostBufferBase : public BufferBase<T>
@@ -128,18 +128,18 @@ namespace mg5amcCpu
     PinnedHostBufferBase( const size_t size )
       : BufferBase<T>( size, false )
     {
-      checkCuda( cudaMallocHost( &( this->m_data ), this->bytes() ) );
+      gpuMallocHost( &( this->m_data ), this->bytes() );
     }
     virtual ~PinnedHostBufferBase()
     {
-      checkCuda( cudaFreeHost( this->m_data ) );
+      gpuFreeHost( this->m_data );
     }
   };
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA device buffer
   template<typename T>
   class DeviceBufferBase : public BufferBase<T>
@@ -148,18 +148,18 @@ namespace mg5amcCpu
     DeviceBufferBase( const size_t size )
       : BufferBase<T>( size, true )
     {
-      checkCuda( cudaMalloc( &( this->m_data ), this->bytes() ) );
+      gpuMalloc( &( this->m_data ), this->bytes() );
     }
     virtual ~DeviceBufferBase()
     {
-      checkCuda( cudaFree( this->m_data ) );
+      gpuFree( this->m_data );
     }
   };
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for a given number of events
   template<typename T, size_t sizePerEvent, bool ismisaligned>
   class HostBuffer : public HostBufferBase<T, ismisaligned>, virtual private NumberOfEvents
@@ -175,7 +175,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA pinned host buffer for a given number of events
   template<typename T, size_t sizePerEvent>
   class PinnedHostBuffer : public PinnedHostBufferBase<T>, virtual private NumberOfEvents
@@ -191,7 +191,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA device buffer for a given number of events
   template<typename T, size_t sizePerEvent>
   class DeviceBuffer : public DeviceBufferBase<T>, virtual private NumberOfEvents
@@ -213,7 +213,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for momenta random numbers
   constexpr size_t sizePerEventRndNumMomenta = MemoryBuffers::np4 * MemoryBuffers::nparf;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for momenta random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumMomenta, HostBufferALIGNED> HostBufferRndNumMomenta;
 #else
@@ -232,7 +232,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer with ONE fptype per event
   constexpr size_t sizePerEventOneFp = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer with ONE fptype per event
   typedef HostBuffer<fptype, sizePerEventOneFp, HostBufferALIGNED> HostBufferOneFp;
 #else
@@ -257,7 +257,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for Gs
   constexpr size_t sizePerEventGs = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventGs, HostBufferALIGNED> HostBufferGs;
 #else
@@ -276,7 +276,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for numerators
   constexpr size_t sizePerEventNumerators = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventNumerators, HostBufferALIGNED> HostBufferNumerators;
 #else
@@ -296,7 +296,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for denominators
   constexpr size_t sizePerEventDenominators = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventDenominators, HostBufferALIGNED> HostBufferDenominators;
 #else
@@ -315,7 +315,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for random numbers
   constexpr size_t sizePerEventCouplings = MemoryBuffers::ndcoup * MemoryBuffers::nx2;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventCouplings, HostBufferALIGNED> HostBufferCouplings;
 #else
@@ -333,7 +333,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for momenta
   constexpr size_t sizePerEventMomenta = MemoryBuffers::np4 * MemoryBuffers::npar;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for momenta
   typedef HostBuffer<fptype, sizePerEventMomenta, HostBufferALIGNED> HostBufferMomenta;
   //typedef HostBuffer<fptype, sizePerEventMomenta, HostBufferMISALIGNED> HostBufferMomenta; // TEST MISALIGNMENT!
@@ -352,7 +352,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for sampling weights
   constexpr size_t sizePerEventWeights = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for sampling weights
   typedef HostBuffer<fptype, sizePerEventWeights, HostBufferALIGNED> HostBufferWeights;
 #else
@@ -370,7 +370,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for matrix elements
   constexpr size_t sizePerEventMatrixElements = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for matrix elements
   typedef HostBuffer<fptype, sizePerEventMatrixElements, HostBufferALIGNED> HostBufferMatrixElements;
 #else
@@ -385,7 +385,7 @@ namespace mg5amcCpu
   // A base class encapsulating a memory buffer for the helicity mask
   typedef BufferBase<bool> BufferHelicityMask;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for the helicity mask
   typedef HostBufferBase<bool, HostBufferALIGNED> HostBufferHelicityMask;
 #else
@@ -403,7 +403,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for wavefunctions
   constexpr size_t sizePerEventWavefunctions = MemoryBuffers::nw6 * MemoryBuffers::nx2;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for wavefunctions
   typedef HostBuffer<fptype, sizePerEventWavefunctions, HostBufferALIGNED> HostBufferWavefunctions;
 #else
@@ -421,7 +421,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for helicity random numbers
   constexpr size_t sizePerEventRndNumHelicity = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for helicity random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumHelicity, HostBufferALIGNED> HostBufferRndNumHelicity;
 #else
@@ -439,7 +439,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for color random numbers
   constexpr size_t sizePerEventRndNumColor = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for color random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumColor, HostBufferALIGNED> HostBufferRndNumColor;
 #else
@@ -457,7 +457,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for helicity selection
   constexpr size_t sizePerEventSelectedHelicity = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for helicity selection
   typedef HostBuffer<int, sizePerEventSelectedHelicity, HostBufferALIGNED> HostBufferSelectedHelicity;
 #else
@@ -475,7 +475,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for color selection
   constexpr size_t sizePerEventSelectedColor = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for color selection
   typedef HostBuffer<int, sizePerEventSelectedColor, HostBufferALIGNED> HostBufferSelectedColor;
 #else
@@ -487,7 +487,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<class Tdst, class Tsrc>
   void copyDeviceFromHost( Tdst& dst, const Tsrc& src ) // keep the same order of arguments as in memcpy
   {
@@ -504,13 +504,13 @@ namespace mg5amcCpu
       throw std::runtime_error( sstr.str() );
     }
     // NB (PR #45): cudaMemcpy involves an intermediate memcpy to pinned memory if host array is a not a pinned host array
-    checkCuda( cudaMemcpy( dst.data(), src.data(), src.bytes(), cudaMemcpyHostToDevice ) );
+    gpuMemcpy( dst.data(), src.data(), src.bytes(), gpuMemcpyHostToDevice );
   }
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<class Tdst, class Tsrc>
   void copyHostFromDevice( Tdst& dst, const Tsrc& src ) // keep the same order of arguments as in memcpy
   {
@@ -527,7 +527,7 @@ namespace mg5amcCpu
       throw std::runtime_error( sstr.str() );
     }
     // NB (PR #45): cudaMemcpy involves an intermediate memcpy to pinned memory if host array is a not a pinned host array
-    checkCuda( cudaMemcpy( dst.data(), src.data(), src.bytes(), cudaMemcpyDeviceToHost ) );
+    gpuMemcpy( dst.data(), src.data(), src.bytes(), gpuMemcpyDeviceToHost );
   }
 #endif
 
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/P1_Sigma_sm_epem_mupmum/CPPProcess.cc b/epochX/cudacpp/ee_mumu.sa/SubProcesses/P1_Sigma_sm_epem_mupmum/CPPProcess.cc
index 87bcecccd9..b87b14d41f 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/P1_Sigma_sm_epem_mupmum/CPPProcess.cc
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/P1_Sigma_sm_epem_mupmum/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_sm.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -46,7 +45,7 @@
 // Class member functions for calculating the matrix elements for
 // Process: e+ e- > mu+ mu- WEIGHTED<=4 @1
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -80,7 +79,7 @@ namespace mg5amcCpu
   __device__ const fptype cIPD[2] = { (fptype)Parameters_sm::mdl_MZ, (fptype)Parameters_sm::mdl_WZ };
   __device__ const fptype cIPC[6] = { (fptype)Parameters_sm::GC_3.real(), (fptype)Parameters_sm::GC_3.imag(), (fptype)Parameters_sm::GC_50.real(), (fptype)Parameters_sm::GC_50.imag(), (fptype)Parameters_sm::GC_59.real(), (fptype)Parameters_sm::GC_59.imag() };
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ fptype cIPD[2];
   __device__ __constant__ fptype cIPC[6];
 #else
@@ -90,7 +89,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -118,13 +117,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -151,7 +150,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -187,7 +186,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -200,8 +199,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -284,7 +285,7 @@ namespace mg5amcCpu
       // [NB do keep 'static' for these constexpr arrays, see issue #283]
       static constexpr fptype2 cf[ncolor][ncolor] = { { 1 } }; // 2-D array[1][1]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -341,7 +342,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -400,7 +401,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -447,8 +448,8 @@ namespace mg5amcCpu
       { 1, -1, 1, 1 },
       { 1, -1, -1, -1 },
       { 1, -1, -1, 1 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -488,9 +489,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     const fptype tIPD[2] = { (fptype)m_pars->mdl_MZ, (fptype)m_pars->mdl_WZ };
     const cxtype tIPC[3] = { cxmake( m_pars->GC_3 ), cxmake( m_pars->GC_50 ), cxmake( m_pars->GC_59 ) };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) ) );
-    checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 3 * sizeof( cxtype ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) );
+    gpuMemcpyToSymbol( cIPC, tIPC, 3 * sizeof( cxtype ) );
 #else
     memcpy( cIPD, tIPD, 2 * sizeof( fptype ) );
     memcpy( cIPC, tIPC, 3 * sizeof( cxtype ) );
@@ -527,7 +528,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -556,6 +557,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -569,6 +574,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -592,12 +599,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -618,7 +625,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -744,9 +751,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -770,7 +777,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -790,7 +797,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 4 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -804,9 +811,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -834,7 +844,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -1044,7 +1054,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/P1_Sigma_sm_epem_mupmum/CPPProcess.h b/epochX/cudacpp/ee_mumu.sa/SubProcesses/P1_Sigma_sm_epem_mupmum/CPPProcess.h
index 77b610753c..0b29ffb3ff 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/P1_Sigma_sm_epem_mupmum/CPPProcess.h
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/P1_Sigma_sm_epem_mupmum/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -107,7 +107,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -120,7 +120,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -150,7 +150,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/P1_Sigma_sm_epem_mupmum/CudaRuntime.h b/epochX/cudacpp/ee_mumu.sa/SubProcesses/P1_Sigma_sm_epem_mupmum/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/P1_Sigma_sm_epem_mupmum/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/P1_Sigma_sm_epem_mupmum/GpuAbstraction.h b/epochX/cudacpp/ee_mumu.sa/SubProcesses/P1_Sigma_sm_epem_mupmum/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/P1_Sigma_sm_epem_mupmum/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/P1_Sigma_sm_epem_mupmum/GpuRuntime.h b/epochX/cudacpp/ee_mumu.sa/SubProcesses/P1_Sigma_sm_epem_mupmum/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/P1_Sigma_sm_epem_mupmum/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/P1_Sigma_sm_epem_mupmum/check_sa.cc b/epochX/cudacpp/ee_mumu.sa/SubProcesses/P1_Sigma_sm_epem_mupmum/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/P1_Sigma_sm_epem_mupmum/check_sa.cc
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/P1_Sigma_sm_epem_mupmum/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/RamboSamplingKernels.cc b/epochX/cudacpp/ee_mumu.sa/SubProcesses/RamboSamplingKernels.cc
index da68aa9255..79abbcc4f8 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/RamboSamplingKernels.cc
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/RamboSamplingKernels.cc
@@ -1,11 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "RamboSamplingKernels.h"
 
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "MemoryAccessMomenta.h"
 #include "MemoryAccessRandomNumbers.h"
 #include "MemoryAccessWeights.h"
@@ -14,7 +14,7 @@
 
 #include <sstream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -92,7 +92,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingKernelDevice::RamboSamplingKernelDevice( const fptype energy,               // input: energy
                                                         const BufferRndNumMomenta& rndmom, // input: random numbers in [0,1]
                                                         BufferMomenta& momenta,            // output: momenta
@@ -135,7 +135,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   getMomentaInitialDevice( const fptype energy,
                            fptype* momenta )
@@ -147,17 +147,17 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   void
   RamboSamplingKernelDevice::getMomentaInitial()
   {
-    getMomentaInitialDevice<<<m_gpublocks, m_gputhreads>>>( m_energy, m_momenta.data() );
+    gpuLaunchKernel( getMomentaInitialDevice, m_gpublocks, m_gputhreads, m_energy, m_momenta.data() );
   }
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   getMomentaFinalDevice( const fptype energy,
                          const fptype* rndmom,
@@ -171,11 +171,11 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   void
   RamboSamplingKernelDevice::getMomentaFinal()
   {
-    getMomentaFinalDevice<<<m_gpublocks, m_gputhreads>>>( m_energy, m_rndmom.data(), m_momenta.data(), m_weights.data() );
+    gpuLaunchKernel( getMomentaFinalDevice, m_gpublocks, m_gputhreads, m_energy, m_rndmom.data(), m_momenta.data(), m_weights.data() );
   }
 #endif
 
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/RamboSamplingKernels.h b/epochX/cudacpp/ee_mumu.sa/SubProcesses/RamboSamplingKernels.h
index 184089efd7..7c214cd74b 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/RamboSamplingKernels.h
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/RamboSamplingKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef RAMBOSAMPLINGKERNELS_H
 #define RAMBOSAMPLINGKERNELS_H 1
@@ -10,7 +10,7 @@
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -93,7 +93,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating RAMBO phase space sampling on a GPU device
   class RamboSamplingKernelDevice final : public SamplingKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/RandomNumberKernels.h b/epochX/cudacpp/ee_mumu.sa/SubProcesses/RandomNumberKernels.h
index 188a72c2c9..21d63beeac 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/RandomNumberKernels.h
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/RandomNumberKernels.h
@@ -1,14 +1,14 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef RANDOMNUMBERKERNELS_H
 #define RANDOMNUMBERKERNELS_H 1
 
 #include "mgOnGpuConfig.h"
 
-// NB This must come AFTER mgOnGpuConfig.h which contains our definition of __global__ when __CUDACC__ is not defined
+// NB This must come AFTER mgOnGpuConfig.h which contains our definition of __global__ when MGONGPUCPP_GPUIMPL is not defined
 #ifndef MGONGPU_HAS_NO_CURAND
 //#include "curand.h"
 struct curandGenerator_st; // forward definition from curand.h
@@ -16,7 +16,7 @@ struct curandGenerator_st; // forward definition from curand.h
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/cudacpp.mk b/epochX/cudacpp/ee_mumu.sa/SubProcesses/cudacpp.mk
index 509307506b..df74dfc284 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/cudacpp.mk
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/cudacpp.mk
@@ -1,7 +1,7 @@
 # Copyright (C) 2020-2023 CERN and UCLouvain.
 # Licensed under the GNU Lesser General Public License (version 3 or later).
 # Created by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-# Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+# Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #=== Determine the name of this makefile (https://ftp.gnu.org/old-gnu/Manuals/make-3.80/html_node/make_17.html)
 #=== NB: use ':=' to ensure that the value of CUDACPP_MAKEFILE is not modified further down after including make_opts
@@ -42,10 +42,10 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure common compiler flags for C++ and CUDA
+#=== Configure common compiler flags for C++ and CUDA/HIP
 
 INCFLAGS = -I.
-OPTFLAGS = -O3 # this ends up in CUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
+OPTFLAGS = -O3 # this ends up in GPUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
 
 # Dependency on src directory
 MG5AMC_COMMONLIB = mg5amc_common
@@ -121,24 +121,45 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure the CUDA compiler
+#=== Configure the GPU compiler (CUDA or HIP)
+
+# FIXME! (AV 24.01.2024)
+# In the current implementation (without separate builds for C++ and CUDA/HIP), we first check for cudacc and hipcc in CUDA_HOME and HIP_HOME.
+# If CUDA_HOME or HIP_HOME are not set, try to determine them from the path to cudacc and hipcc.
+# While convoluted, this is currently necessary to allow disabling CUDA/HIP builds by setting CUDA_HOME or HIP_HOME to invalid paths.
+# This will (probably?) be fixed when separate C++ and CUDA/HIP builds are implemented (PR #775).
 
 # If CXX is not a single word (example "clang++ --gcc-toolchain...") then disable CUDA builds (issue #505)
-# This is because it is impossible to pass this to "CUFLAGS += -ccbin <host-compiler>" below
+# This is because it is impossible to pass this to "GPUFLAGS += -ccbin <host-compiler>" below
 ifneq ($(words $(subst ccache ,,$(CXX))),1) # allow at most "CXX=ccache <host-compiler>" from outside
   $(warning CUDA builds are not supported for multi-word CXX "$(CXX)")
   override CUDA_HOME=disabled
 endif
 
-# If CUDA_HOME is not set, try to set it from the location of nvcc
+# If CUDA_HOME is not set, try to set it from the path to nvcc
 ifndef CUDA_HOME
-  CUDA_HOME = $(patsubst %bin/nvcc,%,$(shell which nvcc 2>/dev/null))
+  CUDA_HOME = $(patsubst %/bin/nvcc,%,$(shell which nvcc 2>/dev/null))
   $(warning CUDA_HOME was not set: using "$(CUDA_HOME)")
 endif
 
-# Set NVCC as $(CUDA_HOME)/bin/nvcc if it exists
+# If HIP_HOME is not set, try to set it from the path to hipcc
+ifndef HIP_HOME
+  HIP_HOME = $(patsubst %/bin/hipcc,%,$(shell which hipcc 2>/dev/null))
+  $(warning HIP_HOME was not set: using "$(HIP_HOME)")
+endif
+
+# FIXME! (AV 24.01.2024)
+# In the current implementation (without separate builds for C++ and CUDA/HIP),
+# builds are performed for HIP only if CUDA is not found in the path.
+# If both CUDA and HIP are installed, HIP builds can be triggered by unsetting CUDA_HOME.
+# This will be fixed when separate C++ and CUDA/HIP builds are implemented (PR #775).
+
+#--- Option 1: CUDA exists -> use CUDA
+
+# Set GPUCC as $(CUDA_HOME)/bin/nvcc if it exists
 ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
-  NVCC = $(CUDA_HOME)/bin/nvcc
+
+  GPUCC = $(CUDA_HOME)/bin/nvcc
   USE_NVTX ?=-DUSE_NVTX
   # See https://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/index.html
   # See https://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/
@@ -148,7 +169,7 @@ ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
   # Examples: use 60 for P100 (Piz Daint), 80 for A100 (Juwels Booster, NVidia raplab/Curiosity).
   MADGRAPH_CUDA_ARCHITECTURE ?= 70
   ###CUARCHFLAGS = -gencode arch=compute_$(MADGRAPH_CUDA_ARCHITECTURE),code=compute_$(MADGRAPH_CUDA_ARCHITECTURE) -gencode arch=compute_$(MADGRAPH_CUDA_ARCHITECTURE),code=sm_$(MADGRAPH_CUDA_ARCHITECTURE) # Older implementation (AV): go back to this one for multi-GPU support #533
-  ###CUARCHFLAGS = --gpu-architecture=compute_$(MADGRAPH_CUDA_ARCHITECTURE) --gpu-code=sm_$(MADGRAPH_CUDA_ARCHITECTURE),compute_$(MADGRAPH_CUDA_ARCHITECTURE) # Newer implementation (SH): cannot use this as-is for multi-GPU support #533
+  ###CUARCHFLAGS = --gpu-architecture=compute_$(MADGRAPH_CUDA_ARCHITECTURE) --gpu-code=sm_$(MADGRAPH_CUDA_ARCHITECTURE),compute_$(MADGRAPH_CUDA_ARCHITECTURE)  # Newer implementation (SH): cannot use this as-is for multi-GPU support #533
   comma:=,
   CUARCHFLAGS = $(foreach arch,$(subst $(comma), ,$(MADGRAPH_CUDA_ARCHITECTURE)),-gencode arch=compute_$(arch),code=compute_$(arch) -gencode arch=compute_$(arch),code=sm_$(arch))
   CUINC = -I$(CUDA_HOME)/include/
@@ -158,41 +179,79 @@ ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
     CURANDLIBFLAGS = -L$(CUDA_HOME)/lib64/ -lcurand # NB: -lcuda is not needed here!
   endif
   CUOPTFLAGS = -lineinfo
-  CUFLAGS = $(foreach opt, $(OPTFLAGS), -Xcompiler $(opt)) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
-  ###CUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
-  ###NVCC_VERSION = $(shell $(NVCC) --version | grep 'Cuda compilation tools' | cut -d' ' -f5 | cut -d, -f1)
-  CUFLAGS += -std=c++17 # need CUDA >= 11.2 (see #333): this is enforced in mgOnGpuConfig.h
+  ###GPUFLAGS = $(OPTFLAGS) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
+  GPUFLAGS = $(foreach opt, $(OPTFLAGS), -Xcompiler $(opt)) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
+  ###GPUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
+  ###GPUCC_VERSION = $(shell $(GPUCC) --version | grep 'Cuda compilation tools' | cut -d' ' -f5 | cut -d, -f1)
+  GPUFLAGS += -std=c++17 # need CUDA >= 11.2 (see #333): this is enforced in mgOnGpuConfig.h
   # Without -maxrregcount: baseline throughput: 6.5E8 (16384 32 12) up to 7.3E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 160 # improves throughput: 6.9E8 (16384 32 12) up to 7.7E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 128 # improves throughput: 7.3E8 (16384 32 12) up to 7.6E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 96 # degrades throughput: 4.1E8 (16384 32 12) up to 4.5E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 64 # degrades throughput: 1.7E8 (16384 32 12) flat at 1.7E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 160 # improves throughput: 6.9E8 (16384 32 12) up to 7.7E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 128 # improves throughput: 7.3E8 (16384 32 12) up to 7.6E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 96 # degrades throughput: 4.1E8 (16384 32 12) up to 4.5E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 64 # degrades throughput: 1.7E8 (16384 32 12) flat at 1.7E8 (65536 128 12)
+  CUBUILDRULEFLAGS = -Xcompiler -fPIC -c
+  CCBUILDRULEFLAGS = -Xcompiler -fPIC -c -x cu
+  CUDATESTFLAGS = -lcuda
+
+  # Set the host C++ compiler for GPUCC via "-ccbin <host-compiler>"
+  # (NB issue #505: this must be a single word, "clang++ --gcc-toolchain..." is not supported)
+  GPUFLAGS += -ccbin $(shell which $(subst ccache ,,$(CXX)))
+
+  # Allow newer (unsupported) C++ compilers with older versions of CUDA if ALLOW_UNSUPPORTED_COMPILER_IN_CUDA is set (#504)
+  ifneq ($(origin ALLOW_UNSUPPORTED_COMPILER_IN_CUDA),undefined)
+  GPUFLAGS += -allow-unsupported-compiler
+  endif
+
 else ifneq ($(origin REQUIRE_CUDA),undefined)
+
   # If REQUIRE_CUDA is set but no cuda is found, stop here (e.g. for CI tests on GPU #443)
-  $(error No cuda installation found (set CUDA_HOME or make nvcc visible in PATH))
+  $(error No cuda installation found (set CUDA_HOME or make GPUCC visible in PATH))
+
+#--- Option 2: CUDA does not exist, HIP exists -> use HIP
+
+# Set GPUCC as $(HIP_HOME)/bin/hipcc if it exists
+else ifneq ($(wildcard $(HIP_HOME)/bin/hipcc),)
+
+  GPUCC = $(HIP_HOME)/bin/hipcc
+  #USE_NVTX ?=-DUSE_NVTX # should maybe find something equivalent to this in HIP?
+  HIPARCHFLAGS = -target x86_64-linux-gnu --offload-arch=gfx90a
+  HIPINC = -I$(HIP_HOME)/include/
+  # Note: -DHIP_FAST_MATH is equivalent to -use_fast_math in HIP 
+  # (but only for single precision line 208: https://rocm-developer-tools.github.io/HIP/hcc__detail_2math__functions_8h_source.html)
+  # Note: CUOPTFLAGS should not be used for HIP, it had been added here but was then removed (#808)
+  GPUFLAGS = $(OPTFLAGS) $(INCFLAGS) $(HIPINC) $(HIPARCHFLAGS) -DHIP_FAST_MATH -DHIP_PLATFORM=amd -fPIC
+  ###GPUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
+  GPUFLAGS += -std=c++17
+  ###GPUFLAGS+= --maxrregcount 255 # (AV: is this option valid on HIP and meaningful on AMD GPUs?)
+  CUBUILDRULEFLAGS = -fPIC -c
+  CCBUILDRULEFLAGS = -fPIC -c -x hip
+
+else ifneq ($(origin REQUIRE_HIP),undefined)
+
+  # If REQUIRE_HIP is set but no HIP is found, stop here (e.g. for CI tests on GPU #443)
+  $(error No hip installation found (set HIP_HOME or make GPUCC visible in PATH))
+
+#--- Option 3: CUDA does not exist, HIP does not exist -> switch off both CUDA and HIP
+
 else
-  # No cuda. Switch cuda compilation off and go to common random numbers in C++
+
+  # No cudacc and no hipcc: switch CUDA and HIP compilation off and go to common random numbers in C++
   $(warning CUDA_HOME is not set or is invalid: export CUDA_HOME to compile with cuda)
-  override NVCC=
+  $(warning HIP_HOME is not set or is invalid: export HIP_HOME to compile with hip)
+  override GPUCC=
   override USE_NVTX=
   override CUINC=
   override CURANDLIBFLAGS=
-endif
-export NVCC
-export CUFLAGS
 
-# Set the host C++ compiler for nvcc via "-ccbin <host-compiler>"
-# (NB issue #505: this must be a single word, "clang++ --gcc-toolchain..." is not supported)
-CUFLAGS += -ccbin $(shell which $(subst ccache ,,$(CXX)))
-
-# Allow newer (unsupported) C++ compilers with older versions of CUDA if ALLOW_UNSUPPORTED_COMPILER_IN_CUDA is set (#504)
-ifneq ($(origin ALLOW_UNSUPPORTED_COMPILER_IN_CUDA),undefined)
-CUFLAGS += -allow-unsupported-compiler
 endif
 
+# Export GPUCC (so that it can also be used in cudacpp_src.mk?)
+export GPUCC
+export GPUFLAGS
+
 #-------------------------------------------------------------------------------
 
-#=== Configure ccache for C++ and CUDA builds
+#=== Configure ccache for C++ and CUDA/HIP builds
 
 # Enable ccache if USECCACHE=1
 ifeq ($(USECCACHE)$(shell echo $(CXX) | grep ccache),1)
@@ -201,15 +260,15 @@ endif
 #ifeq ($(USECCACHE)$(shell echo $(AR) | grep ccache),1)
 #  override AR:=ccache $(AR)
 #endif
-ifneq ($(NVCC),)
-  ifeq ($(USECCACHE)$(shell echo $(NVCC) | grep ccache),1)
-    override NVCC:=ccache $(NVCC)
+ifneq ($(GPUCC),)
+  ifeq ($(USECCACHE)$(shell echo $(GPUCC) | grep ccache),1)
+    override GPUCC:=ccache $(GPUCC)
   endif
 endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure PowerPC-specific compiler flags for C++ and CUDA
+#=== Configure PowerPC-specific compiler flags for C++ and CUDA/HIP
 
 # PowerPC-specific CXX compiler flags (being reviewed)
 ifeq ($(UNAME_P),ppc64le)
@@ -225,9 +284,9 @@ else
   ######CXXFLAGS+= -fno-semantic-interposition # no benefit (neither alone, nor combined with -flto)
 endif
 
-# PowerPC-specific CUDA compiler flags (to be reviewed!)
+# PowerPC-specific CUDA/HIP compiler flags (to be reviewed!)
 ifeq ($(UNAME_P),ppc64le)
-  CUFLAGS+= -Xcompiler -mno-float128
+  GPUFLAGS+= -Xcompiler -mno-float128
 endif
 
 #-------------------------------------------------------------------------------
@@ -235,9 +294,11 @@ endif
 #=== Configure defaults and check if user-defined choices exist for OMPFLAGS, AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
 
 # Set the default OMPFLAGS choice
-ifneq ($(shell $(CXX) --version | egrep '^Intel'),)
+ifneq ($(findstring hipcc,$(GPUCC)),)
+override OMPFLAGS = # disable OpenMP MT when using hipcc #802
+else ifneq ($(shell $(CXX) --version | egrep '^Intel'),)
 override OMPFLAGS = -fopenmp
-###override OMPFLAGS = # disable OpenMP MT on Intel (was ok without nvcc but not ok with nvcc before #578)
+###override OMPFLAGS = # disable OpenMP MT on Intel (was ok without GPUCC but not ok with GPUCC before #578)
 else ifneq ($(shell $(CXX) --version | egrep '^(clang)'),)
 override OMPFLAGS = -fopenmp
 ###override OMPFLAGS = # disable OpenMP MT on clang (was not ok without or with nvcc before #578)
@@ -293,7 +354,10 @@ endif
 
 # Set the default RNDGEN (random number generator) choice
 ifeq ($(RNDGEN),)
-  ifeq ($(NVCC),)
+  ifeq ($(GPUCC),)
+    override RNDGEN = hasNoCurand
+  # Edgecase for HIP compilation
+  else ifeq ($(findstring hipcc,$(GPUCC)),hipcc)
     override RNDGEN = hasNoCurand
   else ifeq ($(RNDGEN),)
     override RNDGEN = hasCurand
@@ -310,7 +374,7 @@ export OMPFLAGS
 
 #-------------------------------------------------------------------------------
 
-#=== Set the CUDA/C++ compiler flags appropriate to user-defined choices of AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
+#=== Set the CUDA/HIP/C++ compiler flags appropriate to user-defined choices of AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
 
 # Set the build flags appropriate to OMPFLAGS
 $(info OMPFLAGS=$(OMPFLAGS))
@@ -368,13 +432,13 @@ CXXFLAGS+= $(AVXFLAGS)
 $(info FPTYPE=$(FPTYPE))
 ifeq ($(FPTYPE),d)
   CXXFLAGS += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
-  CUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
+  GPUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
 else ifeq ($(FPTYPE),f)
   CXXFLAGS += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
-  CUFLAGS  += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
+  GPUFLAGS  += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
 else ifeq ($(FPTYPE),m)
   CXXFLAGS += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
-  CUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
+  GPUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
 else
   $(error Unknown FPTYPE='$(FPTYPE)': only 'd', 'f' and 'm' are supported)
 endif
@@ -383,7 +447,7 @@ endif
 $(info HELINL=$(HELINL))
 ifeq ($(HELINL),1)
   CXXFLAGS += -DMGONGPU_INLINE_HELAMPS
-  CUFLAGS  += -DMGONGPU_INLINE_HELAMPS
+  GPUFLAGS  += -DMGONGPU_INLINE_HELAMPS
 else ifneq ($(HELINL),0)
   $(error Unknown HELINL='$(HELINL)': only '0' and '1' are supported)
 endif
@@ -392,7 +456,7 @@ endif
 $(info HRDCOD=$(HRDCOD))
 ifeq ($(HRDCOD),1)
   CXXFLAGS += -DMGONGPU_HARDCODE_PARAM
-  CUFLAGS  += -DMGONGPU_HARDCODE_PARAM
+  GPUFLAGS  += -DMGONGPU_HARDCODE_PARAM
 else ifneq ($(HRDCOD),0)
   $(error Unknown HRDCOD='$(HRDCOD)': only '0' and '1' are supported)
 endif
@@ -444,11 +508,11 @@ ifeq ($(UNAME_S),Darwin)
   override CULIBFLAGSRPATH2 =
 else
   # RPATH to cuda/cpp libs when linking executables
-  override CXXLIBFLAGSRPATH = -Wl,-rpath,$(LIBDIRRPATH)
-  override CULIBFLAGSRPATH = -Xlinker -rpath,$(LIBDIRRPATH)
+  override CXXLIBFLAGSRPATH = -Wl,-rpath=$(LIBDIRRPATH)
+  override CULIBFLAGSRPATH = -Xlinker -rpath=$(LIBDIRRPATH)
   # RPATH to common lib when linking cuda/cpp libs
-  override CXXLIBFLAGSRPATH2 = -Wl,-rpath,'$$ORIGIN'
-  override CULIBFLAGSRPATH2 = -Xlinker -rpath,'$$ORIGIN'
+  override CXXLIBFLAGSRPATH2 = -Wl,-rpath='$$ORIGIN'
+  override CULIBFLAGSRPATH2 = -Xlinker -rpath='$$ORIGIN'
 endif
 
 # Setting LD_LIBRARY_PATH or DYLD_LIBRARY_PATH in the RUNTIME is no longer necessary (neither on Linux nor on Mac)
@@ -461,7 +525,7 @@ override RUNTIME =
 cxx_main=$(BUILDDIR)/check.exe
 fcxx_main=$(BUILDDIR)/fcheck.exe
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 cu_main=$(BUILDDIR)/gcheck.exe
 fcu_main=$(BUILDDIR)/fgcheck.exe
 else
@@ -492,14 +556,15 @@ $(BUILDDIR)/.build.$(TAG):
 	@touch $(BUILDDIR)/.build.$(TAG)
 
 # Generic target and build rules: objects from CUDA compilation
-ifneq ($(NVCC),)
+# NB: CCBUILDRULEFLAGS includes "-x cu" for nvcc and "-x hip" for hipcc (#810)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/%.o : %.cu *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $(CUBUILDRULEFLAGS) $< -o $@
 
 $(BUILDDIR)/%_cu.o : %.cc *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c -x cu $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $(CCBUILDRULEFLAGS) $< -o $@
 endif
 
 # Generic target and build rules: objects from C++ compilation
@@ -509,11 +574,14 @@ $(BUILDDIR)/%.o : %.cc *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	$(CXX) $(CPPFLAGS) $(CXXFLAGS) -fPIC -c $< -o $@
 
 # Apply special build flags only to CrossSectionKernel.cc and gCrossSectionKernel.cu (no fast math, see #117 and #516)
+# Added edgecase for HIP compilation
 ifeq ($(shell $(CXX) --version | grep ^nvc++),)
 $(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS := $(filter-out -ffast-math,$(CXXFLAGS))
 $(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS += -fno-fast-math
-ifneq ($(NVCC),)
-$(BUILDDIR)/gCrossSectionKernels.o: CUFLAGS += -Xcompiler -fno-fast-math
+ifeq ($(findstring nvcc,$(GPUCC)),nvcc)
+  $(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -Xcompiler -fno-fast-math
+else
+  $(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -fno-fast-math
 endif
 endif
 
@@ -530,10 +598,10 @@ ifeq ($(RNDGEN),hasCurand)
 $(BUILDDIR)/CurandRandomNumberKernel.o: CXXFLAGS += $(CUINC)
 endif
 
-# Avoid "warning: builtin __has_trivial_... is deprecated; use __is_trivially_... instead" in nvcc with icx2023 (#592)
+# Avoid "warning: builtin __has_trivial_... is deprecated; use __is_trivially_... instead" in GPUCC with icx2023 (#592)
 ifneq ($(shell $(CXX) --version | egrep '^(Intel)'),)
-ifneq ($(NVCC),)
-CUFLAGS += -Xcompiler -Wno-deprecated-builtins
+ifneq ($(GPUCC),)
+GPUFLAGS += -Wno-deprecated-builtins
 endif
 endif
 
@@ -541,8 +609,8 @@ endif
 # This patch does remove the warning, but I prefer to keep it disabled for the moment...
 ###ifneq ($(shell $(CXX) --version | egrep '^(clang|Apple clang|Intel)'),)
 ###$(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS += -Wno-overriding-t-option
-###ifneq ($(NVCC),)
-###$(BUILDDIR)/gCrossSectionKernels.o: CUFLAGS += -Xcompiler -Wno-overriding-t-option
+###ifneq ($(GPUCC),)
+###$(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -Xcompiler -Wno-overriding-t-option
 ###endif
 ###endif
 
@@ -569,7 +637,7 @@ MG5AMC_CXXLIB = mg5amc_$(processid_short)_cpp
 cxx_objects_lib=$(BUILDDIR)/CPPProcess.o $(BUILDDIR)/MatrixElementKernels.o $(BUILDDIR)/BridgeKernels.o $(BUILDDIR)/CrossSectionKernels.o
 cxx_objects_exe=$(BUILDDIR)/CommonRandomNumberKernel.o $(BUILDDIR)/RamboSamplingKernels.o
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 MG5AMC_CULIB = mg5amc_$(processid_short)_cuda
 cu_objects_lib=$(BUILDDIR)/gCPPProcess.o $(BUILDDIR)/gMatrixElementKernels.o $(BUILDDIR)/gBridgeKernels.o $(BUILDDIR)/gCrossSectionKernels.o
 cu_objects_exe=$(BUILDDIR)/gCommonRandomNumberKernel.o $(BUILDDIR)/gRamboSamplingKernels.o
@@ -581,11 +649,17 @@ $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: cxx_objects_lib += $(BUILDDIR)/fbridge.o
 $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib)
 	$(CXX) -shared -o $@ $(cxx_objects_lib) $(CXXLIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: $(BUILDDIR)/fbridge_cu.o
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: cu_objects_lib += $(BUILDDIR)/fbridge_cu.o
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cu_objects_lib)
-	$(NVCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+# Bypass std::filesystem completely to ease portability on LUMI #803
+#ifneq ($(findstring hipcc,$(GPUCC)),)
+#	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB) -lstdc++fs
+#else
+#	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+#endif
 endif
 
 #-------------------------------------------------------------------------------
@@ -602,16 +676,16 @@ $(cxx_main): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PAT
 $(cxx_main): $(BUILDDIR)/check_sa.o $(LIBDIR)/lib$(MG5AMC_CXXLIB).so $(cxx_objects_exe) $(BUILDDIR)/CurandRandomNumberKernel.o
 	$(CXX) -o $@ $(BUILDDIR)/check_sa.o $(OMPFLAGS) -ldl -pthread $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe) $(BUILDDIR)/CurandRandomNumberKernel.o $(CURANDLIBFLAGS)
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(cu_main): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(cu_main): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(cu_main): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(cu_main): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 else ifneq ($(shell $(CXX) --version | grep ^nvc++),) # support nvc++ #531
 $(cu_main): LIBFLAGS += -L$(patsubst %bin/nvc++,%lib,$(subst ccache ,,$(CXX))) -lnvhpcatm -lnvcpumath -lnvc
 endif
 $(cu_main): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(cu_main): $(BUILDDIR)/gcheck_sa.o $(LIBDIR)/lib$(MG5AMC_CULIB).so $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o
-	$(NVCC) -o $@ $(BUILDDIR)/gcheck_sa.o $(CUARCHFLAGS) $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o $(CURANDLIBFLAGS)
+	$(GPUCC) -o $@ $(BUILDDIR)/gcheck_sa.o $(CUARCHFLAGS) $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o $(CURANDLIBFLAGS)
 endif
 
 #-------------------------------------------------------------------------------
@@ -635,19 +709,27 @@ $(fcxx_main): LIBFLAGS += -L$(shell dirname $(shell $(FC) --print-file-name libg
 endif
 $(fcxx_main): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(fcxx_main): $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler.o $(LIBDIR)/lib$(MG5AMC_CXXLIB).so $(cxx_objects_exe)
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(BUILDDIR)/fcheck_sa.o $(OMPFLAGS) $(BUILDDIR)/fsampler.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe) -lstdc++
+else
 	$(CXX) -o $@ $(BUILDDIR)/fcheck_sa.o $(OMPFLAGS) $(BUILDDIR)/fsampler.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe)
+endif
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(fcu_main): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(fcu_main): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(fcu_main): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(fcu_main): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 endif
 ifeq ($(UNAME_S),Darwin)
 $(fcu_main): LIBFLAGS += -L$(shell dirname $(shell $(FC) --print-file-name libgfortran.dylib)) # add path to libgfortran on Mac #375
 endif
 $(fcu_main): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(fcu_main): $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBDIR)/lib$(MG5AMC_CULIB).so $(cu_objects_exe)
-	$(NVCC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe)
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) -lstdc++ -L$(shell dirname $(shell $(GPUCC) -print-prog-name=clang))/../../lib -lamdhip64
+else
+	$(GPUCC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe)
+endif
 endif
 
 #-------------------------------------------------------------------------------
@@ -659,7 +741,7 @@ $(BUILDDIR)/testxxx.o: testxxx_cc_ref.txt
 $(testmain): $(BUILDDIR)/testxxx.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/testxxx.o # Comment out this line to skip the C++ test of xxx functions
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/testxxx_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/testxxx_cu.o: INCFLAGS += $(GTESTINC)
 $(BUILDDIR)/testxxx_cu.o: testxxx_cc_ref.txt
@@ -672,7 +754,7 @@ $(BUILDDIR)/testmisc.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/testmisc.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/testmisc.o # Comment out this line to skip the C++ miscellaneous tests
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/testmisc_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/testmisc_cu.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/testmisc_cu.o
@@ -684,12 +766,12 @@ $(BUILDDIR)/runTest.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/runTest.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/runTest.o
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/runTest_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/runTest_cu.o: INCFLAGS += $(GTESTINC)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(testmain): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(testmain): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(testmain): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(testmain): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 else ifneq ($(shell $(CXX) --version | grep ^nvc++),) # support nvc++ #531
 $(testmain): LIBFLAGS += -L$(patsubst %bin/nvc++,%lib,$(subst ccache ,,$(CXX))) -lnvhpcatm -lnvcpumath -lnvc
 endif
@@ -713,14 +795,23 @@ $(testmain): LIBFLAGS += -lgomp
 endif
 endif
 
-ifeq ($(NVCC),) # link only runTest.o
+# Bypass std::filesystem completely to ease portability on LUMI #803
+#ifneq ($(findstring hipcc,$(GPUCC)),)
+#$(testmain): LIBFLAGS += -lstdc++fs
+#endif
+
+ifeq ($(GPUCC),) # link only runTest.o
 $(testmain): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(testmain): $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib) $(cxx_objects_exe) $(GTESTLIBS)
 	$(CXX) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) -ldl -pthread $(LIBFLAGS)
 else # link both runTest.o and runTest_cu.o
 $(testmain): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(testmain): $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) $(GTESTLIBS)
-	$(NVCC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) -lcuda
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) $(CUDATESTFLAGS) -lstdc++ -lpthread  -L$(shell dirname $(shell $(GPUCC) -print-prog-name=clang))/../../lib -lamdhip64
+else
+	$(GPUCC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) $(CUDATESTFLAGS)
+endif
 endif
 
 # Use target gtestlibs to build only googletest
@@ -829,9 +920,9 @@ ifeq ($(USECCACHE),1)
 	ccache --version | head -1
 endif
 	@echo ""
-	@echo NVCC=$(NVCC)
-ifneq ($(NVCC),)
-	$(NVCC) --version
+	@echo GPUCC=$(GPUCC)
+ifneq ($(GPUCC),)
+	$(GPUCC) --version
 endif
 	@echo ""
 	@echo CXX=$(CXX)
@@ -850,7 +941,7 @@ endif
 
 # Target: check (run the C++ test executable)
 # [NB THIS IS WHAT IS USED IN THE GITHUB CI!]
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 check: runTest cmpFcheck cmpFGcheck
 else
 check: runTest cmpFcheck
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/fbridge.cc b/epochX/cudacpp/ee_mumu.sa/SubProcesses/fbridge.cc
index 2d2b36d560..22ce3f5115 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/fbridge.cc
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/fbridge.cc
@@ -1,11 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Roiser (Oct 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "Bridge.h"
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 
 extern "C"
 {
@@ -22,7 +22,7 @@ extern "C"
    * Using the same Fortran MadEvent code, linking to the hetrerogeneous library would allow access to both CPU and GPU implementations.
    * The specific heterogeneous configuration (how many GPUs, how many threads on each CPU, etc) could be loaded in CUDA/C++ from a data file.
    */
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -46,8 +46,8 @@ extern "C"
    */
   void fbridgecreate_( CppObjectInFortran** ppbridge, const int* pnevtF, const int* pnparF, const int* pnp4F )
   {
-#ifdef __CUDACC__
-    CudaRuntime::setUp();
+#ifdef MGONGPUCPP_GPUIMPL
+    GpuRuntime::setUp();
 #endif
     // (NB: CPPProcess::initProc no longer needs to be executed here because it is called in the Bridge constructor)
     // FIXME: disable OMP in Bridge when called from Fortran
@@ -65,8 +65,8 @@ extern "C"
     Bridge<FORTRANFPTYPE>* pbridge = dynamic_cast<Bridge<FORTRANFPTYPE>*>( *ppbridge );
     if( pbridge == 0 ) throw std::runtime_error( "fbridgedelete_: invalid Bridge address" );
     delete pbridge;
-#ifdef __CUDACC__
-    CudaRuntime::tearDown();
+#ifdef MGONGPUCPP_GPUIMPL
+    GpuRuntime::tearDown();
 #endif
   }
 
@@ -96,7 +96,7 @@ extern "C"
   {
     Bridge<FORTRANFPTYPE>* pbridge = dynamic_cast<Bridge<FORTRANFPTYPE>*>( *ppbridge );
     if( pbridge == 0 ) throw std::runtime_error( "fbridgesequence_: invalid Bridge address" );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Use the device/GPU implementation in the CUDA library
     // (there is also a host implementation in this library)
     pbridge->gpu_sequence( momenta, gs, rndhel, rndcol, *pchannelId, mes, selhel, selcol );
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/fsampler.cc b/epochX/cudacpp/ee_mumu.sa/SubProcesses/fsampler.cc
index 2fb445372d..3743934f41 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/fsampler.cc
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/fsampler.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Feb 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "mgOnGpuConfig.h"
 
@@ -13,7 +13,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -40,7 +40,7 @@ namespace mg5amcCpu
   private:
     const int m_nevt; // The number of events in each iteration
     int m_iiter;      // The iteration counter (for random number seeding)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     HostBufferRndNumMomenta m_hstRndmom; // Memory buffers for random numbers
     HostBufferMomenta m_hstMomenta;      // Memory buffers for momenta
     HostBufferWeights m_hstWeights;      // Memory buffers for sampling weights
@@ -105,7 +105,7 @@ namespace mg5amcCpu
 
 extern "C"
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/runTest.cc b/epochX/cudacpp/ee_mumu.sa/SubProcesses/runTest.cc
index d4a760a71b..de327f2321 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/runTest.cc
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/runTest.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Hageboeck (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*xxx to run only testxxx.cc tests
 //----------------------------------------------------------------------------
@@ -18,7 +18,7 @@
 #include "RandomNumberKernels.h"
 #include "epoch_process_id.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using namespace mg5amcGpu;
 #else
 using namespace mg5amcCpu;
@@ -35,7 +35,7 @@ struct CUDA_CPU_TestBase : public TestDriverBase
     : TestDriverBase( npar, refFileName ) {}
 };
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 struct CPUTest : public CUDA_CPU_TestBase
 {
   // Struct data members (process, and memory structures for random numbers, momenta, matrix elements and weights on host and device)
@@ -119,7 +119,7 @@ struct CPUTest : public CUDA_CPU_TestBase
 };
 #endif
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 struct CUDATest : public CUDA_CPU_TestBase
 {
   // Reset the device when our test goes out of scope. Note that this should happen after
@@ -128,7 +128,7 @@ struct CUDATest : public CUDA_CPU_TestBase
   {
     ~DeviceReset()
     {
-      checkCuda( cudaDeviceReset() ); // this is needed by cuda-memcheck --leak-check full
+      checkGpu( gpuDeviceReset() ); // this is needed by cuda-memcheck --leak-check full
     }
   } deviceResetter;
 
@@ -256,7 +256,7 @@ INSTANTIATE_TEST_SUITE_P( prefix, \
                           test_suite_name, \
                           testing::Values( new CUDATest( MG_EPOCH_REFERENCE_FILE_NAME ) ) );
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 MG_INSTANTIATE_TEST_SUITE_GPU( XTESTID_GPU( MG_EPOCH_PROCESS_ID ), MadgraphTest );
 #else
 MG_INSTANTIATE_TEST_SUITE_CPU( XTESTID_CPU( MG_EPOCH_PROCESS_ID ), MadgraphTest );
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/testmisc.cc b/epochX/cudacpp/ee_mumu.sa/SubProcesses/testmisc.cc
index 895d6eeb56..ba9e59a8a3 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/testmisc.cc
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/testmisc.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*misc to run only testmisc.cc tests
 //----------------------------------------------------------------------------
@@ -17,7 +17,7 @@
 #include <sstream>
 #include <typeinfo>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #define TESTID( s ) s##_GPU_MISC
 #else
 #define TESTID( s ) s##_CPU_MISC
@@ -26,7 +26,7 @@
 #define XTESTID( s ) TESTID( s )
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -59,7 +59,7 @@ namespace mg5amcCpu
 
 TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testmisc )
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/testxxx.cc b/epochX/cudacpp/ee_mumu.sa/SubProcesses/testxxx.cc
index 3361fe5aa9..e5167de00c 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/testxxx.cc
+++ b/epochX/cudacpp/ee_mumu.sa/SubProcesses/testxxx.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Apr 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*xxx to run only testxxx.cc tests
 //----------------------------------------------------------------------------
@@ -24,7 +24,7 @@
 #include <iomanip>
 #include <iostream>
 #include <vector>
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #define TESTID( s ) s##_GPU_XXX
 #else
 #define TESTID( s ) s##_CPU_XXX
@@ -32,7 +32,7 @@
 
 #define XTESTID( s ) TESTID( s )
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -42,7 +42,7 @@ namespace mg5amcCpu
   int FPEhandlerIevt = -1;
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU): '" << FPEhandlerMessage << "' ievt=" << FPEhandlerIevt << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU neppV=" << neppV << "): '" << FPEhandlerMessage << "' ievt=" << FPEhandlerIevt << std::endl;
@@ -53,7 +53,7 @@ namespace mg5amcCpu
 
 TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -77,7 +77,7 @@ TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
   assert( nevt % neppM == 0 ); // nevt must be a multiple of neppM
   assert( nevt % neppV == 0 ); // nevt must be a multiple of neppV
   // Fill in the input momenta
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   mg5amcGpu::PinnedHostBufferMomenta hstMomenta( nevt ); // AOSOA[npagM][npar=4][np4=4][neppM]
 #else
   mg5amcCpu::HostBufferMomenta hstMomenta( nevt ); // AOSOA[npagM][npar=4][np4=4][neppM]
@@ -322,7 +322,7 @@ TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
   {
     for( int ievt = 0; ievt < nevt; ievt++ )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       using namespace mg5amcGpu;
 #else
       using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/ee_mumu.sa/src/HelAmps_sm.h b/epochX/cudacpp/ee_mumu.sa/src/HelAmps_sm.h
index 9fa30cfd7f..e878fcd28e 100644
--- a/epochX/cudacpp/ee_mumu.sa/src/HelAmps_sm.h
+++ b/epochX/cudacpp/ee_mumu.sa/src/HelAmps_sm.h
@@ -5,7 +5,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -28,7 +28,7 @@
 //#include <iomanip>
 //#include <iostream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/ee_mumu.sa/src/Parameters_sm.cc b/epochX/cudacpp/ee_mumu.sa/src/Parameters_sm.cc
index 0b4be4d5ed..cffc5d3bff 100644
--- a/epochX/cudacpp/ee_mumu.sa/src/Parameters_sm.cc
+++ b/epochX/cudacpp/ee_mumu.sa/src/Parameters_sm.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -17,7 +17,7 @@
 #include <iomanip>
 #include <iostream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using namespace mg5amcGpu;
 #else
 using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/ee_mumu.sa/src/Parameters_sm.h b/epochX/cudacpp/ee_mumu.sa/src/Parameters_sm.h
index 64d0b8e761..2a6d960581 100644
--- a/epochX/cudacpp/ee_mumu.sa/src/Parameters_sm.h
+++ b/epochX/cudacpp/ee_mumu.sa/src/Parameters_sm.h
@@ -27,7 +27,7 @@
 #include "read_slha.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -93,7 +93,7 @@ namespace mg5amcCpu
 #include <limits>
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -218,7 +218,7 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -235,7 +235,7 @@ namespace mg5amcCpu
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wunused-variable"  // e.g. <<warning: unused variable ‘mdl_G__exp__2’ [-Wunused-variable]>>
 #pragma GCC diagnostic ignored "-Wunused-parameter" // e.g. <<warning: unused parameter ‘G’ [-Wunused-parameter]>>
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #pragma nv_diagnostic push
 #pragma nv_diag_suppress 177 // e.g. <<warning #177-D: variable "mdl_G__exp__2" was declared but never referenced>>
 #endif
@@ -259,7 +259,7 @@ namespace mg5amcCpu
       // End SM implementation - no special handling of vectors of floats as in EFT (#439)
       return out;
     }
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #pragma GCC diagnostic pop
 #pragma nv_diagnostic pop
 #endif
diff --git a/epochX/cudacpp/ee_mumu.sa/src/cudacpp_src.mk b/epochX/cudacpp/ee_mumu.sa/src/cudacpp_src.mk
index d4cc628aec..b2b9da5288 100644
--- a/epochX/cudacpp/ee_mumu.sa/src/cudacpp_src.mk
+++ b/epochX/cudacpp/ee_mumu.sa/src/cudacpp_src.mk
@@ -19,7 +19,7 @@ SHELL := /bin/bash
 #=== Configure common compiler flags for CUDA and C++
 
 INCFLAGS = -I.
-OPTFLAGS = -O3 # this ends up in CUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
+OPTFLAGS = -O3 # this ends up in GPUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
 
 #-------------------------------------------------------------------------------
 
@@ -45,13 +45,13 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure the CUDA compiler (note: NVCC is already exported including ccache)
+#=== Configure the CUDA compiler (note: GPUCC is already exported including ccache)
 
-###$(info NVCC=$(NVCC))
+###$(info GPUCC=$(GPUCC))
 
 #-------------------------------------------------------------------------------
 
-#=== Configure ccache for C++ builds (note: NVCC is already exported including ccache)
+#=== Configure ccache for C++ builds (note: GPUCC is already exported including ccache)
 
 # Enable ccache if USECCACHE=1
 ifeq ($(USECCACHE)$(shell echo $(CXX) | grep ccache),1)
@@ -92,6 +92,13 @@ endif
 ###$(info OMPFLAGS=$(OMPFLAGS))
 CXXFLAGS += $(OMPFLAGS)
 
+# Add correct flags for nvcc (-x cu) and hipcc (-x hip) for GPU code (see #810)
+ifeq ($(findstring nvcc,$(GPUCC)),nvcc)
+  GPUFLAGS += -Xcompiler -fPIC -c -x cu
+else ifeq ($(findstring hipcc,$(GPUCC)),hipcc)
+  GPUFLAGS += -fPIC -c -x hip
+endif
+
 # Set the build flags appropriate to each AVX choice (example: "make AVX=none")
 # [NB MGONGPU_PVW512 is needed because "-mprefer-vector-width=256" is not exposed in a macro]
 # [See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96476]
@@ -253,20 +260,20 @@ $(BUILDDIR)/%.o : %.cc *.h $(BUILDDIR)/.build.$(TAG)
 # Generic target and build rules: objects from CUDA compilation
 $(BUILDDIR)/%_cu.o : %.cc *.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c -x cu $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $< -o $@
 
 #-------------------------------------------------------------------------------
 
 cxx_objects=$(addprefix $(BUILDDIR)/, Parameters_sm.o read_slha.o)
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 cu_objects=$(addprefix $(BUILDDIR)/, Parameters_sm_cu.o)
 endif
 
 # Target (and build rules): common (src) library
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so : $(cxx_objects) $(cu_objects)
 	@if [ ! -d $(LIBDIR) ]; then echo "mkdir -p $(LIBDIR)"; mkdir -p $(LIBDIR); fi
-	$(NVCC) -shared -o $@ $(cxx_objects) $(cu_objects) $(LDFLAGS)
+	$(GPUCC) -shared -o $@ $(cxx_objects) $(cu_objects) $(LDFLAGS)
 else
 $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so : $(cxx_objects)
 	@if [ ! -d $(LIBDIR) ]; then echo "mkdir -p $(LIBDIR)"; mkdir -p $(LIBDIR); fi
diff --git a/epochX/cudacpp/ee_mumu.sa/src/mgOnGpuConfig.h b/epochX/cudacpp/ee_mumu.sa/src/mgOnGpuConfig.h
index b247654dcf..475749ca7c 100644
--- a/epochX/cudacpp/ee_mumu.sa/src/mgOnGpuConfig.h
+++ b/epochX/cudacpp/ee_mumu.sa/src/mgOnGpuConfig.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jul 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUCONFIG_H
 #define MGONGPUCONFIG_H 1
@@ -10,12 +10,26 @@
 // There are two different code bases for standalone_cudacpp (without multichannel) and madevent+cudacpp (with multichannel)
 #undef MGONGPU_SUPPORTS_MULTICHANNEL
 
+// Is this a GPU (CUDA, HIP) or CPU implementation?
+#ifdef __CUDACC__
+#define MGONGPUCPP_GPUIMPL cuda
+#elif defined __HIPCC__
+#define MGONGPUCPP_GPUIMPL hip
+#include "hip/hip_runtime.h" // needed for blockDim, blockIdx, threadIdx: better in mgOnGpuConfig.h than in GpuAbstraction.h
+#else
+#undef MGONGPUCPP_GPUIMPL
+#endif
+
 // ** NB1 Throughputs (e.g. 6.8E8) are events/sec for "./gcheck.exe -p 65536 128 12"
 // ** NB2 Baseline on b7g47n0004 fluctuates (probably depends on load on other VMs)
 
 // Choose if curand is supported for generating random numbers
+// For HIP, by default, do not use curand (common random numbers will be used instead)
 // For both CUDA and C++, by default, do not inline, but allow this macro to be set from outside with e.g. -DMGONGPU_HAS_NO_CURAND
-// (there exist CUDA installations, e.g. using the HPC package, which do not include curand - see PR #784)
+// (there exist CUDA installations, e.g. using the HPC package, which do not include curand - see PR #784 and #785)
+#if defined __HIPCC__
+#define MGONGPU_HAS_NO_CURAND 1
+#else
 //#ifdef __CUDACC__
 //#undef MGONGPU_HAS_NO_CURAND // default
 ////#define MGONGPU_HAS_NO_CURAND 1
@@ -23,6 +37,7 @@
 //#undef MGONGPU_HAS_NO_CURAND // default
 ////#define MGONGPU_HAS_NO_CURAND 1
 //#endif
+#endif
 
 // Choose floating point precision (for everything but color algebra #537)
 // If one of these macros has been set from outside with e.g. -DMGONGPU_FPTYPE_FLOAT, nothing happens (issue #167)
@@ -54,23 +69,31 @@
 //#undef MGONGPU_HARDCODE_PARAM // default
 ////#define MGONGPU_HARDCODE_PARAM 1
 
-// Complex type in c++: std::complex or cxsmpl (CHOOSE ONLY ONE)
-#ifndef __CUDACC__
-//#define MGONGPU_CPPCXTYPE_STDCOMPLEX 1 // ~8 percent slower on float, same on double (5.1E6/double, 9.4E6/float)
-#define MGONGPU_CPPCXTYPE_CXSMPL 1 // new default (5.1E6/double, 10.2E6/float)
-#endif
-
-// Complex type in cuda: thrust or cucomplex or cxsmpl (CHOOSE ONLY ONE)
+// Complex type in CUDA: thrust or cucomplex or cxsmpl (CHOOSE ONLY ONE)
+// (NB THIS IS MGONGPU_*CU*CXTYPE_xxx)
 #ifdef __CUDACC__
 #define MGONGPU_CUCXTYPE_THRUST 1 // default (~1.15E9/double, ~3.2E9/float)
 //#define MGONGPU_CUCXTYPE_CUCOMPLEX 1 // ~10 percent slower (1.03E9/double, ~2.8E9/float)
 //#define MGONGPU_CUCXTYPE_CXSMPL 1 // ~10 percent slower (1.00E9/double, ~2.9E9/float)
+
+// Complex type in HIP: cxsmpl (ONLY ONE OPTION POSSIBLE? #810)
+// (NB THIS IS MGONGPU_*HIP*CXTYPE_xxx)
+#elif defined __HIPCC__
+#define MGONGPU_HIPCXTYPE_CXSMPL 1 // default for HIP
+
+// Complex type in C++: std::complex or cxsmpl (CHOOSE ONLY ONE)
+// (NB THIS IS MGONGPU_*CPP*CXTYPE_xxx)
+#else
+//#define MGONGPU_CPPCXTYPE_STDCOMPLEX 1 // ~8 percent slower on float, same on double (5.1E6/double, 9.4E6/float)
+#define MGONGPU_CPPCXTYPE_CXSMPL 1 // new default (5.1E6/double, 10.2E6/float)
 #endif
 
-// Cuda nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
+// CUDA nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
 #ifdef __CUDACC__
-#undef MGONGPU_NSIGHT_DEBUG // default
+#undef MGONGPU_NSIGHT_DEBUG // default in CUDA
 //#define MGONGPU_NSIGHT_DEBUG 1
+#else
+#undef MGONGPU_NSIGHT_DEBUG // only option in HIP or C++
 #endif
 
 // SANITY CHECKS (floating point precision for everything but color algebra #537)
@@ -86,17 +109,21 @@
 #error You cannot use double precision for color algebra and single precision elsewhere
 #endif
 
-// SANITY CHECKS (c++ complex number implementation)
-#ifndef __CUDACC__
-#if defined MGONGPU_CPPCXTYPE_STDCOMPLEX and defined MGONGPU_CPPCXTYPE_CXSMPL
-#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CPPCXTYPE_STDCOMPLEX or MGONGPU_CPPCXTYPE_CXSMPL
+// SANITY CHECKS (CUDA complex number implementation)
+#ifdef __CUDACC__
+#if defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CUCOMPLEX
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX for CUDA
+#elif defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CXSMPL for CUDA
+#elif defined MGONGPU_CUCXTYPE_CUCOMPLEX and defined MGONGPU_CUCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE OF MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL for CUDA
 #endif
 #endif
 
-// SANITY CHECKS (cuda complex number implementation)
-#ifdef __CUDACC__
-#if defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CUCOMPLEX and defined MGONGPU_CUCXTYPE_CXSMPL
-#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL
+// SANITY CHECKS (C++ complex number implementation)
+#ifndef MGONGPUCPP_GPUIMPL
+#if defined MGONGPU_CPPCXTYPE_STDCOMPLEX and defined MGONGPU_CPPCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CPPCXTYPE_STDCOMPLEX or MGONGPU_CPPCXTYPE_CXSMPL for C++
 #endif
 #endif
 
@@ -134,7 +161,7 @@ namespace mgOnGpu
   // Alignment requirement for using reinterpret_cast with SIMD vectorized code
   // (using reinterpret_cast with non aligned memory may lead to segmentation faults!)
   // Only needed for C++ code but can be enforced also in NVCC builds of C++ code using CUDA>=11.2 and C++17 (#318, #319, #333)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   constexpr int cppAlign = 64; // alignment requirement for SIMD vectorization (64-byte i.e. 512-bit)
 #endif
 
@@ -145,7 +172,7 @@ using mgOnGpu::fptype;
 using mgOnGpu::fptype2;
 
 // C++ SIMD vectorization width (this will be used to set neppV)
-#ifdef __CUDACC__ // CUDA implementation has no SIMD
+#ifdef MGONGPUCPP_GPUIMPL // CUDA and HIP implementations have no SIMD
 #undef MGONGPU_CPPSIMD
 #elif defined __AVX512VL__ && defined MGONGPU_PVW512 // C++ "512z" AVX512 with 512 width (512-bit ie 64-byte): 8 (DOUBLE) or 16 (FLOAT)
 #ifdef MGONGPU_FPTYPE_DOUBLE
@@ -175,9 +202,9 @@ using mgOnGpu::fptype2;
 #undef MGONGPU_CPPSIMD
 #endif
 
-// Cuda nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
+// CUDA nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
 // Arguments (not used so far): text is __FUNCTION__, code is 0 (start) or 1 (end)
-#if defined __CUDACC__ && defined MGONGPU_NSIGHT_DEBUG /* clang-format off */
+#if defined __CUDA__ && defined MGONGPU_NSIGHT_DEBUG /* clang-format off */
 #define mgDebugDeclare() __shared__ float mgDebugCounter[mgOnGpu::ntpbMAX];
 #define mgDebugInitialise() { mgDebugCounter[threadIdx.x] = 0; }
 #define mgDebug( code, text ) { mgDebugCounter[threadIdx.x] += 1; }
@@ -189,8 +216,8 @@ using mgOnGpu::fptype2;
 #define mgDebugFinalise() { /*noop*/ }
 #endif /* clang-format on */
 
-// Define empty CUDA declaration specifiers for C++
-#ifndef __CUDACC__
+// Define empty CUDA/HIP declaration specifiers for C++
+#ifndef MGONGPUCPP_GPUIMPL
 #define __global__
 #define __host__
 #define __device__
diff --git a/epochX/cudacpp/ee_mumu.sa/src/mgOnGpuCxtypes.h b/epochX/cudacpp/ee_mumu.sa/src/mgOnGpuCxtypes.h
index ca9a9f00c0..7ede1dbfae 100644
--- a/epochX/cudacpp/ee_mumu.sa/src/mgOnGpuCxtypes.h
+++ b/epochX/cudacpp/ee_mumu.sa/src/mgOnGpuCxtypes.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022, based on earlier work by D. Smith) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUCXTYPES_H
 #define MGONGPUCXTYPES_H 1
@@ -19,7 +19,7 @@
 #include <complex>
 
 // Complex type in cuda: thrust or cucomplex or cxsmpl
-#ifdef __CUDACC__
+#ifdef __CUDACC__ // this must be __CUDAC__ (not MGONGPUCPP_GPUIMPL)
 #if defined MGONGPU_CUCXTYPE_THRUST
 #pragma clang diagnostic push
 #pragma clang diagnostic ignored "-Wtautological-compare" // for icpx2021/clang13 (https://stackoverflow.com/a/15864661)
@@ -30,8 +30,13 @@
 #elif not defined MGONGPU_CUCXTYPE_CXSMPL
 #error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL
 #endif
+// Complex type in HIP: cxsmpl
+#elif defined __HIPCC__
+#if not defined MGONGPU_HIPCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_HIPCXTYPE_CXSMPL
+#endif
 #else
-// Complex type in c++: std::complex or cxsmpl
+// Complex type in c++ or HIP: std::complex or cxsmpl
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
 #include <cmath>
 #elif not defined MGONGPU_CPPCXTYPE_CXSMPL
@@ -82,7 +87,7 @@ namespace mgOnGpu /* clang-format off */
 using mgOnGpu::cxsmpl;
 
 // Printout to stream for user defined types
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -92,7 +97,7 @@ namespace mg5amcCpu
   inline __host__ std::ostream&
   operator<<( std::ostream& out, const cxsmpl<FP>& c )
   {
-    out << std::complex( c.real(), c.imag() );
+    out << std::complex<FP>( c.real(), c.imag() );
     return out;
   }
 
@@ -215,14 +220,14 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
   // --- Type definitions (complex type: cxtype)
-#ifdef __CUDACC__ // cuda
+#ifdef __CUDACC__ // this must be __CUDAC__ (not MGONGPUCPP_GPUIMPL)
 #if defined MGONGPU_CUCXTYPE_THRUST
   typedef thrust::complex<fptype> cxtype;
 #elif defined MGONGPU_CUCXTYPE_CUCOMPLEX
@@ -255,13 +260,13 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
-#if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+#if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
 
   //------------------------------
   // CUDA or C++ - using cxsmpl
@@ -303,11 +308,11 @@ namespace mg5amcCpu
     return cxmake( c.real(), c.imag() );
   }
 
-#endif // #if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+#endif // #if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
 
   //==========================================================================
 
-#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_THRUST // cuda + thrust
+#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_THRUST // cuda + thrust (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // CUDA - using thrust::complex
@@ -347,7 +352,7 @@ namespace mg5amcCpu
 
   //==========================================================================
 
-#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_CUCOMPLEX // cuda + cucomplex
+#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_CUCOMPLEX // cuda + cucomplex (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // CUDA - using cuComplex
@@ -566,7 +571,7 @@ namespace mg5amcCpu
 
   //==========================================================================
 
-#if not defined __CUDACC__ and defined MGONGPU_CPPCXTYPE_STDCOMPLEX // c++ + stdcomplex
+#if not defined __CUDACC__ and defined MGONGPU_CPPCXTYPE_STDCOMPLEX // c++/hip + stdcomplex (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // C++ - using std::complex
@@ -633,7 +638,7 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/ee_mumu.sa/src/mgOnGpuFptypes.h b/epochX/cudacpp/ee_mumu.sa/src/mgOnGpuFptypes.h
index 905c97d700..fa3a02664b 100644
--- a/epochX/cudacpp/ee_mumu.sa/src/mgOnGpuFptypes.h
+++ b/epochX/cudacpp/ee_mumu.sa/src/mgOnGpuFptypes.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUFPTYPES_H
 #define MGONGPUFPTYPES_H 1
@@ -12,7 +12,7 @@
 #include <cmath>
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL // cuda
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -20,7 +20,7 @@ namespace mg5amcCpu
 {
   //==========================================================================
 
-#ifdef __CUDACC__ // cuda
+#ifdef MGONGPUCPP_GPUIMPL // cuda
 
   //------------------------------
   // Floating point types - Cuda
@@ -64,11 +64,11 @@ namespace mg5amcCpu
 #endif
   }
 
-#endif // #ifdef __CUDACC__
+#endif // #ifdef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
   //------------------------------
   // Floating point types - C++
@@ -92,7 +92,7 @@ namespace mg5amcCpu
     return std::sqrt( f );
   }
 
-#endif // #ifndef __CUDACC__
+#endif // #ifndef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
diff --git a/epochX/cudacpp/ee_mumu.sa/src/mgOnGpuVectors.h b/epochX/cudacpp/ee_mumu.sa/src/mgOnGpuVectors.h
index e1299ba81e..cdae04326b 100644
--- a/epochX/cudacpp/ee_mumu.sa/src/mgOnGpuVectors.h
+++ b/epochX/cudacpp/ee_mumu.sa/src/mgOnGpuVectors.h
@@ -32,7 +32,7 @@
 #endif
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -131,7 +131,7 @@ namespace mg5amcCpu
 #endif
 #endif
 
-#else // i.e #ifndef MGONGPU_CPPSIMD (this includes #ifdef __CUDACC__)
+#else // i.e #ifndef MGONGPU_CPPSIMD (this includes #ifdef MGONGPUCPP_GPUIMPL)
 
   const int neppV = 1;
 
@@ -153,13 +153,13 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
   // Printout to stream for user defined types
 
@@ -805,11 +805,11 @@ namespace mg5amcCpu
 
 #endif // #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
 
-#endif // #ifndef __CUDACC__
+#endif // #ifndef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
   //------------------------------
   // Vector types - CUDA
@@ -853,12 +853,12 @@ namespace mg5amcCpu
     return mask;
   }
 
-#endif // #ifdef __CUDACC__
+#endif // #ifdef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
   // Scalar-or-vector types: scalar in CUDA, vector or scalar in C++
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   typedef bool bool_sv;
   typedef fptype fptype_sv;
   typedef fptype2 fptype2_sv;
@@ -879,7 +879,7 @@ namespace mg5amcCpu
 #endif
 
   // Scalar-or-vector zeros: scalar in CUDA, vector or scalar in C++
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   inline __host__ __device__ cxtype cxzero_sv(){ return cxtype( 0, 0 ); }
 #elif defined MGONGPU_CPPSIMD
   inline cxtype_v cxzero_sv() { return cxtype_v(); } // RRRR=0000 IIII=0000
diff --git a/epochX/cudacpp/ee_mumu.sa/src/rambo.h b/epochX/cudacpp/ee_mumu.sa/src/rambo.h
index e02ea52496..cd7e1008ea 100644
--- a/epochX/cudacpp/ee_mumu.sa/src/rambo.h
+++ b/epochX/cudacpp/ee_mumu.sa/src/rambo.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -18,7 +18,7 @@
 #include <iostream>
 
 // Simplified rambo version for 2 to N (with N>=2) processes with massless particles
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +83,7 @@ namespace mg5amcCpu
       static bool first = true;
       if( first )
       {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
         if constexpr( M_ACCESS::isOnDevice() ) // avoid
         {
           const int ievt0 = 0;
@@ -166,7 +166,7 @@ namespace mg5amcCpu
     wt = po2log;
     if( nparf != 2 ) wt = ( 2. * nparf - 4. ) * log( energy ) + z[nparf - 1];
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     // issue warnings if weight is too small or too large
     static int iwarn[5] = { 0, 0, 0, 0, 0 };
     if( wt < -180. )
diff --git a/epochX/cudacpp/ee_mumu.sa/src/read_slha.cc b/epochX/cudacpp/ee_mumu.sa/src/read_slha.cc
index 055b19a779..f8e46f2e66 100644
--- a/epochX/cudacpp/ee_mumu.sa/src/read_slha.cc
+++ b/epochX/cudacpp/ee_mumu.sa/src/read_slha.cc
@@ -11,7 +11,11 @@
 
 #include <algorithm>
 #include <cstdlib>
-#include <filesystem>
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
 #include <fstream>
 #include <iostream>
 
@@ -60,7 +64,14 @@ SLHAReader::read_slha_file( std::string file_name, bool verbose )
     {
       std::cout << "WARNING! Card file '" << file_name << "' does not exist:"
                 << " look for the file in directory $" << envpath << "='" << getenv( envpath ) << "'" << std::endl;
+      /*
+#ifdef __HIPCC__
+      const std::string file_name2 = std::experimental::filesystem::path( getenv( envpath ) ) / std::experimental::filesystem::path( file_name ).filename();
+#else
       const std::string file_name2 = std::filesystem::path( getenv( envpath ) ) / std::filesystem::path( file_name ).filename();
+#endif
+      */
+      const std::string file_name2 = std::string( getenv( envpath ) ) + "/" + file_name; // bypass std::filesystem #803
       param_card.open( file_name2.c_str(), std::ifstream::in );
       if( param_card.good() )
       {
diff --git a/epochX/cudacpp/gg_tt.mad/CODEGEN_mad_gg_tt_log.txt b/epochX/cudacpp/gg_tt.mad/CODEGEN_mad_gg_tt_log.txt
index a477013568..5782086b56 100644
--- a/epochX/cudacpp/gg_tt.mad/CODEGEN_mad_gg_tt_log.txt
+++ b/epochX/cudacpp/gg_tt.mad/CODEGEN_mad_gg_tt_log.txt
@@ -52,7 +52,7 @@ Note that you can still compile and run aMC@NLO with the built-in PDFs
 
 Using default text editor "vi". Set another one in ./input/mg5_configuration.txt
 Using default eps viewer "evince". Set another one in ./input/mg5_configuration.txt
-No valid web browser found. Please set in ./input/mg5_configuration.txt
+Using default web browser "firefox". Set another one in ./input/mg5_configuration.txt
 import /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_mad_gg_tt.mg
 The import format was not given, so we guess it as command
 set stdout_level DEBUG
@@ -62,7 +62,7 @@ generate g g > t t~
 No model currently active, so we import the Standard Model
 INFO: load particles 
 INFO: load vertices 
-DEBUG: model prefixing  takes 0.005816459655761719 
+DEBUG: model prefixing  takes 0.005559206008911133 
 INFO: Restrict model sm with file models/sm/restrict_default.dat . 
 DEBUG: Simplifying conditional expressions 
 DEBUG: remove interactions: u s w+ at order: QED=1 
@@ -175,7 +175,7 @@ INFO: Generating Helas calls for process: g g > t t~ WEIGHTED<=2 @1
 INFO: Processing color information for process: g g > t t~ @1 
 INFO: Creating files in directory P1_gg_ttx 
 DEBUG:  kwargs[prefix] = 0 [model_handling.py at line 1058] 
-DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7f089da5e700> [export_v4.py at line 6262] 
+DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7f0131f456a0> [export_v4.py at line 6262] 
 INFO: Creating files in directory . 
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.h
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.cc
@@ -191,16 +191,16 @@ INFO: Created files CPPProcess.h and CPPProcess.cc in directory ./.
 INFO: Generating Feynman diagrams for Process: g g > t t~ WEIGHTED<=2 @1 
 INFO: Finding symmetric diagrams for subprocess group gg_ttx 
 Generated helas calls for 1 subprocesses (3 diagrams) in 0.006 s
-Wrote files for 10 helas calls in 0.103 s
+Wrote files for 10 helas calls in 0.100 s
 ALOHA: aloha starts to compute helicity amplitudes
 ALOHA: aloha creates VVV1 set of routines with options: P0
 ALOHA: aloha creates FFV1 routines
-ALOHA: aloha creates 2 routines in  0.155 s
+ALOHA: aloha creates 2 routines in  0.146 s
 DEBUG:  Entering PLUGIN_ProcessExporter.convert_model (create the model) [output.py at line 202] 
 ALOHA: aloha starts to compute helicity amplitudes
 ALOHA: aloha creates VVV1 set of routines with options: P0
 ALOHA: aloha creates FFV1 routines
-ALOHA: aloha creates 4 routines in  0.135 s
+ALOHA: aloha creates 4 routines in  0.132 s
 <class 'aloha.create_aloha.AbstractRoutine'> VVV1
 <class 'aloha.create_aloha.AbstractRoutine'> FFV1
 <class 'aloha.create_aloha.AbstractRoutine'> FFV1
@@ -237,9 +237,9 @@ Type "launch" to generate events from this process, or see
 Run "open index.html" to see more information about this process.
 quit
 
-real	0m1.729s
-user	0m1.515s
-sys	0m0.204s
+real	0m1.699s
+user	0m1.467s
+sys	0m0.225s
 Code generation completed in 2 seconds
 ************************************************************
 *                                                          *
@@ -266,7 +266,7 @@ INFO: load configuration from /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/mg5amc
 INFO: load configuration from /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_mad_gg_tt/Cards/me5_configuration.txt  
 Using default text editor "vi". Set another one in ./input/mg5_configuration.txt
 Using default eps viewer "evince". Set another one in ./input/mg5_configuration.txt
-No valid web browser found. Please set in ./input/mg5_configuration.txt
+Using default web browser "firefox". Set another one in ./input/mg5_configuration.txt
 treatcards run
 quit
 INFO:  
@@ -296,7 +296,7 @@ INFO: load configuration from /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/mg5amc
 INFO: load configuration from /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_mad_gg_tt/Cards/me5_configuration.txt  
 Using default text editor "vi". Set another one in ./input/mg5_configuration.txt
 Using default eps viewer "evince". Set another one in ./input/mg5_configuration.txt
-No valid web browser found. Please set in ./input/mg5_configuration.txt
+Using default web browser "firefox". Set another one in ./input/mg5_configuration.txt
 treatcards param
 quit
 INFO:  
diff --git a/epochX/cudacpp/gg_tt.mad/COPYRIGHT b/epochX/cudacpp/gg_tt.mad/COPYRIGHT
index a134b5fef9..9036d9260a 100644
--- a/epochX/cudacpp/gg_tt.mad/COPYRIGHT
+++ b/epochX/cudacpp/gg_tt.mad/COPYRIGHT
@@ -15,6 +15,7 @@ The full development team currently includes the following authors :
   Stephan Hageboeck (CERN)
   Olivier Mattelaer (Universite Catholique de Louvain, original author)
   Stefan Roiser (CERN, original author)
+  Jorgen Teig (CERN)
   Andrea Valassi (CERN, original author)
   Zenny Wettersten (CERN)
 See https://github.com/madgraph5/madgraph4gpu for more details. For the full
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/Bridge.h b/epochX/cudacpp/gg_tt.mad/SubProcesses/Bridge.h
index bf8b5e024d..f9ed70dfde 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/Bridge.h
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/Bridge.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Roiser (Nov 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef BRIDGE_H
 #define BRIDGE_H 1
@@ -14,16 +14,23 @@
 #include "MemoryAccessMomenta.h"  // for MemoryAccessMomenta::neppM
 #include "MemoryBuffers.h"        // for HostBufferMomenta, DeviceBufferMomenta etc
 
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
+
+#include <sys/stat.h> // bypass std::filesystem #803
+
 #include <algorithm>
 #include <cassert>
 #include <cmath>
 #include <cstring>
-#include <filesystem>
 #include <iostream>
 #include <memory>
 #include <type_traits>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +90,7 @@ namespace mg5amcCpu
     Bridge& operator=( const Bridge& ) = delete;
     Bridge& operator=( Bridge&& ) = delete;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     /**
      * Set the gpublocks and gputhreads for the gpusequence - throws if evnt != gpublocks*gputhreads
      * (this is needed for BridgeKernel tests rather than for actual production use in Fortran)
@@ -150,7 +157,7 @@ namespace mg5amcCpu
     unsigned int m_nevt; // number of events
     int m_nGoodHel;      // the number of good helicities (-1 initially when they have not yet been calculated)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     int m_gputhreads; // number of gpu threads (default set from number of events, can be modified)
     int m_gpublocks;  // number of gpu blocks (default set from number of events, can be modified)
     DeviceBuffer<FORTRANFPTYPE, sizePerEventMomenta> m_devMomentaF;
@@ -187,12 +194,12 @@ namespace mg5amcCpu
   // Forward declare transposition methods
   //
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
   template<typename Tin, typename Tout>
   __global__ void dev_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt );
 
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
 
   template<typename Tin, typename Tout>
   void hst_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt );
@@ -209,7 +216,7 @@ namespace mg5amcCpu
   Bridge<FORTRANFPTYPE>::Bridge( unsigned int nevtF, unsigned int nparF, unsigned int np4F )
     : m_nevt( nevtF )
     , m_nGoodHel( -1 )
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     , m_gputhreads( 256 )                  // default number of gpu threads
     , m_gpublocks( m_nevt / m_gputhreads ) // this ensures m_nevt <= m_gpublocks*m_gputhreads
     , m_devMomentaF( m_nevt )
@@ -233,7 +240,7 @@ namespace mg5amcCpu
   {
     if( nparF != CPPProcess::npar ) throw std::runtime_error( "Bridge constructor: npar mismatch" );
     if( np4F != CPPProcess::np4 ) throw std::runtime_error( "Bridge constructor: np4 mismatch" );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( ( m_nevt < s_gputhreadsmin ) || ( m_nevt % s_gputhreadsmin != 0 ) )
       throw std::runtime_error( "Bridge constructor: nevt should be a multiple of " + std::to_string( s_gputhreadsmin ) );
     while( m_nevt != m_gpublocks * m_gputhreads )
@@ -249,20 +256,28 @@ namespace mg5amcCpu
 #else
     std::cout << "WARNING! Instantiate host Bridge (nevt=" << m_nevt << ")" << std::endl;
     m_pmek.reset( new MatrixElementKernelHost( m_hstMomentaC, m_hstGs, m_hstRndHel, m_hstRndCol, m_hstMEs, m_hstSelHel, m_hstSelCol, m_nevt ) );
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
     // Create a process object, read param card and set parameters
     // FIXME: the process instance can happily go out of scope because it is only needed to read parameters?
     // FIXME: the CPPProcess should really be a singleton? what if fbridgecreate is called from several Fortran threads?
     CPPProcess process( /*verbose=*/false );
     std::string paramCard = "../../Cards/param_card.dat";
-    if( !std::filesystem::exists( paramCard ) )
-    {
-      paramCard = "../" + paramCard;
-    }
+    /*
+#ifdef __HIPCC__
+    if( !std::experimental::filesystem::exists( paramCard ) ) paramCard = "../" + paramCard;
+#else
+    if( !std::filesystem::exists( paramCard ) ) paramCard = "../" + paramCard;
+#endif
+    */
+    //struct stat dummybuffer; // bypass std::filesystem #803
+    //if( !( stat( paramCard.c_str(), &dummyBuffer ) == 0 ) ) paramCard = "../" + paramCard; //
+    auto fileExists = []( std::string& fileName )
+    { struct stat buffer; return stat( fileName.c_str(), &buffer ) == 0; };
+    if( !fileExists( paramCard ) ) paramCard = "../" + paramCard; // bypass std::filesystem #803
     process.initProc( paramCard );
   }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::set_gpugrid( const int gpublocks, const int gputhreads )
   {
@@ -276,7 +291,7 @@ namespace mg5amcCpu
   }
 #endif
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::gpu_sequence( const FORTRANFPTYPE* momenta,
                                             const FORTRANFPTYPE* gs,
@@ -291,14 +306,14 @@ namespace mg5amcCpu
     constexpr int neppM = MemoryAccessMomenta::neppM;
     if constexpr( neppM == 1 && std::is_same_v<FORTRANFPTYPE, fptype> )
     {
-      checkCuda( cudaMemcpy( m_devMomentaC.data(), momenta, m_devMomentaC.bytes(), cudaMemcpyHostToDevice ) );
+      gpuMemcpy( m_devMomentaC.data(), momenta, m_devMomentaC.bytes(), gpuMemcpyHostToDevice );
     }
     else
     {
-      checkCuda( cudaMemcpy( m_devMomentaF.data(), momenta, m_devMomentaF.bytes(), cudaMemcpyHostToDevice ) );
+      gpuMemcpy( m_devMomentaF.data(), momenta, m_devMomentaF.bytes(), gpuMemcpyHostToDevice );
       const int thrPerEvt = CPPProcess::npar * CPPProcess::np4; // AV: transpose alg does 1 element per thread (NOT 1 event per thread)
       //const int thrPerEvt = 1; // AV: try new alg with 1 event per thread... this seems slower
-      dev_transposeMomentaF2C<<<m_gpublocks * thrPerEvt, m_gputhreads>>>( m_devMomentaF.data(), m_devMomentaC.data(), m_nevt );
+      gpuLaunchKernel( dev_transposeMomentaF2C, m_gpublocks * thrPerEvt, m_gputhreads, m_devMomentaF.data(), m_devMomentaC.data(), m_nevt );
     }
     if constexpr( std::is_same_v<FORTRANFPTYPE, fptype> )
     {
@@ -341,7 +356,7 @@ namespace mg5amcCpu
   }
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::cpu_sequence( const FORTRANFPTYPE* momenta,
                                             const FORTRANFPTYPE* gs,
@@ -396,7 +411,7 @@ namespace mg5amcCpu
   // - C++ array: momenta[npagM][npar][np4][neppM] with nevt=npagM*neppM (AOSOA)
   //
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename Tin, typename Tout>
   __global__ void dev_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt )
   {
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/BridgeKernels.cc b/epochX/cudacpp/gg_tt.mad/SubProcesses/BridgeKernels.cc
index d58066c9c1..eaf4037a24 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/BridgeKernels.cc
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/BridgeKernels.cc
@@ -1,17 +1,18 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "BridgeKernels.h"
 
+#include "GpuAbstraction.h"
 #include "MemoryAccessMomenta.h"
 
 #include <sstream>
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -45,7 +46,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu
 {
 
@@ -96,7 +97,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/BridgeKernels.h b/epochX/cudacpp/gg_tt.mad/SubProcesses/BridgeKernels.h
index 15eb4bff4d..3efef8ce97 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/BridgeKernels.h
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/BridgeKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef BRIDGEKERNELS_H
 #define BRIDGEKERNELS_H 1
@@ -12,7 +12,7 @@
 #include "MatrixElementKernels.h"
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -49,7 +49,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A Bridge wrapper class encapsulating matrix element calculations on a CPU host
   class BridgeKernelHost final : public BridgeKernelBase
   {
@@ -89,7 +89,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A Bridge wrapper class encapsulating matrix element calculations on a GPU device
   class BridgeKernelDevice : public BridgeKernelBase
   {
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/CommonRandomNumberKernel.cc b/epochX/cudacpp/gg_tt.mad/SubProcesses/CommonRandomNumberKernel.cc
index 985b39f576..010bc4cbd0 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/CommonRandomNumberKernel.cc
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/CommonRandomNumberKernel.cc
@@ -1,15 +1,16 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "CommonRandomNumbers.h"
+#include "GpuAbstraction.h"
 #include "MemoryBuffers.h"
 #include "RandomNumberKernels.h"
 
 #include <cassert>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/CrossSectionKernels.cc b/epochX/cudacpp/gg_tt.mad/SubProcesses/CrossSectionKernels.cc
index 0b355a3c8d..c15b39844d 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/CrossSectionKernels.cc
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/CrossSectionKernels.cc
@@ -1,10 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "CrossSectionKernels.h"
 
+#include "GpuAbstraction.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessWeights.h"
 #include "MemoryBuffers.h"
@@ -77,7 +78,7 @@ debug_me_is_abnormal( const fptype& me, size_t ievtALL )
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -185,7 +186,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/CrossSectionKernels.h b/epochX/cudacpp/gg_tt.mad/SubProcesses/CrossSectionKernels.h
index 7933ca4bbf..4d9659e04e 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/CrossSectionKernels.h
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/CrossSectionKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef CROSSSECTIONKERNELS_H
 #define CROSSSECTIONKERNELS_H 1
@@ -13,7 +13,7 @@
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -96,7 +96,7 @@ namespace mg5amcCpu
   //--------------------------------------------------------------------------
 
   /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating the calculation of event statistics on a GPU device
   class CrossSectionKernelDevice : public CrossSectionKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/CurandRandomNumberKernel.cc b/epochX/cudacpp/gg_tt.mad/SubProcesses/CurandRandomNumberKernel.cc
index eb56333b03..08a16f6f2c 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/CurandRandomNumberKernel.cc
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/CurandRandomNumberKernel.cc
@@ -1,9 +1,9 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "MemoryBuffers.h"
 #include "RandomNumberKernels.h"
 
@@ -22,7 +22,7 @@ inline void assertCurand( curandStatus_t code, const char *file, int line, bool
 }
 #endif /* clang-format on */
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -36,7 +36,7 @@ namespace mg5amcCpu
   {
     if( m_isOnDevice )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if( !m_rnarray.isOnDevice() )
         throw std::runtime_error( "CurandRandomNumberKernel on device with a host random number array" );
 #else
@@ -114,7 +114,7 @@ namespace mg5amcCpu
     /*
     printf( "\nCurandRandomNumberKernel::generateRnarray size = %d\n", (int)m_rnarray.size() );
     fptype* data = m_rnarray.data();
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( m_rnarray.isOnDevice() )
     {
       data = new fptype[m_rnarray.size()]();
@@ -123,7 +123,7 @@ namespace mg5amcCpu
 #endif
     for( int i = 0; i < ( (int)m_rnarray.size() / 4 ); i++ )
       printf( "[%4d] %f %f %f %f\n", i * 4, data[i * 4], data[i * 4 + 2], data[i * 4 + 2], data[i * 4 + 3] );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( m_rnarray.isOnDevice() ) delete[] data;
 #endif
     */
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/EventStatistics.h b/epochX/cudacpp/gg_tt.mad/SubProcesses/EventStatistics.h
index 48b51e0a49..b425a5bade 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/EventStatistics.h
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/EventStatistics.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef EventStatistics_H
 #define EventStatistics_H 1
@@ -16,7 +16,7 @@
 #include <limits>
 #include <string>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/GpuAbstraction.h b/epochX/cudacpp/gg_tt.mad/SubProcesses/GpuAbstraction.h
new file mode 100644
index 0000000000..9c467b1e04
--- /dev/null
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/GpuAbstraction.h
@@ -0,0 +1,69 @@
+// Copyright (C) 2020-2023 CERN and UCLouvain.
+// Licensed under the GNU Lesser General Public License (version 3 or later).
+// Created by: J. Teig (Jul 2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+
+#ifndef MG5AMC_GPUABSTRACTION_H
+#define MG5AMC_GPUABSTRACTION_H 1
+
+#include <cassert>
+
+//--------------------------------------------------------------------------
+
+#ifdef __CUDACC__
+
+#define gpuError_t cudaError_t
+#define gpuPeekAtLastError cudaPeekAtLastError
+#define gpuGetErrorString cudaGetErrorString
+#define gpuSuccess cudaSuccess
+
+#define gpuMallocHost( ptr, size ) checkGpu( cudaMallocHost( ptr, size ) )
+#define gpuMalloc( ptr, size ) checkGpu( cudaMalloc( ptr, size ) )
+
+#define gpuMemcpy( dstData, srcData, srcBytes, func ) checkGpu( cudaMemcpy( dstData, srcData, srcBytes, func ) )
+#define gpuMemcpyHostToDevice cudaMemcpyHostToDevice
+#define gpuMemcpyDeviceToHost cudaMemcpyDeviceToHost
+#define gpuMemcpyToSymbol( type1, type2, size ) checkGpu( cudaMemcpyToSymbol( type1, type2, size ) )
+
+#define gpuFree( ptr ) checkGpu( cudaFree( ptr ) )
+#define gpuFreeHost( ptr ) checkGpu( cudaFreeHost( ptr ) )
+
+#define gpuSetDevice cudaSetDevice
+#define gpuDeviceSynchronize cudaDeviceSynchronize
+#define gpuDeviceReset cudaDeviceReset
+
+#define gpuLaunchKernel( kernel, blocks, threads, ... ) kernel<<<blocks, threads>>>( __VA_ARGS__ )
+#define gpuLaunchKernelSharedMem( kernel, blocks, threads, sharedMem, ... ) kernel<<<blocks, threads, sharedMem>>>( __VA_ARGS__ )
+
+//--------------------------------------------------------------------------
+
+#elif defined __HIPCC__
+
+#define gpuError_t hipError_t
+#define gpuPeekAtLastError hipPeekAtLastError
+#define gpuGetErrorString hipGetErrorString
+#define gpuSuccess hipSuccess
+
+#define gpuMallocHost( ptr, size ) checkGpu( hipHostMalloc( ptr, size ) ) // HostMalloc better
+#define gpuMalloc( ptr, size ) checkGpu( hipMalloc( ptr, size ) )
+
+#define gpuMemcpy( dstData, srcData, srcBytes, func ) checkGpu( hipMemcpy( dstData, srcData, srcBytes, func ) )
+#define gpuMemcpyHostToDevice hipMemcpyHostToDevice
+#define gpuMemcpyDeviceToHost hipMemcpyDeviceToHost
+#define gpuMemcpyToSymbol( type1, type2, size ) checkGpu( hipMemcpyToSymbol( type1, type2, size ) )
+
+#define gpuFree( ptr ) checkGpu( hipFree( ptr ) )
+#define gpuFreeHost( ptr ) checkGpu( hipHostFree( ptr ) )
+
+#define gpuSetDevice hipSetDevice
+#define gpuDeviceSynchronize hipDeviceSynchronize
+#define gpuDeviceReset hipDeviceReset
+
+#define gpuLaunchKernel( kernel, blocks, threads, ... ) kernel<<<blocks, threads>>>( __VA_ARGS__ )
+#define gpuLaunchKernelSharedMem( kernel, blocks, threads, sharedMem, ... ) kernel<<<blocks, threads, sharedMem>>>( __VA_ARGS__ )
+
+//--------------------------------------------------------------------------
+
+#endif
+
+#endif // MG5AMC_GPUABSTRACTION_H
diff --git a/epochX/cudacpp/ee_mumu.sa/SubProcesses/CudaRuntime.h b/epochX/cudacpp/gg_tt.mad/SubProcesses/GpuRuntime.h
similarity index 62%
rename from epochX/cudacpp/ee_mumu.sa/SubProcesses/CudaRuntime.h
rename to epochX/cudacpp/gg_tt.mad/SubProcesses/GpuRuntime.h
index 64ce52f4b3..93579ef08b 100644
--- a/epochX/cudacpp/ee_mumu.sa/SubProcesses/CudaRuntime.h
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/GpuRuntime.h
@@ -1,49 +1,50 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
-// Created by: S. Roiser (Jul 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Created by: J. Teig (Jun 2023, based on earlier work by S. Roiser) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
-#ifndef MG5AMC_CUDARUNTIME_H
-#define MG5AMC_CUDARUNTIME_H 1
+#ifndef MG5AMC_GPURUNTIME_H
+#define MG5AMC_GPURUNTIME_H 1
 
 // MG5AMC on GPU uses the CUDA runtime API, not the lower level CUDA driver API
 // See https://docs.nvidia.com/cuda/cuda-runtime-api/driver-vs-runtime-api.html#driver-vs-runtime-api
 
-#include <cassert>
+#include "GpuAbstraction.h"
+
 #include <iostream>
 
 //--------------------------------------------------------------------------
 
 // See https://stackoverflow.com/a/14038590
-#ifdef __CUDACC__ /* clang-format off */
-#define checkCuda( code ) { assertCuda( code, __FILE__, __LINE__ ); }
-inline void assertCuda( cudaError_t code, const char* file, int line, bool abort = true )
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
+#define checkGpu( code ) { assertGpu( code, __FILE__, __LINE__ ); }
+inline void assertGpu( gpuError_t code, const char* file, int line, bool abort = true )
 {
-  if( code != cudaSuccess )
+  if( code != gpuSuccess )
   {
-    printf( "ERROR! assertCuda: '%s' (%d) in %s:%d\n", cudaGetErrorString( code ), code, file, line );
-    if( abort ) assert( code == cudaSuccess );
+    printf( "ERROR! assertGpu: '%s' (%d) in %s:%d\n", gpuGetErrorString( code ), code, file, line );
+    if( abort ) assert( code == gpuSuccess );
   }
 }
 #endif /* clang-format on */
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
+  // Instantiate a GpuRuntime at the beginnining of the application's main to
+  // invoke gpuSetDevice(0) in the constructor and book a gpuDeviceReset() call in the destructor
   // *** FIXME! This will all need to be designed differently when going to multi-GPU nodes! ***
-  struct CudaRuntime final
+  struct GpuRuntime final
   {
-    CudaRuntime( const bool debug = true )
+    GpuRuntime( const bool debug = true )
       : m_debug( debug ) { setUp( m_debug ); }
-    ~CudaRuntime() { tearDown( m_debug ); }
-    CudaRuntime( const CudaRuntime& ) = delete;
-    CudaRuntime( CudaRuntime&& ) = delete;
-    CudaRuntime& operator=( const CudaRuntime& ) = delete;
-    CudaRuntime& operator=( CudaRuntime&& ) = delete;
+    ~GpuRuntime() { tearDown( m_debug ); }
+    GpuRuntime( const GpuRuntime& ) = delete;
+    GpuRuntime( GpuRuntime&& ) = delete;
+    GpuRuntime& operator=( const GpuRuntime& ) = delete;
+    GpuRuntime& operator=( GpuRuntime&& ) = delete;
     bool m_debug;
 
     // Set up CUDA application
@@ -62,8 +63,8 @@ namespace mg5amcGpu
       */
       // Replace cudaFree(0) by cudaSetDevice(0), even if it is not really needed either
       // (but see https://developer.nvidia.com/blog/cuda-pro-tip-always-set-current-device-avoid-multithreading-bugs)
-      if( debug ) std::cout << "__CudaRuntime: calling cudaSetDevice(0)" << std::endl;
-      checkCuda( cudaSetDevice( 0 ) ); // SLOW!
+      if( debug ) std::cout << "__GpuRuntime: calling GpuSetDevice(0)" << std::endl;
+      checkGpu( gpuSetDevice( 0 ) ); // SLOW!
     }
 
     // Tear down CUDA application (call cudaDeviceReset)
@@ -72,14 +73,13 @@ namespace mg5amcGpu
     // See https://docs.nvidia.com/cuda/cuda-memcheck/index.html#leak-checking
     static void tearDown( const bool debug = true )
     {
-      if( debug ) std::cout << "__CudaRuntime: calling cudaDeviceReset()" << std::endl;
-      checkCuda( cudaDeviceReset() );
+      if( debug ) std::cout << "__GpuRuntime: calling GpuDeviceReset()" << std::endl;
+      checkGpu( gpuDeviceReset() );
     }
   };
-
 }
 #endif
 
 //--------------------------------------------------------------------------
 
-#endif // MG5AMC_CUDARUNTIME_H
+#endif // MG5AMC_GPURUNTIME_H
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/MadgraphTest.h b/epochX/cudacpp/gg_tt.mad/SubProcesses/MadgraphTest.h
index ef40624c88..6054185300 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/MadgraphTest.h
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/MadgraphTest.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Hageboeck (Dec 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MADGRAPHTEST_H_
 #define MADGRAPHTEST_H_ 1
@@ -14,7 +14,11 @@
 
 #include <array>
 #include <cmath>
-#include <filesystem>
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
 #include <fstream>
 #include <iomanip>
 #include <memory>
@@ -22,7 +26,7 @@
 #include <string>
 #include <vector>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using mg5amcGpu::CPPProcess;
 #else
 using mg5amcCpu::CPPProcess;
@@ -201,7 +205,7 @@ class MadgraphTest : public testing::TestWithParam<TestDriverBase*>
 
 // Since we link both the CPU-only and GPU tests into the same executable, we prevent
 // a multiply defined symbol by only compiling this in the non-CUDA phase:
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
 /// Compare momenta and matrix elements.
 /// This uses an implementation of TestDriverBase to run a madgraph workflow,
@@ -219,7 +223,14 @@ TEST_P( MadgraphTest, CompareMomentaAndME )
   const char* dumpEventsC = getenv( "CUDACPP_RUNTEST_DUMPEVENTS" );
   const bool dumpEvents = ( dumpEventsC != 0 ) && ( std::string( dumpEventsC ) != "" );
   const std::string refFileName = testDriver->getRefFileName();
+  /*
+#ifdef __HIPCC__
+  const std::string dumpFileName = std::experimental::filesystem::path( refFileName ).filename();
+#else
   const std::string dumpFileName = std::filesystem::path( refFileName ).filename();
+#endif
+  */
+  const std::string dumpFileName = refFileName; // bypass std::filesystem #803
   std::ofstream dumpFile;
   if( dumpEvents )
   {
@@ -307,6 +318,6 @@ TEST_P( MadgraphTest, CompareMomentaAndME )
   }
 }
 
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
 
 #endif /* MADGRAPHTEST_H_ */
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/MatrixElementKernels.cc b/epochX/cudacpp/gg_tt.mad/SubProcesses/MatrixElementKernels.cc
index 74b5239ebf..81699dfea9 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/MatrixElementKernels.cc
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/MatrixElementKernels.cc
@@ -1,12 +1,12 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "MatrixElementKernels.h"
 
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h" // Includes the abstraction for Nvidia/AMD compilation
 #include "MemoryAccessMomenta.h"
 #include "MemoryBuffers.h"
 
@@ -14,7 +14,7 @@
 
 //============================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu
 {
 
@@ -150,7 +150,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
@@ -209,13 +209,13 @@ namespace mg5amcGpu
     PinnedHostBufferHelicityMask hstIsGoodHel( ncomb );
     DeviceBufferHelicityMask devIsGoodHel( ncomb );
     // ... 0d1. Compute good helicity mask on the device
-    computeDependentCouplings<<<m_gpublocks, m_gputhreads>>>( m_gs.data(), m_couplings.data() );
+    gpuLaunchKernel( computeDependentCouplings, m_gpublocks, m_gputhreads, m_gs.data(), m_couplings.data() );
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-    sigmaKin_getGoodHel<<<m_gpublocks, m_gputhreads>>>( m_momenta.data(), m_couplings.data(), m_matrixElements.data(), m_numerators.data(), m_denominators.data(), devIsGoodHel.data() );
+    gpuLaunchKernel( sigmaKin_getGoodHel, m_gpublocks, m_gputhreads, m_momenta.data(), m_couplings.data(), m_matrixElements.data(), m_numerators.data(), m_denominators.data(), devIsGoodHel.data() );
 #else
-    sigmaKin_getGoodHel<<<m_gpublocks, m_gputhreads>>>( m_momenta.data(), m_couplings.data(), m_matrixElements.data(), devIsGoodHel.data() );
+    gpuLaunchKernel( sigmaKin_getGoodHel, m_gpublocks, m_gputhreads, m_momenta.data(), m_couplings.data(), m_matrixElements.data(), devIsGoodHel.data() );
 #endif
-    checkCuda( cudaPeekAtLastError() );
+    checkGpu( gpuPeekAtLastError() );
     // ... 0d2. Copy back good helicity mask to the host
     copyHostFromDevice( hstIsGoodHel, devIsGoodHel );
     // ... 0d3. Copy back good helicity list to constant memory on the device
@@ -226,19 +226,19 @@ namespace mg5amcGpu
 
   void MatrixElementKernelDevice::computeMatrixElements( const unsigned int channelId )
   {
-    computeDependentCouplings<<<m_gpublocks, m_gputhreads>>>( m_gs.data(), m_couplings.data() );
+    gpuLaunchKernel( computeDependentCouplings, m_gpublocks, m_gputhreads, m_gs.data(), m_couplings.data() );
 #ifndef MGONGPU_NSIGHT_DEBUG
     constexpr unsigned int sharedMemSize = 0;
 #else
     constexpr unsigned int sharedMemSize = ntpbMAX * sizeof( float );
 #endif
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-    sigmaKin<<<m_gpublocks, m_gputhreads, sharedMemSize>>>( m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), channelId, m_numerators.data(), m_denominators.data(), m_selhel.data(), m_selcol.data() );
+    gpuLaunchKernelSharedMem( sigmaKin, m_gpublocks, m_gputhreads, sharedMemSize, m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), channelId, m_numerators.data(), m_denominators.data(), m_selhel.data(), m_selcol.data() );
 #else
-    sigmaKin<<<m_gpublocks, m_gputhreads, sharedMemSize>>>( m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), m_selhel.data(), m_selcol.data() );
+    gpuLaunchKernelSharedMem( sigmaKin, m_gpublocks, m_gputhreads, sharedMemSize, m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), m_selhel.data(), m_selcol.data() );
 #endif
-    checkCuda( cudaPeekAtLastError() );
-    checkCuda( cudaDeviceSynchronize() );
+    checkGpu( gpuPeekAtLastError() );
+    checkGpu( gpuDeviceSynchronize() );
   }
 
   //--------------------------------------------------------------------------
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/MatrixElementKernels.h b/epochX/cudacpp/gg_tt.mad/SubProcesses/MatrixElementKernels.h
index 23e84757a2..72bd8f195b 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/MatrixElementKernels.h
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/MatrixElementKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MATRIXELEMENTKERNELS_H
 #define MATRIXELEMENTKERNELS_H 1
@@ -10,7 +10,7 @@
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -81,7 +81,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating matrix element calculations on a CPU host
   class MatrixElementKernelHost final : public MatrixElementKernelBase, public NumberOfEvents
   {
@@ -130,7 +130,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating matrix element calculations on a GPU device
   class MatrixElementKernelDevice : public MatrixElementKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessAmplitudes.h b/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessAmplitudes.h
index 573b3bbbc9..ffb76e93de 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessAmplitudes.h
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessAmplitudes.h
@@ -15,7 +15,7 @@
 #define MGONGPU_TRIVIAL_AMPLITUDES 1
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessCouplings.h b/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessCouplings.h
index 35a3af42e0..3afdf3e554 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessCouplings.h
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessCouplings.h
@@ -15,7 +15,7 @@
 #include "MemoryBuffers.h"       // for HostBufferCouplings::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessCouplingsFixed.h b/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessCouplingsFixed.h
index dc0d93afff..ffcdf4dbef 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessCouplingsFixed.h
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessCouplingsFixed.h
@@ -14,7 +14,7 @@
 //#include "MemoryAccessHelpers.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessDenominators.h b/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessDenominators.h
index 3bce635718..66f2d32a6b 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessDenominators.h
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessDenominators.h
@@ -10,7 +10,7 @@
 #include "MemoryAccessGs.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessGs.h b/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessGs.h
index 31311aa375..4c726b30f3 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessGs.h
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessGs.h
@@ -13,7 +13,7 @@
 #include "MemoryBuffers.h" // for HostBufferMatrixElements::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessHelpers.h b/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessHelpers.h
index c82a6c7635..db73e4e064 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessHelpers.h
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessHelpers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessHelpers_H
 #define MemoryAccessHelpers_H 1
@@ -105,7 +105,7 @@ class KernelAccessHelper : public MemoryAccessHelper<T>
     }
     else
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
       //printf( "kernelAccessRecord: ievt=%d threadId=%d\n", ievt, threadIdx.x );
       return T::ieventAccessRecord( buffer, ievt ); // NB fptype and fptype_sv coincide for CUDA
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessMatrixElements.h b/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessMatrixElements.h
index f32e6fea5b..3741011971 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessMatrixElements.h
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessMatrixElements.h
@@ -13,7 +13,7 @@
 #include "MemoryBuffers.h" // for HostBufferMatrixElements::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessMomenta.h b/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessMomenta.h
index 29266de32c..3be229d392 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessMomenta.h
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessMomenta.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessMomenta_H
 #define MemoryAccessMomenta_H 1
@@ -13,7 +13,7 @@
 #include "MemoryAccessVectors.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -30,7 +30,7 @@ namespace mg5amcCpu
 
     // Number of Events Per Page in the momenta AOSOA memory buffer layout
     // (these are all best kept as a compile-time constants: see issue #23)
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
     // -----------------------------------------------------------------------------------------------
     // --- GPUs: neppM is best set to a power of 2 times the number of fptype's in a 32-byte cacheline
     // --- This is relevant to ensure coalesced access to momenta in global memory
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessNumerators.h b/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessNumerators.h
index b152183b28..18991f4fa6 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessNumerators.h
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessNumerators.h
@@ -10,7 +10,7 @@
 #include "MemoryAccessGs.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessRandomNumbers.h b/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessRandomNumbers.h
index e2988d39f3..40cb089135 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessRandomNumbers.h
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessRandomNumbers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessRandomNumbers_H
 #define MemoryAccessRandomNumbers_H 1
@@ -11,7 +11,7 @@
 #include "CPPProcess.h"
 #include "MemoryAccessHelpers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using mg5amcGpu::CPPProcess;
 #else
 using mg5amcCpu::CPPProcess;
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessVectors.h b/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessVectors.h
index e9b197368e..08faccff0f 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessVectors.h
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessVectors.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessVectors_H
 #define MemoryAccessVectors_H 1
@@ -10,7 +10,7 @@
 
 #include "mgOnGpuVectors.h"
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu // this is only needed for CPU SIMD vectorization
 {
 
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessWavefunctions.h b/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessWavefunctions.h
index 5428aaf933..33bef4559e 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessWavefunctions.h
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryAccessWavefunctions.h
@@ -15,7 +15,7 @@
 #define MGONGPU_TRIVIAL_WAVEFUNCTIONS 1
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryBuffers.h b/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryBuffers.h
index 3093e6ed18..7756a71621 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryBuffers.h
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/MemoryBuffers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021, based on earlier work by S. Hageboeck) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryBuffers_H
 #define MemoryBuffers_H 1
@@ -11,12 +11,12 @@
 #include "mgOnGpuCxtypes.h"
 
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "Parameters_sm.h"
 
 #include <sstream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -87,7 +87,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   constexpr bool HostBufferALIGNED = false;   // ismisaligned=false
   constexpr bool HostBufferMISALIGNED = true; // ismisaligned=true
 
@@ -119,7 +119,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA pinned host buffer
   template<typename T>
   class PinnedHostBufferBase : public BufferBase<T>
@@ -128,18 +128,18 @@ namespace mg5amcCpu
     PinnedHostBufferBase( const size_t size )
       : BufferBase<T>( size, false )
     {
-      checkCuda( cudaMallocHost( &( this->m_data ), this->bytes() ) );
+      gpuMallocHost( &( this->m_data ), this->bytes() );
     }
     virtual ~PinnedHostBufferBase()
     {
-      checkCuda( cudaFreeHost( this->m_data ) );
+      gpuFreeHost( this->m_data );
     }
   };
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA device buffer
   template<typename T>
   class DeviceBufferBase : public BufferBase<T>
@@ -148,18 +148,18 @@ namespace mg5amcCpu
     DeviceBufferBase( const size_t size )
       : BufferBase<T>( size, true )
     {
-      checkCuda( cudaMalloc( &( this->m_data ), this->bytes() ) );
+      gpuMalloc( &( this->m_data ), this->bytes() );
     }
     virtual ~DeviceBufferBase()
     {
-      checkCuda( cudaFree( this->m_data ) );
+      gpuFree( this->m_data );
     }
   };
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for a given number of events
   template<typename T, size_t sizePerEvent, bool ismisaligned>
   class HostBuffer : public HostBufferBase<T, ismisaligned>, virtual private NumberOfEvents
@@ -175,7 +175,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA pinned host buffer for a given number of events
   template<typename T, size_t sizePerEvent>
   class PinnedHostBuffer : public PinnedHostBufferBase<T>, virtual private NumberOfEvents
@@ -191,7 +191,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA device buffer for a given number of events
   template<typename T, size_t sizePerEvent>
   class DeviceBuffer : public DeviceBufferBase<T>, virtual private NumberOfEvents
@@ -213,7 +213,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for momenta random numbers
   constexpr size_t sizePerEventRndNumMomenta = MemoryBuffers::np4 * MemoryBuffers::nparf;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for momenta random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumMomenta, HostBufferALIGNED> HostBufferRndNumMomenta;
 #else
@@ -232,7 +232,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer with ONE fptype per event
   constexpr size_t sizePerEventOneFp = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer with ONE fptype per event
   typedef HostBuffer<fptype, sizePerEventOneFp, HostBufferALIGNED> HostBufferOneFp;
 #else
@@ -257,7 +257,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for Gs
   constexpr size_t sizePerEventGs = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventGs, HostBufferALIGNED> HostBufferGs;
 #else
@@ -276,7 +276,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for numerators
   constexpr size_t sizePerEventNumerators = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventNumerators, HostBufferALIGNED> HostBufferNumerators;
 #else
@@ -296,7 +296,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for denominators
   constexpr size_t sizePerEventDenominators = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventDenominators, HostBufferALIGNED> HostBufferDenominators;
 #else
@@ -315,7 +315,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for random numbers
   constexpr size_t sizePerEventCouplings = MemoryBuffers::ndcoup * MemoryBuffers::nx2;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventCouplings, HostBufferALIGNED> HostBufferCouplings;
 #else
@@ -333,7 +333,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for momenta
   constexpr size_t sizePerEventMomenta = MemoryBuffers::np4 * MemoryBuffers::npar;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for momenta
   typedef HostBuffer<fptype, sizePerEventMomenta, HostBufferALIGNED> HostBufferMomenta;
   //typedef HostBuffer<fptype, sizePerEventMomenta, HostBufferMISALIGNED> HostBufferMomenta; // TEST MISALIGNMENT!
@@ -352,7 +352,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for sampling weights
   constexpr size_t sizePerEventWeights = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for sampling weights
   typedef HostBuffer<fptype, sizePerEventWeights, HostBufferALIGNED> HostBufferWeights;
 #else
@@ -370,7 +370,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for matrix elements
   constexpr size_t sizePerEventMatrixElements = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for matrix elements
   typedef HostBuffer<fptype, sizePerEventMatrixElements, HostBufferALIGNED> HostBufferMatrixElements;
 #else
@@ -385,7 +385,7 @@ namespace mg5amcCpu
   // A base class encapsulating a memory buffer for the helicity mask
   typedef BufferBase<bool> BufferHelicityMask;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for the helicity mask
   typedef HostBufferBase<bool, HostBufferALIGNED> HostBufferHelicityMask;
 #else
@@ -403,7 +403,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for wavefunctions
   constexpr size_t sizePerEventWavefunctions = MemoryBuffers::nw6 * MemoryBuffers::nx2;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for wavefunctions
   typedef HostBuffer<fptype, sizePerEventWavefunctions, HostBufferALIGNED> HostBufferWavefunctions;
 #else
@@ -421,7 +421,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for helicity random numbers
   constexpr size_t sizePerEventRndNumHelicity = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for helicity random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumHelicity, HostBufferALIGNED> HostBufferRndNumHelicity;
 #else
@@ -439,7 +439,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for color random numbers
   constexpr size_t sizePerEventRndNumColor = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for color random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumColor, HostBufferALIGNED> HostBufferRndNumColor;
 #else
@@ -457,7 +457,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for helicity selection
   constexpr size_t sizePerEventSelectedHelicity = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for helicity selection
   typedef HostBuffer<int, sizePerEventSelectedHelicity, HostBufferALIGNED> HostBufferSelectedHelicity;
 #else
@@ -475,7 +475,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for color selection
   constexpr size_t sizePerEventSelectedColor = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for color selection
   typedef HostBuffer<int, sizePerEventSelectedColor, HostBufferALIGNED> HostBufferSelectedColor;
 #else
@@ -487,7 +487,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<class Tdst, class Tsrc>
   void copyDeviceFromHost( Tdst& dst, const Tsrc& src ) // keep the same order of arguments as in memcpy
   {
@@ -504,13 +504,13 @@ namespace mg5amcCpu
       throw std::runtime_error( sstr.str() );
     }
     // NB (PR #45): cudaMemcpy involves an intermediate memcpy to pinned memory if host array is a not a pinned host array
-    checkCuda( cudaMemcpy( dst.data(), src.data(), src.bytes(), cudaMemcpyHostToDevice ) );
+    gpuMemcpy( dst.data(), src.data(), src.bytes(), gpuMemcpyHostToDevice );
   }
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<class Tdst, class Tsrc>
   void copyHostFromDevice( Tdst& dst, const Tsrc& src ) // keep the same order of arguments as in memcpy
   {
@@ -527,7 +527,7 @@ namespace mg5amcCpu
       throw std::runtime_error( sstr.str() );
     }
     // NB (PR #45): cudaMemcpy involves an intermediate memcpy to pinned memory if host array is a not a pinned host array
-    checkCuda( cudaMemcpy( dst.data(), src.data(), src.bytes(), cudaMemcpyDeviceToHost ) );
+    gpuMemcpy( dst.data(), src.data(), src.bytes(), gpuMemcpyDeviceToHost );
   }
 #endif
 
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/CPPProcess.cc b/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/CPPProcess.cc
index 18052b6676..dbaa56b35c 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/CPPProcess.cc
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_sm.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -46,7 +45,7 @@
 // Class member functions for calculating the matrix elements for
 // Process: g g > t t~ WEIGHTED<=2 @1
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -80,7 +79,7 @@ namespace mg5amcCpu
   __device__ const fptype cIPD[2] = { (fptype)Parameters_sm::mdl_MT, (fptype)Parameters_sm::mdl_WT };
   __device__ const fptype* cIPC = nullptr; // unused as nicoup=0
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ fptype cIPD[2];
   __device__ __constant__ fptype* cIPC = nullptr; // unused as nicoup=0
 #else
@@ -90,7 +89,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -118,13 +117,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -151,7 +150,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -187,7 +186,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -200,8 +199,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -302,7 +303,7 @@ namespace mg5amcCpu
         { 16, -2 },
         { -2, 16 } }; // 2-D array[2][2]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -359,7 +360,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -418,7 +419,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -465,8 +466,8 @@ namespace mg5amcCpu
       { 1, 1, -1, -1 },
       { 1, 1, 1, 1 },
       { 1, 1, 1, -1 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -506,9 +507,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     const fptype tIPD[2] = { (fptype)m_pars->mdl_MT, (fptype)m_pars->mdl_WT };
     //const cxtype tIPC[0] = { ... }; // nicoup=0
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) ) );
-    //checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ) ); // nicoup=0
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) );
+    //gpuMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
 #else
     memcpy( cIPD, tIPD, 2 * sizeof( fptype ) );
     //memcpy( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
@@ -544,7 +545,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -573,6 +574,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -586,6 +591,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -609,12 +616,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -635,7 +642,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -761,9 +768,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -787,7 +794,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -807,7 +814,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 256 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -821,9 +828,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -851,7 +861,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -1061,7 +1071,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/CPPProcess.h b/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/CPPProcess.h
index 3ebd92c038..4a88a07226 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/CPPProcess.h
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -107,7 +107,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -120,7 +120,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -150,7 +150,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/CudaRuntime.h b/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/GpuAbstraction.h b/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/GpuRuntime.h b/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/check_sa.cc b/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/check_sa.cc
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/RamboSamplingKernels.cc b/epochX/cudacpp/gg_tt.mad/SubProcesses/RamboSamplingKernels.cc
index da68aa9255..79abbcc4f8 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/RamboSamplingKernels.cc
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/RamboSamplingKernels.cc
@@ -1,11 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "RamboSamplingKernels.h"
 
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "MemoryAccessMomenta.h"
 #include "MemoryAccessRandomNumbers.h"
 #include "MemoryAccessWeights.h"
@@ -14,7 +14,7 @@
 
 #include <sstream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -92,7 +92,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingKernelDevice::RamboSamplingKernelDevice( const fptype energy,               // input: energy
                                                         const BufferRndNumMomenta& rndmom, // input: random numbers in [0,1]
                                                         BufferMomenta& momenta,            // output: momenta
@@ -135,7 +135,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   getMomentaInitialDevice( const fptype energy,
                            fptype* momenta )
@@ -147,17 +147,17 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   void
   RamboSamplingKernelDevice::getMomentaInitial()
   {
-    getMomentaInitialDevice<<<m_gpublocks, m_gputhreads>>>( m_energy, m_momenta.data() );
+    gpuLaunchKernel( getMomentaInitialDevice, m_gpublocks, m_gputhreads, m_energy, m_momenta.data() );
   }
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   getMomentaFinalDevice( const fptype energy,
                          const fptype* rndmom,
@@ -171,11 +171,11 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   void
   RamboSamplingKernelDevice::getMomentaFinal()
   {
-    getMomentaFinalDevice<<<m_gpublocks, m_gputhreads>>>( m_energy, m_rndmom.data(), m_momenta.data(), m_weights.data() );
+    gpuLaunchKernel( getMomentaFinalDevice, m_gpublocks, m_gputhreads, m_energy, m_rndmom.data(), m_momenta.data(), m_weights.data() );
   }
 #endif
 
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/RamboSamplingKernels.h b/epochX/cudacpp/gg_tt.mad/SubProcesses/RamboSamplingKernels.h
index 184089efd7..7c214cd74b 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/RamboSamplingKernels.h
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/RamboSamplingKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef RAMBOSAMPLINGKERNELS_H
 #define RAMBOSAMPLINGKERNELS_H 1
@@ -10,7 +10,7 @@
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -93,7 +93,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating RAMBO phase space sampling on a GPU device
   class RamboSamplingKernelDevice final : public SamplingKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/RandomNumberKernels.h b/epochX/cudacpp/gg_tt.mad/SubProcesses/RandomNumberKernels.h
index 188a72c2c9..21d63beeac 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/RandomNumberKernels.h
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/RandomNumberKernels.h
@@ -1,14 +1,14 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef RANDOMNUMBERKERNELS_H
 #define RANDOMNUMBERKERNELS_H 1
 
 #include "mgOnGpuConfig.h"
 
-// NB This must come AFTER mgOnGpuConfig.h which contains our definition of __global__ when __CUDACC__ is not defined
+// NB This must come AFTER mgOnGpuConfig.h which contains our definition of __global__ when MGONGPUCPP_GPUIMPL is not defined
 #ifndef MGONGPU_HAS_NO_CURAND
 //#include "curand.h"
 struct curandGenerator_st; // forward definition from curand.h
@@ -16,7 +16,7 @@ struct curandGenerator_st; // forward definition from curand.h
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/cudacpp.mk b/epochX/cudacpp/gg_tt.mad/SubProcesses/cudacpp.mk
index 509307506b..df74dfc284 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/cudacpp.mk
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/cudacpp.mk
@@ -1,7 +1,7 @@
 # Copyright (C) 2020-2023 CERN and UCLouvain.
 # Licensed under the GNU Lesser General Public License (version 3 or later).
 # Created by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-# Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+# Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #=== Determine the name of this makefile (https://ftp.gnu.org/old-gnu/Manuals/make-3.80/html_node/make_17.html)
 #=== NB: use ':=' to ensure that the value of CUDACPP_MAKEFILE is not modified further down after including make_opts
@@ -42,10 +42,10 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure common compiler flags for C++ and CUDA
+#=== Configure common compiler flags for C++ and CUDA/HIP
 
 INCFLAGS = -I.
-OPTFLAGS = -O3 # this ends up in CUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
+OPTFLAGS = -O3 # this ends up in GPUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
 
 # Dependency on src directory
 MG5AMC_COMMONLIB = mg5amc_common
@@ -121,24 +121,45 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure the CUDA compiler
+#=== Configure the GPU compiler (CUDA or HIP)
+
+# FIXME! (AV 24.01.2024)
+# In the current implementation (without separate builds for C++ and CUDA/HIP), we first check for cudacc and hipcc in CUDA_HOME and HIP_HOME.
+# If CUDA_HOME or HIP_HOME are not set, try to determine them from the path to cudacc and hipcc.
+# While convoluted, this is currently necessary to allow disabling CUDA/HIP builds by setting CUDA_HOME or HIP_HOME to invalid paths.
+# This will (probably?) be fixed when separate C++ and CUDA/HIP builds are implemented (PR #775).
 
 # If CXX is not a single word (example "clang++ --gcc-toolchain...") then disable CUDA builds (issue #505)
-# This is because it is impossible to pass this to "CUFLAGS += -ccbin <host-compiler>" below
+# This is because it is impossible to pass this to "GPUFLAGS += -ccbin <host-compiler>" below
 ifneq ($(words $(subst ccache ,,$(CXX))),1) # allow at most "CXX=ccache <host-compiler>" from outside
   $(warning CUDA builds are not supported for multi-word CXX "$(CXX)")
   override CUDA_HOME=disabled
 endif
 
-# If CUDA_HOME is not set, try to set it from the location of nvcc
+# If CUDA_HOME is not set, try to set it from the path to nvcc
 ifndef CUDA_HOME
-  CUDA_HOME = $(patsubst %bin/nvcc,%,$(shell which nvcc 2>/dev/null))
+  CUDA_HOME = $(patsubst %/bin/nvcc,%,$(shell which nvcc 2>/dev/null))
   $(warning CUDA_HOME was not set: using "$(CUDA_HOME)")
 endif
 
-# Set NVCC as $(CUDA_HOME)/bin/nvcc if it exists
+# If HIP_HOME is not set, try to set it from the path to hipcc
+ifndef HIP_HOME
+  HIP_HOME = $(patsubst %/bin/hipcc,%,$(shell which hipcc 2>/dev/null))
+  $(warning HIP_HOME was not set: using "$(HIP_HOME)")
+endif
+
+# FIXME! (AV 24.01.2024)
+# In the current implementation (without separate builds for C++ and CUDA/HIP),
+# builds are performed for HIP only if CUDA is not found in the path.
+# If both CUDA and HIP are installed, HIP builds can be triggered by unsetting CUDA_HOME.
+# This will be fixed when separate C++ and CUDA/HIP builds are implemented (PR #775).
+
+#--- Option 1: CUDA exists -> use CUDA
+
+# Set GPUCC as $(CUDA_HOME)/bin/nvcc if it exists
 ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
-  NVCC = $(CUDA_HOME)/bin/nvcc
+
+  GPUCC = $(CUDA_HOME)/bin/nvcc
   USE_NVTX ?=-DUSE_NVTX
   # See https://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/index.html
   # See https://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/
@@ -148,7 +169,7 @@ ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
   # Examples: use 60 for P100 (Piz Daint), 80 for A100 (Juwels Booster, NVidia raplab/Curiosity).
   MADGRAPH_CUDA_ARCHITECTURE ?= 70
   ###CUARCHFLAGS = -gencode arch=compute_$(MADGRAPH_CUDA_ARCHITECTURE),code=compute_$(MADGRAPH_CUDA_ARCHITECTURE) -gencode arch=compute_$(MADGRAPH_CUDA_ARCHITECTURE),code=sm_$(MADGRAPH_CUDA_ARCHITECTURE) # Older implementation (AV): go back to this one for multi-GPU support #533
-  ###CUARCHFLAGS = --gpu-architecture=compute_$(MADGRAPH_CUDA_ARCHITECTURE) --gpu-code=sm_$(MADGRAPH_CUDA_ARCHITECTURE),compute_$(MADGRAPH_CUDA_ARCHITECTURE) # Newer implementation (SH): cannot use this as-is for multi-GPU support #533
+  ###CUARCHFLAGS = --gpu-architecture=compute_$(MADGRAPH_CUDA_ARCHITECTURE) --gpu-code=sm_$(MADGRAPH_CUDA_ARCHITECTURE),compute_$(MADGRAPH_CUDA_ARCHITECTURE)  # Newer implementation (SH): cannot use this as-is for multi-GPU support #533
   comma:=,
   CUARCHFLAGS = $(foreach arch,$(subst $(comma), ,$(MADGRAPH_CUDA_ARCHITECTURE)),-gencode arch=compute_$(arch),code=compute_$(arch) -gencode arch=compute_$(arch),code=sm_$(arch))
   CUINC = -I$(CUDA_HOME)/include/
@@ -158,41 +179,79 @@ ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
     CURANDLIBFLAGS = -L$(CUDA_HOME)/lib64/ -lcurand # NB: -lcuda is not needed here!
   endif
   CUOPTFLAGS = -lineinfo
-  CUFLAGS = $(foreach opt, $(OPTFLAGS), -Xcompiler $(opt)) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
-  ###CUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
-  ###NVCC_VERSION = $(shell $(NVCC) --version | grep 'Cuda compilation tools' | cut -d' ' -f5 | cut -d, -f1)
-  CUFLAGS += -std=c++17 # need CUDA >= 11.2 (see #333): this is enforced in mgOnGpuConfig.h
+  ###GPUFLAGS = $(OPTFLAGS) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
+  GPUFLAGS = $(foreach opt, $(OPTFLAGS), -Xcompiler $(opt)) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
+  ###GPUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
+  ###GPUCC_VERSION = $(shell $(GPUCC) --version | grep 'Cuda compilation tools' | cut -d' ' -f5 | cut -d, -f1)
+  GPUFLAGS += -std=c++17 # need CUDA >= 11.2 (see #333): this is enforced in mgOnGpuConfig.h
   # Without -maxrregcount: baseline throughput: 6.5E8 (16384 32 12) up to 7.3E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 160 # improves throughput: 6.9E8 (16384 32 12) up to 7.7E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 128 # improves throughput: 7.3E8 (16384 32 12) up to 7.6E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 96 # degrades throughput: 4.1E8 (16384 32 12) up to 4.5E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 64 # degrades throughput: 1.7E8 (16384 32 12) flat at 1.7E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 160 # improves throughput: 6.9E8 (16384 32 12) up to 7.7E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 128 # improves throughput: 7.3E8 (16384 32 12) up to 7.6E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 96 # degrades throughput: 4.1E8 (16384 32 12) up to 4.5E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 64 # degrades throughput: 1.7E8 (16384 32 12) flat at 1.7E8 (65536 128 12)
+  CUBUILDRULEFLAGS = -Xcompiler -fPIC -c
+  CCBUILDRULEFLAGS = -Xcompiler -fPIC -c -x cu
+  CUDATESTFLAGS = -lcuda
+
+  # Set the host C++ compiler for GPUCC via "-ccbin <host-compiler>"
+  # (NB issue #505: this must be a single word, "clang++ --gcc-toolchain..." is not supported)
+  GPUFLAGS += -ccbin $(shell which $(subst ccache ,,$(CXX)))
+
+  # Allow newer (unsupported) C++ compilers with older versions of CUDA if ALLOW_UNSUPPORTED_COMPILER_IN_CUDA is set (#504)
+  ifneq ($(origin ALLOW_UNSUPPORTED_COMPILER_IN_CUDA),undefined)
+  GPUFLAGS += -allow-unsupported-compiler
+  endif
+
 else ifneq ($(origin REQUIRE_CUDA),undefined)
+
   # If REQUIRE_CUDA is set but no cuda is found, stop here (e.g. for CI tests on GPU #443)
-  $(error No cuda installation found (set CUDA_HOME or make nvcc visible in PATH))
+  $(error No cuda installation found (set CUDA_HOME or make GPUCC visible in PATH))
+
+#--- Option 2: CUDA does not exist, HIP exists -> use HIP
+
+# Set GPUCC as $(HIP_HOME)/bin/hipcc if it exists
+else ifneq ($(wildcard $(HIP_HOME)/bin/hipcc),)
+
+  GPUCC = $(HIP_HOME)/bin/hipcc
+  #USE_NVTX ?=-DUSE_NVTX # should maybe find something equivalent to this in HIP?
+  HIPARCHFLAGS = -target x86_64-linux-gnu --offload-arch=gfx90a
+  HIPINC = -I$(HIP_HOME)/include/
+  # Note: -DHIP_FAST_MATH is equivalent to -use_fast_math in HIP 
+  # (but only for single precision line 208: https://rocm-developer-tools.github.io/HIP/hcc__detail_2math__functions_8h_source.html)
+  # Note: CUOPTFLAGS should not be used for HIP, it had been added here but was then removed (#808)
+  GPUFLAGS = $(OPTFLAGS) $(INCFLAGS) $(HIPINC) $(HIPARCHFLAGS) -DHIP_FAST_MATH -DHIP_PLATFORM=amd -fPIC
+  ###GPUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
+  GPUFLAGS += -std=c++17
+  ###GPUFLAGS+= --maxrregcount 255 # (AV: is this option valid on HIP and meaningful on AMD GPUs?)
+  CUBUILDRULEFLAGS = -fPIC -c
+  CCBUILDRULEFLAGS = -fPIC -c -x hip
+
+else ifneq ($(origin REQUIRE_HIP),undefined)
+
+  # If REQUIRE_HIP is set but no HIP is found, stop here (e.g. for CI tests on GPU #443)
+  $(error No hip installation found (set HIP_HOME or make GPUCC visible in PATH))
+
+#--- Option 3: CUDA does not exist, HIP does not exist -> switch off both CUDA and HIP
+
 else
-  # No cuda. Switch cuda compilation off and go to common random numbers in C++
+
+  # No cudacc and no hipcc: switch CUDA and HIP compilation off and go to common random numbers in C++
   $(warning CUDA_HOME is not set or is invalid: export CUDA_HOME to compile with cuda)
-  override NVCC=
+  $(warning HIP_HOME is not set or is invalid: export HIP_HOME to compile with hip)
+  override GPUCC=
   override USE_NVTX=
   override CUINC=
   override CURANDLIBFLAGS=
-endif
-export NVCC
-export CUFLAGS
 
-# Set the host C++ compiler for nvcc via "-ccbin <host-compiler>"
-# (NB issue #505: this must be a single word, "clang++ --gcc-toolchain..." is not supported)
-CUFLAGS += -ccbin $(shell which $(subst ccache ,,$(CXX)))
-
-# Allow newer (unsupported) C++ compilers with older versions of CUDA if ALLOW_UNSUPPORTED_COMPILER_IN_CUDA is set (#504)
-ifneq ($(origin ALLOW_UNSUPPORTED_COMPILER_IN_CUDA),undefined)
-CUFLAGS += -allow-unsupported-compiler
 endif
 
+# Export GPUCC (so that it can also be used in cudacpp_src.mk?)
+export GPUCC
+export GPUFLAGS
+
 #-------------------------------------------------------------------------------
 
-#=== Configure ccache for C++ and CUDA builds
+#=== Configure ccache for C++ and CUDA/HIP builds
 
 # Enable ccache if USECCACHE=1
 ifeq ($(USECCACHE)$(shell echo $(CXX) | grep ccache),1)
@@ -201,15 +260,15 @@ endif
 #ifeq ($(USECCACHE)$(shell echo $(AR) | grep ccache),1)
 #  override AR:=ccache $(AR)
 #endif
-ifneq ($(NVCC),)
-  ifeq ($(USECCACHE)$(shell echo $(NVCC) | grep ccache),1)
-    override NVCC:=ccache $(NVCC)
+ifneq ($(GPUCC),)
+  ifeq ($(USECCACHE)$(shell echo $(GPUCC) | grep ccache),1)
+    override GPUCC:=ccache $(GPUCC)
   endif
 endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure PowerPC-specific compiler flags for C++ and CUDA
+#=== Configure PowerPC-specific compiler flags for C++ and CUDA/HIP
 
 # PowerPC-specific CXX compiler flags (being reviewed)
 ifeq ($(UNAME_P),ppc64le)
@@ -225,9 +284,9 @@ else
   ######CXXFLAGS+= -fno-semantic-interposition # no benefit (neither alone, nor combined with -flto)
 endif
 
-# PowerPC-specific CUDA compiler flags (to be reviewed!)
+# PowerPC-specific CUDA/HIP compiler flags (to be reviewed!)
 ifeq ($(UNAME_P),ppc64le)
-  CUFLAGS+= -Xcompiler -mno-float128
+  GPUFLAGS+= -Xcompiler -mno-float128
 endif
 
 #-------------------------------------------------------------------------------
@@ -235,9 +294,11 @@ endif
 #=== Configure defaults and check if user-defined choices exist for OMPFLAGS, AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
 
 # Set the default OMPFLAGS choice
-ifneq ($(shell $(CXX) --version | egrep '^Intel'),)
+ifneq ($(findstring hipcc,$(GPUCC)),)
+override OMPFLAGS = # disable OpenMP MT when using hipcc #802
+else ifneq ($(shell $(CXX) --version | egrep '^Intel'),)
 override OMPFLAGS = -fopenmp
-###override OMPFLAGS = # disable OpenMP MT on Intel (was ok without nvcc but not ok with nvcc before #578)
+###override OMPFLAGS = # disable OpenMP MT on Intel (was ok without GPUCC but not ok with GPUCC before #578)
 else ifneq ($(shell $(CXX) --version | egrep '^(clang)'),)
 override OMPFLAGS = -fopenmp
 ###override OMPFLAGS = # disable OpenMP MT on clang (was not ok without or with nvcc before #578)
@@ -293,7 +354,10 @@ endif
 
 # Set the default RNDGEN (random number generator) choice
 ifeq ($(RNDGEN),)
-  ifeq ($(NVCC),)
+  ifeq ($(GPUCC),)
+    override RNDGEN = hasNoCurand
+  # Edgecase for HIP compilation
+  else ifeq ($(findstring hipcc,$(GPUCC)),hipcc)
     override RNDGEN = hasNoCurand
   else ifeq ($(RNDGEN),)
     override RNDGEN = hasCurand
@@ -310,7 +374,7 @@ export OMPFLAGS
 
 #-------------------------------------------------------------------------------
 
-#=== Set the CUDA/C++ compiler flags appropriate to user-defined choices of AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
+#=== Set the CUDA/HIP/C++ compiler flags appropriate to user-defined choices of AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
 
 # Set the build flags appropriate to OMPFLAGS
 $(info OMPFLAGS=$(OMPFLAGS))
@@ -368,13 +432,13 @@ CXXFLAGS+= $(AVXFLAGS)
 $(info FPTYPE=$(FPTYPE))
 ifeq ($(FPTYPE),d)
   CXXFLAGS += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
-  CUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
+  GPUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
 else ifeq ($(FPTYPE),f)
   CXXFLAGS += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
-  CUFLAGS  += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
+  GPUFLAGS  += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
 else ifeq ($(FPTYPE),m)
   CXXFLAGS += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
-  CUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
+  GPUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
 else
   $(error Unknown FPTYPE='$(FPTYPE)': only 'd', 'f' and 'm' are supported)
 endif
@@ -383,7 +447,7 @@ endif
 $(info HELINL=$(HELINL))
 ifeq ($(HELINL),1)
   CXXFLAGS += -DMGONGPU_INLINE_HELAMPS
-  CUFLAGS  += -DMGONGPU_INLINE_HELAMPS
+  GPUFLAGS  += -DMGONGPU_INLINE_HELAMPS
 else ifneq ($(HELINL),0)
   $(error Unknown HELINL='$(HELINL)': only '0' and '1' are supported)
 endif
@@ -392,7 +456,7 @@ endif
 $(info HRDCOD=$(HRDCOD))
 ifeq ($(HRDCOD),1)
   CXXFLAGS += -DMGONGPU_HARDCODE_PARAM
-  CUFLAGS  += -DMGONGPU_HARDCODE_PARAM
+  GPUFLAGS  += -DMGONGPU_HARDCODE_PARAM
 else ifneq ($(HRDCOD),0)
   $(error Unknown HRDCOD='$(HRDCOD)': only '0' and '1' are supported)
 endif
@@ -444,11 +508,11 @@ ifeq ($(UNAME_S),Darwin)
   override CULIBFLAGSRPATH2 =
 else
   # RPATH to cuda/cpp libs when linking executables
-  override CXXLIBFLAGSRPATH = -Wl,-rpath,$(LIBDIRRPATH)
-  override CULIBFLAGSRPATH = -Xlinker -rpath,$(LIBDIRRPATH)
+  override CXXLIBFLAGSRPATH = -Wl,-rpath=$(LIBDIRRPATH)
+  override CULIBFLAGSRPATH = -Xlinker -rpath=$(LIBDIRRPATH)
   # RPATH to common lib when linking cuda/cpp libs
-  override CXXLIBFLAGSRPATH2 = -Wl,-rpath,'$$ORIGIN'
-  override CULIBFLAGSRPATH2 = -Xlinker -rpath,'$$ORIGIN'
+  override CXXLIBFLAGSRPATH2 = -Wl,-rpath='$$ORIGIN'
+  override CULIBFLAGSRPATH2 = -Xlinker -rpath='$$ORIGIN'
 endif
 
 # Setting LD_LIBRARY_PATH or DYLD_LIBRARY_PATH in the RUNTIME is no longer necessary (neither on Linux nor on Mac)
@@ -461,7 +525,7 @@ override RUNTIME =
 cxx_main=$(BUILDDIR)/check.exe
 fcxx_main=$(BUILDDIR)/fcheck.exe
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 cu_main=$(BUILDDIR)/gcheck.exe
 fcu_main=$(BUILDDIR)/fgcheck.exe
 else
@@ -492,14 +556,15 @@ $(BUILDDIR)/.build.$(TAG):
 	@touch $(BUILDDIR)/.build.$(TAG)
 
 # Generic target and build rules: objects from CUDA compilation
-ifneq ($(NVCC),)
+# NB: CCBUILDRULEFLAGS includes "-x cu" for nvcc and "-x hip" for hipcc (#810)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/%.o : %.cu *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $(CUBUILDRULEFLAGS) $< -o $@
 
 $(BUILDDIR)/%_cu.o : %.cc *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c -x cu $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $(CCBUILDRULEFLAGS) $< -o $@
 endif
 
 # Generic target and build rules: objects from C++ compilation
@@ -509,11 +574,14 @@ $(BUILDDIR)/%.o : %.cc *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	$(CXX) $(CPPFLAGS) $(CXXFLAGS) -fPIC -c $< -o $@
 
 # Apply special build flags only to CrossSectionKernel.cc and gCrossSectionKernel.cu (no fast math, see #117 and #516)
+# Added edgecase for HIP compilation
 ifeq ($(shell $(CXX) --version | grep ^nvc++),)
 $(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS := $(filter-out -ffast-math,$(CXXFLAGS))
 $(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS += -fno-fast-math
-ifneq ($(NVCC),)
-$(BUILDDIR)/gCrossSectionKernels.o: CUFLAGS += -Xcompiler -fno-fast-math
+ifeq ($(findstring nvcc,$(GPUCC)),nvcc)
+  $(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -Xcompiler -fno-fast-math
+else
+  $(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -fno-fast-math
 endif
 endif
 
@@ -530,10 +598,10 @@ ifeq ($(RNDGEN),hasCurand)
 $(BUILDDIR)/CurandRandomNumberKernel.o: CXXFLAGS += $(CUINC)
 endif
 
-# Avoid "warning: builtin __has_trivial_... is deprecated; use __is_trivially_... instead" in nvcc with icx2023 (#592)
+# Avoid "warning: builtin __has_trivial_... is deprecated; use __is_trivially_... instead" in GPUCC with icx2023 (#592)
 ifneq ($(shell $(CXX) --version | egrep '^(Intel)'),)
-ifneq ($(NVCC),)
-CUFLAGS += -Xcompiler -Wno-deprecated-builtins
+ifneq ($(GPUCC),)
+GPUFLAGS += -Wno-deprecated-builtins
 endif
 endif
 
@@ -541,8 +609,8 @@ endif
 # This patch does remove the warning, but I prefer to keep it disabled for the moment...
 ###ifneq ($(shell $(CXX) --version | egrep '^(clang|Apple clang|Intel)'),)
 ###$(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS += -Wno-overriding-t-option
-###ifneq ($(NVCC),)
-###$(BUILDDIR)/gCrossSectionKernels.o: CUFLAGS += -Xcompiler -Wno-overriding-t-option
+###ifneq ($(GPUCC),)
+###$(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -Xcompiler -Wno-overriding-t-option
 ###endif
 ###endif
 
@@ -569,7 +637,7 @@ MG5AMC_CXXLIB = mg5amc_$(processid_short)_cpp
 cxx_objects_lib=$(BUILDDIR)/CPPProcess.o $(BUILDDIR)/MatrixElementKernels.o $(BUILDDIR)/BridgeKernels.o $(BUILDDIR)/CrossSectionKernels.o
 cxx_objects_exe=$(BUILDDIR)/CommonRandomNumberKernel.o $(BUILDDIR)/RamboSamplingKernels.o
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 MG5AMC_CULIB = mg5amc_$(processid_short)_cuda
 cu_objects_lib=$(BUILDDIR)/gCPPProcess.o $(BUILDDIR)/gMatrixElementKernels.o $(BUILDDIR)/gBridgeKernels.o $(BUILDDIR)/gCrossSectionKernels.o
 cu_objects_exe=$(BUILDDIR)/gCommonRandomNumberKernel.o $(BUILDDIR)/gRamboSamplingKernels.o
@@ -581,11 +649,17 @@ $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: cxx_objects_lib += $(BUILDDIR)/fbridge.o
 $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib)
 	$(CXX) -shared -o $@ $(cxx_objects_lib) $(CXXLIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: $(BUILDDIR)/fbridge_cu.o
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: cu_objects_lib += $(BUILDDIR)/fbridge_cu.o
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cu_objects_lib)
-	$(NVCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+# Bypass std::filesystem completely to ease portability on LUMI #803
+#ifneq ($(findstring hipcc,$(GPUCC)),)
+#	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB) -lstdc++fs
+#else
+#	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+#endif
 endif
 
 #-------------------------------------------------------------------------------
@@ -602,16 +676,16 @@ $(cxx_main): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PAT
 $(cxx_main): $(BUILDDIR)/check_sa.o $(LIBDIR)/lib$(MG5AMC_CXXLIB).so $(cxx_objects_exe) $(BUILDDIR)/CurandRandomNumberKernel.o
 	$(CXX) -o $@ $(BUILDDIR)/check_sa.o $(OMPFLAGS) -ldl -pthread $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe) $(BUILDDIR)/CurandRandomNumberKernel.o $(CURANDLIBFLAGS)
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(cu_main): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(cu_main): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(cu_main): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(cu_main): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 else ifneq ($(shell $(CXX) --version | grep ^nvc++),) # support nvc++ #531
 $(cu_main): LIBFLAGS += -L$(patsubst %bin/nvc++,%lib,$(subst ccache ,,$(CXX))) -lnvhpcatm -lnvcpumath -lnvc
 endif
 $(cu_main): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(cu_main): $(BUILDDIR)/gcheck_sa.o $(LIBDIR)/lib$(MG5AMC_CULIB).so $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o
-	$(NVCC) -o $@ $(BUILDDIR)/gcheck_sa.o $(CUARCHFLAGS) $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o $(CURANDLIBFLAGS)
+	$(GPUCC) -o $@ $(BUILDDIR)/gcheck_sa.o $(CUARCHFLAGS) $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o $(CURANDLIBFLAGS)
 endif
 
 #-------------------------------------------------------------------------------
@@ -635,19 +709,27 @@ $(fcxx_main): LIBFLAGS += -L$(shell dirname $(shell $(FC) --print-file-name libg
 endif
 $(fcxx_main): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(fcxx_main): $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler.o $(LIBDIR)/lib$(MG5AMC_CXXLIB).so $(cxx_objects_exe)
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(BUILDDIR)/fcheck_sa.o $(OMPFLAGS) $(BUILDDIR)/fsampler.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe) -lstdc++
+else
 	$(CXX) -o $@ $(BUILDDIR)/fcheck_sa.o $(OMPFLAGS) $(BUILDDIR)/fsampler.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe)
+endif
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(fcu_main): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(fcu_main): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(fcu_main): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(fcu_main): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 endif
 ifeq ($(UNAME_S),Darwin)
 $(fcu_main): LIBFLAGS += -L$(shell dirname $(shell $(FC) --print-file-name libgfortran.dylib)) # add path to libgfortran on Mac #375
 endif
 $(fcu_main): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(fcu_main): $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBDIR)/lib$(MG5AMC_CULIB).so $(cu_objects_exe)
-	$(NVCC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe)
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) -lstdc++ -L$(shell dirname $(shell $(GPUCC) -print-prog-name=clang))/../../lib -lamdhip64
+else
+	$(GPUCC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe)
+endif
 endif
 
 #-------------------------------------------------------------------------------
@@ -659,7 +741,7 @@ $(BUILDDIR)/testxxx.o: testxxx_cc_ref.txt
 $(testmain): $(BUILDDIR)/testxxx.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/testxxx.o # Comment out this line to skip the C++ test of xxx functions
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/testxxx_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/testxxx_cu.o: INCFLAGS += $(GTESTINC)
 $(BUILDDIR)/testxxx_cu.o: testxxx_cc_ref.txt
@@ -672,7 +754,7 @@ $(BUILDDIR)/testmisc.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/testmisc.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/testmisc.o # Comment out this line to skip the C++ miscellaneous tests
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/testmisc_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/testmisc_cu.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/testmisc_cu.o
@@ -684,12 +766,12 @@ $(BUILDDIR)/runTest.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/runTest.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/runTest.o
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/runTest_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/runTest_cu.o: INCFLAGS += $(GTESTINC)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(testmain): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(testmain): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(testmain): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(testmain): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 else ifneq ($(shell $(CXX) --version | grep ^nvc++),) # support nvc++ #531
 $(testmain): LIBFLAGS += -L$(patsubst %bin/nvc++,%lib,$(subst ccache ,,$(CXX))) -lnvhpcatm -lnvcpumath -lnvc
 endif
@@ -713,14 +795,23 @@ $(testmain): LIBFLAGS += -lgomp
 endif
 endif
 
-ifeq ($(NVCC),) # link only runTest.o
+# Bypass std::filesystem completely to ease portability on LUMI #803
+#ifneq ($(findstring hipcc,$(GPUCC)),)
+#$(testmain): LIBFLAGS += -lstdc++fs
+#endif
+
+ifeq ($(GPUCC),) # link only runTest.o
 $(testmain): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(testmain): $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib) $(cxx_objects_exe) $(GTESTLIBS)
 	$(CXX) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) -ldl -pthread $(LIBFLAGS)
 else # link both runTest.o and runTest_cu.o
 $(testmain): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(testmain): $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) $(GTESTLIBS)
-	$(NVCC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) -lcuda
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) $(CUDATESTFLAGS) -lstdc++ -lpthread  -L$(shell dirname $(shell $(GPUCC) -print-prog-name=clang))/../../lib -lamdhip64
+else
+	$(GPUCC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) $(CUDATESTFLAGS)
+endif
 endif
 
 # Use target gtestlibs to build only googletest
@@ -829,9 +920,9 @@ ifeq ($(USECCACHE),1)
 	ccache --version | head -1
 endif
 	@echo ""
-	@echo NVCC=$(NVCC)
-ifneq ($(NVCC),)
-	$(NVCC) --version
+	@echo GPUCC=$(GPUCC)
+ifneq ($(GPUCC),)
+	$(GPUCC) --version
 endif
 	@echo ""
 	@echo CXX=$(CXX)
@@ -850,7 +941,7 @@ endif
 
 # Target: check (run the C++ test executable)
 # [NB THIS IS WHAT IS USED IN THE GITHUB CI!]
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 check: runTest cmpFcheck cmpFGcheck
 else
 check: runTest cmpFcheck
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/fbridge.cc b/epochX/cudacpp/gg_tt.mad/SubProcesses/fbridge.cc
index 2d2b36d560..22ce3f5115 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/fbridge.cc
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/fbridge.cc
@@ -1,11 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Roiser (Oct 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "Bridge.h"
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 
 extern "C"
 {
@@ -22,7 +22,7 @@ extern "C"
    * Using the same Fortran MadEvent code, linking to the hetrerogeneous library would allow access to both CPU and GPU implementations.
    * The specific heterogeneous configuration (how many GPUs, how many threads on each CPU, etc) could be loaded in CUDA/C++ from a data file.
    */
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -46,8 +46,8 @@ extern "C"
    */
   void fbridgecreate_( CppObjectInFortran** ppbridge, const int* pnevtF, const int* pnparF, const int* pnp4F )
   {
-#ifdef __CUDACC__
-    CudaRuntime::setUp();
+#ifdef MGONGPUCPP_GPUIMPL
+    GpuRuntime::setUp();
 #endif
     // (NB: CPPProcess::initProc no longer needs to be executed here because it is called in the Bridge constructor)
     // FIXME: disable OMP in Bridge when called from Fortran
@@ -65,8 +65,8 @@ extern "C"
     Bridge<FORTRANFPTYPE>* pbridge = dynamic_cast<Bridge<FORTRANFPTYPE>*>( *ppbridge );
     if( pbridge == 0 ) throw std::runtime_error( "fbridgedelete_: invalid Bridge address" );
     delete pbridge;
-#ifdef __CUDACC__
-    CudaRuntime::tearDown();
+#ifdef MGONGPUCPP_GPUIMPL
+    GpuRuntime::tearDown();
 #endif
   }
 
@@ -96,7 +96,7 @@ extern "C"
   {
     Bridge<FORTRANFPTYPE>* pbridge = dynamic_cast<Bridge<FORTRANFPTYPE>*>( *ppbridge );
     if( pbridge == 0 ) throw std::runtime_error( "fbridgesequence_: invalid Bridge address" );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Use the device/GPU implementation in the CUDA library
     // (there is also a host implementation in this library)
     pbridge->gpu_sequence( momenta, gs, rndhel, rndcol, *pchannelId, mes, selhel, selcol );
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/fsampler.cc b/epochX/cudacpp/gg_tt.mad/SubProcesses/fsampler.cc
index 2fb445372d..3743934f41 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/fsampler.cc
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/fsampler.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Feb 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "mgOnGpuConfig.h"
 
@@ -13,7 +13,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -40,7 +40,7 @@ namespace mg5amcCpu
   private:
     const int m_nevt; // The number of events in each iteration
     int m_iiter;      // The iteration counter (for random number seeding)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     HostBufferRndNumMomenta m_hstRndmom; // Memory buffers for random numbers
     HostBufferMomenta m_hstMomenta;      // Memory buffers for momenta
     HostBufferWeights m_hstWeights;      // Memory buffers for sampling weights
@@ -105,7 +105,7 @@ namespace mg5amcCpu
 
 extern "C"
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/runTest.cc b/epochX/cudacpp/gg_tt.mad/SubProcesses/runTest.cc
index d4a760a71b..de327f2321 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/runTest.cc
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/runTest.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Hageboeck (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*xxx to run only testxxx.cc tests
 //----------------------------------------------------------------------------
@@ -18,7 +18,7 @@
 #include "RandomNumberKernels.h"
 #include "epoch_process_id.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using namespace mg5amcGpu;
 #else
 using namespace mg5amcCpu;
@@ -35,7 +35,7 @@ struct CUDA_CPU_TestBase : public TestDriverBase
     : TestDriverBase( npar, refFileName ) {}
 };
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 struct CPUTest : public CUDA_CPU_TestBase
 {
   // Struct data members (process, and memory structures for random numbers, momenta, matrix elements and weights on host and device)
@@ -119,7 +119,7 @@ struct CPUTest : public CUDA_CPU_TestBase
 };
 #endif
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 struct CUDATest : public CUDA_CPU_TestBase
 {
   // Reset the device when our test goes out of scope. Note that this should happen after
@@ -128,7 +128,7 @@ struct CUDATest : public CUDA_CPU_TestBase
   {
     ~DeviceReset()
     {
-      checkCuda( cudaDeviceReset() ); // this is needed by cuda-memcheck --leak-check full
+      checkGpu( gpuDeviceReset() ); // this is needed by cuda-memcheck --leak-check full
     }
   } deviceResetter;
 
@@ -256,7 +256,7 @@ INSTANTIATE_TEST_SUITE_P( prefix, \
                           test_suite_name, \
                           testing::Values( new CUDATest( MG_EPOCH_REFERENCE_FILE_NAME ) ) );
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 MG_INSTANTIATE_TEST_SUITE_GPU( XTESTID_GPU( MG_EPOCH_PROCESS_ID ), MadgraphTest );
 #else
 MG_INSTANTIATE_TEST_SUITE_CPU( XTESTID_CPU( MG_EPOCH_PROCESS_ID ), MadgraphTest );
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/testmisc.cc b/epochX/cudacpp/gg_tt.mad/SubProcesses/testmisc.cc
index 895d6eeb56..ba9e59a8a3 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/testmisc.cc
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/testmisc.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*misc to run only testmisc.cc tests
 //----------------------------------------------------------------------------
@@ -17,7 +17,7 @@
 #include <sstream>
 #include <typeinfo>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #define TESTID( s ) s##_GPU_MISC
 #else
 #define TESTID( s ) s##_CPU_MISC
@@ -26,7 +26,7 @@
 #define XTESTID( s ) TESTID( s )
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -59,7 +59,7 @@ namespace mg5amcCpu
 
 TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testmisc )
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_tt.mad/SubProcesses/testxxx.cc b/epochX/cudacpp/gg_tt.mad/SubProcesses/testxxx.cc
index 3361fe5aa9..e5167de00c 100644
--- a/epochX/cudacpp/gg_tt.mad/SubProcesses/testxxx.cc
+++ b/epochX/cudacpp/gg_tt.mad/SubProcesses/testxxx.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Apr 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*xxx to run only testxxx.cc tests
 //----------------------------------------------------------------------------
@@ -24,7 +24,7 @@
 #include <iomanip>
 #include <iostream>
 #include <vector>
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #define TESTID( s ) s##_GPU_XXX
 #else
 #define TESTID( s ) s##_CPU_XXX
@@ -32,7 +32,7 @@
 
 #define XTESTID( s ) TESTID( s )
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -42,7 +42,7 @@ namespace mg5amcCpu
   int FPEhandlerIevt = -1;
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU): '" << FPEhandlerMessage << "' ievt=" << FPEhandlerIevt << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU neppV=" << neppV << "): '" << FPEhandlerMessage << "' ievt=" << FPEhandlerIevt << std::endl;
@@ -53,7 +53,7 @@ namespace mg5amcCpu
 
 TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -77,7 +77,7 @@ TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
   assert( nevt % neppM == 0 ); // nevt must be a multiple of neppM
   assert( nevt % neppV == 0 ); // nevt must be a multiple of neppV
   // Fill in the input momenta
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   mg5amcGpu::PinnedHostBufferMomenta hstMomenta( nevt ); // AOSOA[npagM][npar=4][np4=4][neppM]
 #else
   mg5amcCpu::HostBufferMomenta hstMomenta( nevt ); // AOSOA[npagM][npar=4][np4=4][neppM]
@@ -322,7 +322,7 @@ TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
   {
     for( int ievt = 0; ievt < nevt; ievt++ )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       using namespace mg5amcGpu;
 #else
       using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_tt.mad/bin/internal/launch_plugin.py b/epochX/cudacpp/gg_tt.mad/bin/internal/launch_plugin.py
index 3b09713e12..90af6b7053 100644
--- a/epochX/cudacpp/gg_tt.mad/bin/internal/launch_plugin.py
+++ b/epochX/cudacpp/gg_tt.mad/bin/internal/launch_plugin.py
@@ -103,7 +103,7 @@ def default_setup(self):
                        fct_mod=(self.reset_makeopts,(),{}),
                        allowed=['auto', 'none', 'sse4', 'avx2','512y','512z'])
         self.add_param('cudacpp_backend', 'CPP', include=False, hidden=False,
-                       allowed=['Fortan', 'CPP', 'CUDA'])
+                       allowed=['Fortran', 'CPP', 'CUDA'])
         self['vector_size'] = 16 # already setup in default class (just change value)
         self['aloha_flag'] = '--fast-math'
         self['matrix_flag'] = '-O3'
diff --git a/epochX/cudacpp/gg_tt.mad/src/HelAmps_sm.h b/epochX/cudacpp/gg_tt.mad/src/HelAmps_sm.h
index 55f43bb43a..add8fce575 100644
--- a/epochX/cudacpp/gg_tt.mad/src/HelAmps_sm.h
+++ b/epochX/cudacpp/gg_tt.mad/src/HelAmps_sm.h
@@ -5,7 +5,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -28,7 +28,7 @@
 //#include <iomanip>
 //#include <iostream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt.mad/src/Parameters_sm.cc b/epochX/cudacpp/gg_tt.mad/src/Parameters_sm.cc
index a9bc93ff98..c5dd6e7e4c 100644
--- a/epochX/cudacpp/gg_tt.mad/src/Parameters_sm.cc
+++ b/epochX/cudacpp/gg_tt.mad/src/Parameters_sm.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -17,7 +17,7 @@
 #include <iomanip>
 #include <iostream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using namespace mg5amcGpu;
 #else
 using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_tt.mad/src/Parameters_sm.h b/epochX/cudacpp/gg_tt.mad/src/Parameters_sm.h
index 932f123fea..5f2f4391b9 100644
--- a/epochX/cudacpp/gg_tt.mad/src/Parameters_sm.h
+++ b/epochX/cudacpp/gg_tt.mad/src/Parameters_sm.h
@@ -27,7 +27,7 @@
 #include "read_slha.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -93,7 +93,7 @@ namespace mg5amcCpu
 #include <limits>
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -217,7 +217,7 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -236,7 +236,7 @@ namespace mg5amcCpu
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wunused-variable"  // e.g. <<warning: unused variable ‘mdl_G__exp__2’ [-Wunused-variable]>>
 #pragma GCC diagnostic ignored "-Wunused-parameter" // e.g. <<warning: unused parameter ‘G’ [-Wunused-parameter]>>
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #pragma nv_diagnostic push
 #pragma nv_diag_suppress 177 // e.g. <<warning #177-D: variable "mdl_G__exp__2" was declared but never referenced>>
 #endif
@@ -263,7 +263,7 @@ namespace mg5amcCpu
       // End SM implementation - no special handling of vectors of floats as in EFT (#439)
       return out;
     }
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #pragma GCC diagnostic pop
 #pragma nv_diagnostic pop
 #endif
diff --git a/epochX/cudacpp/gg_tt.mad/src/cudacpp_src.mk b/epochX/cudacpp/gg_tt.mad/src/cudacpp_src.mk
index d4cc628aec..b2b9da5288 100644
--- a/epochX/cudacpp/gg_tt.mad/src/cudacpp_src.mk
+++ b/epochX/cudacpp/gg_tt.mad/src/cudacpp_src.mk
@@ -19,7 +19,7 @@ SHELL := /bin/bash
 #=== Configure common compiler flags for CUDA and C++
 
 INCFLAGS = -I.
-OPTFLAGS = -O3 # this ends up in CUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
+OPTFLAGS = -O3 # this ends up in GPUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
 
 #-------------------------------------------------------------------------------
 
@@ -45,13 +45,13 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure the CUDA compiler (note: NVCC is already exported including ccache)
+#=== Configure the CUDA compiler (note: GPUCC is already exported including ccache)
 
-###$(info NVCC=$(NVCC))
+###$(info GPUCC=$(GPUCC))
 
 #-------------------------------------------------------------------------------
 
-#=== Configure ccache for C++ builds (note: NVCC is already exported including ccache)
+#=== Configure ccache for C++ builds (note: GPUCC is already exported including ccache)
 
 # Enable ccache if USECCACHE=1
 ifeq ($(USECCACHE)$(shell echo $(CXX) | grep ccache),1)
@@ -92,6 +92,13 @@ endif
 ###$(info OMPFLAGS=$(OMPFLAGS))
 CXXFLAGS += $(OMPFLAGS)
 
+# Add correct flags for nvcc (-x cu) and hipcc (-x hip) for GPU code (see #810)
+ifeq ($(findstring nvcc,$(GPUCC)),nvcc)
+  GPUFLAGS += -Xcompiler -fPIC -c -x cu
+else ifeq ($(findstring hipcc,$(GPUCC)),hipcc)
+  GPUFLAGS += -fPIC -c -x hip
+endif
+
 # Set the build flags appropriate to each AVX choice (example: "make AVX=none")
 # [NB MGONGPU_PVW512 is needed because "-mprefer-vector-width=256" is not exposed in a macro]
 # [See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96476]
@@ -253,20 +260,20 @@ $(BUILDDIR)/%.o : %.cc *.h $(BUILDDIR)/.build.$(TAG)
 # Generic target and build rules: objects from CUDA compilation
 $(BUILDDIR)/%_cu.o : %.cc *.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c -x cu $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $< -o $@
 
 #-------------------------------------------------------------------------------
 
 cxx_objects=$(addprefix $(BUILDDIR)/, Parameters_sm.o read_slha.o)
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 cu_objects=$(addprefix $(BUILDDIR)/, Parameters_sm_cu.o)
 endif
 
 # Target (and build rules): common (src) library
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so : $(cxx_objects) $(cu_objects)
 	@if [ ! -d $(LIBDIR) ]; then echo "mkdir -p $(LIBDIR)"; mkdir -p $(LIBDIR); fi
-	$(NVCC) -shared -o $@ $(cxx_objects) $(cu_objects) $(LDFLAGS)
+	$(GPUCC) -shared -o $@ $(cxx_objects) $(cu_objects) $(LDFLAGS)
 else
 $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so : $(cxx_objects)
 	@if [ ! -d $(LIBDIR) ]; then echo "mkdir -p $(LIBDIR)"; mkdir -p $(LIBDIR); fi
diff --git a/epochX/cudacpp/gg_tt.mad/src/mgOnGpuConfig.h b/epochX/cudacpp/gg_tt.mad/src/mgOnGpuConfig.h
index 80032e528b..6bde4466d0 100644
--- a/epochX/cudacpp/gg_tt.mad/src/mgOnGpuConfig.h
+++ b/epochX/cudacpp/gg_tt.mad/src/mgOnGpuConfig.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jul 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUCONFIG_H
 #define MGONGPUCONFIG_H 1
@@ -10,12 +10,26 @@
 // There are two different code bases for standalone_cudacpp (without multichannel) and madevent+cudacpp (with multichannel)
 #define MGONGPU_SUPPORTS_MULTICHANNEL 1
 
+// Is this a GPU (CUDA, HIP) or CPU implementation?
+#ifdef __CUDACC__
+#define MGONGPUCPP_GPUIMPL cuda
+#elif defined __HIPCC__
+#define MGONGPUCPP_GPUIMPL hip
+#include "hip/hip_runtime.h" // needed for blockDim, blockIdx, threadIdx: better in mgOnGpuConfig.h than in GpuAbstraction.h
+#else
+#undef MGONGPUCPP_GPUIMPL
+#endif
+
 // ** NB1 Throughputs (e.g. 6.8E8) are events/sec for "./gcheck.exe -p 65536 128 12"
 // ** NB2 Baseline on b7g47n0004 fluctuates (probably depends on load on other VMs)
 
 // Choose if curand is supported for generating random numbers
+// For HIP, by default, do not use curand (common random numbers will be used instead)
 // For both CUDA and C++, by default, do not inline, but allow this macro to be set from outside with e.g. -DMGONGPU_HAS_NO_CURAND
-// (there exist CUDA installations, e.g. using the HPC package, which do not include curand - see PR #784)
+// (there exist CUDA installations, e.g. using the HPC package, which do not include curand - see PR #784 and #785)
+#if defined __HIPCC__
+#define MGONGPU_HAS_NO_CURAND 1
+#else
 //#ifdef __CUDACC__
 //#undef MGONGPU_HAS_NO_CURAND // default
 ////#define MGONGPU_HAS_NO_CURAND 1
@@ -23,6 +37,7 @@
 //#undef MGONGPU_HAS_NO_CURAND // default
 ////#define MGONGPU_HAS_NO_CURAND 1
 //#endif
+#endif
 
 // Choose floating point precision (for everything but color algebra #537)
 // If one of these macros has been set from outside with e.g. -DMGONGPU_FPTYPE_FLOAT, nothing happens (issue #167)
@@ -54,23 +69,31 @@
 //#undef MGONGPU_HARDCODE_PARAM // default
 ////#define MGONGPU_HARDCODE_PARAM 1
 
-// Complex type in c++: std::complex or cxsmpl (CHOOSE ONLY ONE)
-#ifndef __CUDACC__
-//#define MGONGPU_CPPCXTYPE_STDCOMPLEX 1 // ~8 percent slower on float, same on double (5.1E6/double, 9.4E6/float)
-#define MGONGPU_CPPCXTYPE_CXSMPL 1 // new default (5.1E6/double, 10.2E6/float)
-#endif
-
-// Complex type in cuda: thrust or cucomplex or cxsmpl (CHOOSE ONLY ONE)
+// Complex type in CUDA: thrust or cucomplex or cxsmpl (CHOOSE ONLY ONE)
+// (NB THIS IS MGONGPU_*CU*CXTYPE_xxx)
 #ifdef __CUDACC__
 #define MGONGPU_CUCXTYPE_THRUST 1 // default (~1.15E9/double, ~3.2E9/float)
 //#define MGONGPU_CUCXTYPE_CUCOMPLEX 1 // ~10 percent slower (1.03E9/double, ~2.8E9/float)
 //#define MGONGPU_CUCXTYPE_CXSMPL 1 // ~10 percent slower (1.00E9/double, ~2.9E9/float)
+
+// Complex type in HIP: cxsmpl (ONLY ONE OPTION POSSIBLE? #810)
+// (NB THIS IS MGONGPU_*HIP*CXTYPE_xxx)
+#elif defined __HIPCC__
+#define MGONGPU_HIPCXTYPE_CXSMPL 1 // default for HIP
+
+// Complex type in C++: std::complex or cxsmpl (CHOOSE ONLY ONE)
+// (NB THIS IS MGONGPU_*CPP*CXTYPE_xxx)
+#else
+//#define MGONGPU_CPPCXTYPE_STDCOMPLEX 1 // ~8 percent slower on float, same on double (5.1E6/double, 9.4E6/float)
+#define MGONGPU_CPPCXTYPE_CXSMPL 1 // new default (5.1E6/double, 10.2E6/float)
 #endif
 
-// Cuda nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
+// CUDA nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
 #ifdef __CUDACC__
-#undef MGONGPU_NSIGHT_DEBUG // default
+#undef MGONGPU_NSIGHT_DEBUG // default in CUDA
 //#define MGONGPU_NSIGHT_DEBUG 1
+#else
+#undef MGONGPU_NSIGHT_DEBUG // only option in HIP or C++
 #endif
 
 // SANITY CHECKS (floating point precision for everything but color algebra #537)
@@ -86,17 +109,21 @@
 #error You cannot use double precision for color algebra and single precision elsewhere
 #endif
 
-// SANITY CHECKS (c++ complex number implementation)
-#ifndef __CUDACC__
-#if defined MGONGPU_CPPCXTYPE_STDCOMPLEX and defined MGONGPU_CPPCXTYPE_CXSMPL
-#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CPPCXTYPE_STDCOMPLEX or MGONGPU_CPPCXTYPE_CXSMPL
+// SANITY CHECKS (CUDA complex number implementation)
+#ifdef __CUDACC__
+#if defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CUCOMPLEX
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX for CUDA
+#elif defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CXSMPL for CUDA
+#elif defined MGONGPU_CUCXTYPE_CUCOMPLEX and defined MGONGPU_CUCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE OF MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL for CUDA
 #endif
 #endif
 
-// SANITY CHECKS (cuda complex number implementation)
-#ifdef __CUDACC__
-#if defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CUCOMPLEX and defined MGONGPU_CUCXTYPE_CXSMPL
-#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL
+// SANITY CHECKS (C++ complex number implementation)
+#ifndef MGONGPUCPP_GPUIMPL
+#if defined MGONGPU_CPPCXTYPE_STDCOMPLEX and defined MGONGPU_CPPCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CPPCXTYPE_STDCOMPLEX or MGONGPU_CPPCXTYPE_CXSMPL for C++
 #endif
 #endif
 
@@ -134,7 +161,7 @@ namespace mgOnGpu
   // Alignment requirement for using reinterpret_cast with SIMD vectorized code
   // (using reinterpret_cast with non aligned memory may lead to segmentation faults!)
   // Only needed for C++ code but can be enforced also in NVCC builds of C++ code using CUDA>=11.2 and C++17 (#318, #319, #333)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   constexpr int cppAlign = 64; // alignment requirement for SIMD vectorization (64-byte i.e. 512-bit)
 #endif
 
@@ -145,7 +172,7 @@ using mgOnGpu::fptype;
 using mgOnGpu::fptype2;
 
 // C++ SIMD vectorization width (this will be used to set neppV)
-#ifdef __CUDACC__ // CUDA implementation has no SIMD
+#ifdef MGONGPUCPP_GPUIMPL // CUDA and HIP implementations have no SIMD
 #undef MGONGPU_CPPSIMD
 #elif defined __AVX512VL__ && defined MGONGPU_PVW512 // C++ "512z" AVX512 with 512 width (512-bit ie 64-byte): 8 (DOUBLE) or 16 (FLOAT)
 #ifdef MGONGPU_FPTYPE_DOUBLE
@@ -175,9 +202,9 @@ using mgOnGpu::fptype2;
 #undef MGONGPU_CPPSIMD
 #endif
 
-// Cuda nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
+// CUDA nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
 // Arguments (not used so far): text is __FUNCTION__, code is 0 (start) or 1 (end)
-#if defined __CUDACC__ && defined MGONGPU_NSIGHT_DEBUG /* clang-format off */
+#if defined __CUDA__ && defined MGONGPU_NSIGHT_DEBUG /* clang-format off */
 #define mgDebugDeclare() __shared__ float mgDebugCounter[mgOnGpu::ntpbMAX];
 #define mgDebugInitialise() { mgDebugCounter[threadIdx.x] = 0; }
 #define mgDebug( code, text ) { mgDebugCounter[threadIdx.x] += 1; }
@@ -189,8 +216,8 @@ using mgOnGpu::fptype2;
 #define mgDebugFinalise() { /*noop*/ }
 #endif /* clang-format on */
 
-// Define empty CUDA declaration specifiers for C++
-#ifndef __CUDACC__
+// Define empty CUDA/HIP declaration specifiers for C++
+#ifndef MGONGPUCPP_GPUIMPL
 #define __global__
 #define __host__
 #define __device__
diff --git a/epochX/cudacpp/gg_tt.mad/src/mgOnGpuCxtypes.h b/epochX/cudacpp/gg_tt.mad/src/mgOnGpuCxtypes.h
index ca9a9f00c0..7ede1dbfae 100644
--- a/epochX/cudacpp/gg_tt.mad/src/mgOnGpuCxtypes.h
+++ b/epochX/cudacpp/gg_tt.mad/src/mgOnGpuCxtypes.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022, based on earlier work by D. Smith) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUCXTYPES_H
 #define MGONGPUCXTYPES_H 1
@@ -19,7 +19,7 @@
 #include <complex>
 
 // Complex type in cuda: thrust or cucomplex or cxsmpl
-#ifdef __CUDACC__
+#ifdef __CUDACC__ // this must be __CUDAC__ (not MGONGPUCPP_GPUIMPL)
 #if defined MGONGPU_CUCXTYPE_THRUST
 #pragma clang diagnostic push
 #pragma clang diagnostic ignored "-Wtautological-compare" // for icpx2021/clang13 (https://stackoverflow.com/a/15864661)
@@ -30,8 +30,13 @@
 #elif not defined MGONGPU_CUCXTYPE_CXSMPL
 #error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL
 #endif
+// Complex type in HIP: cxsmpl
+#elif defined __HIPCC__
+#if not defined MGONGPU_HIPCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_HIPCXTYPE_CXSMPL
+#endif
 #else
-// Complex type in c++: std::complex or cxsmpl
+// Complex type in c++ or HIP: std::complex or cxsmpl
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
 #include <cmath>
 #elif not defined MGONGPU_CPPCXTYPE_CXSMPL
@@ -82,7 +87,7 @@ namespace mgOnGpu /* clang-format off */
 using mgOnGpu::cxsmpl;
 
 // Printout to stream for user defined types
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -92,7 +97,7 @@ namespace mg5amcCpu
   inline __host__ std::ostream&
   operator<<( std::ostream& out, const cxsmpl<FP>& c )
   {
-    out << std::complex( c.real(), c.imag() );
+    out << std::complex<FP>( c.real(), c.imag() );
     return out;
   }
 
@@ -215,14 +220,14 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
   // --- Type definitions (complex type: cxtype)
-#ifdef __CUDACC__ // cuda
+#ifdef __CUDACC__ // this must be __CUDAC__ (not MGONGPUCPP_GPUIMPL)
 #if defined MGONGPU_CUCXTYPE_THRUST
   typedef thrust::complex<fptype> cxtype;
 #elif defined MGONGPU_CUCXTYPE_CUCOMPLEX
@@ -255,13 +260,13 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
-#if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+#if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
 
   //------------------------------
   // CUDA or C++ - using cxsmpl
@@ -303,11 +308,11 @@ namespace mg5amcCpu
     return cxmake( c.real(), c.imag() );
   }
 
-#endif // #if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+#endif // #if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
 
   //==========================================================================
 
-#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_THRUST // cuda + thrust
+#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_THRUST // cuda + thrust (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // CUDA - using thrust::complex
@@ -347,7 +352,7 @@ namespace mg5amcCpu
 
   //==========================================================================
 
-#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_CUCOMPLEX // cuda + cucomplex
+#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_CUCOMPLEX // cuda + cucomplex (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // CUDA - using cuComplex
@@ -566,7 +571,7 @@ namespace mg5amcCpu
 
   //==========================================================================
 
-#if not defined __CUDACC__ and defined MGONGPU_CPPCXTYPE_STDCOMPLEX // c++ + stdcomplex
+#if not defined __CUDACC__ and defined MGONGPU_CPPCXTYPE_STDCOMPLEX // c++/hip + stdcomplex (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // C++ - using std::complex
@@ -633,7 +638,7 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt.mad/src/mgOnGpuFptypes.h b/epochX/cudacpp/gg_tt.mad/src/mgOnGpuFptypes.h
index 905c97d700..fa3a02664b 100644
--- a/epochX/cudacpp/gg_tt.mad/src/mgOnGpuFptypes.h
+++ b/epochX/cudacpp/gg_tt.mad/src/mgOnGpuFptypes.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUFPTYPES_H
 #define MGONGPUFPTYPES_H 1
@@ -12,7 +12,7 @@
 #include <cmath>
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL // cuda
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -20,7 +20,7 @@ namespace mg5amcCpu
 {
   //==========================================================================
 
-#ifdef __CUDACC__ // cuda
+#ifdef MGONGPUCPP_GPUIMPL // cuda
 
   //------------------------------
   // Floating point types - Cuda
@@ -64,11 +64,11 @@ namespace mg5amcCpu
 #endif
   }
 
-#endif // #ifdef __CUDACC__
+#endif // #ifdef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
   //------------------------------
   // Floating point types - C++
@@ -92,7 +92,7 @@ namespace mg5amcCpu
     return std::sqrt( f );
   }
 
-#endif // #ifndef __CUDACC__
+#endif // #ifndef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
diff --git a/epochX/cudacpp/gg_tt.mad/src/mgOnGpuVectors.h b/epochX/cudacpp/gg_tt.mad/src/mgOnGpuVectors.h
index e1299ba81e..cdae04326b 100644
--- a/epochX/cudacpp/gg_tt.mad/src/mgOnGpuVectors.h
+++ b/epochX/cudacpp/gg_tt.mad/src/mgOnGpuVectors.h
@@ -32,7 +32,7 @@
 #endif
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -131,7 +131,7 @@ namespace mg5amcCpu
 #endif
 #endif
 
-#else // i.e #ifndef MGONGPU_CPPSIMD (this includes #ifdef __CUDACC__)
+#else // i.e #ifndef MGONGPU_CPPSIMD (this includes #ifdef MGONGPUCPP_GPUIMPL)
 
   const int neppV = 1;
 
@@ -153,13 +153,13 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
   // Printout to stream for user defined types
 
@@ -805,11 +805,11 @@ namespace mg5amcCpu
 
 #endif // #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
 
-#endif // #ifndef __CUDACC__
+#endif // #ifndef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
   //------------------------------
   // Vector types - CUDA
@@ -853,12 +853,12 @@ namespace mg5amcCpu
     return mask;
   }
 
-#endif // #ifdef __CUDACC__
+#endif // #ifdef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
   // Scalar-or-vector types: scalar in CUDA, vector or scalar in C++
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   typedef bool bool_sv;
   typedef fptype fptype_sv;
   typedef fptype2 fptype2_sv;
@@ -879,7 +879,7 @@ namespace mg5amcCpu
 #endif
 
   // Scalar-or-vector zeros: scalar in CUDA, vector or scalar in C++
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   inline __host__ __device__ cxtype cxzero_sv(){ return cxtype( 0, 0 ); }
 #elif defined MGONGPU_CPPSIMD
   inline cxtype_v cxzero_sv() { return cxtype_v(); } // RRRR=0000 IIII=0000
diff --git a/epochX/cudacpp/gg_tt.mad/src/rambo.h b/epochX/cudacpp/gg_tt.mad/src/rambo.h
index e02ea52496..cd7e1008ea 100644
--- a/epochX/cudacpp/gg_tt.mad/src/rambo.h
+++ b/epochX/cudacpp/gg_tt.mad/src/rambo.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -18,7 +18,7 @@
 #include <iostream>
 
 // Simplified rambo version for 2 to N (with N>=2) processes with massless particles
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +83,7 @@ namespace mg5amcCpu
       static bool first = true;
       if( first )
       {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
         if constexpr( M_ACCESS::isOnDevice() ) // avoid
         {
           const int ievt0 = 0;
@@ -166,7 +166,7 @@ namespace mg5amcCpu
     wt = po2log;
     if( nparf != 2 ) wt = ( 2. * nparf - 4. ) * log( energy ) + z[nparf - 1];
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     // issue warnings if weight is too small or too large
     static int iwarn[5] = { 0, 0, 0, 0, 0 };
     if( wt < -180. )
diff --git a/epochX/cudacpp/gg_tt.mad/src/read_slha.cc b/epochX/cudacpp/gg_tt.mad/src/read_slha.cc
index 055b19a779..f8e46f2e66 100644
--- a/epochX/cudacpp/gg_tt.mad/src/read_slha.cc
+++ b/epochX/cudacpp/gg_tt.mad/src/read_slha.cc
@@ -11,7 +11,11 @@
 
 #include <algorithm>
 #include <cstdlib>
-#include <filesystem>
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
 #include <fstream>
 #include <iostream>
 
@@ -60,7 +64,14 @@ SLHAReader::read_slha_file( std::string file_name, bool verbose )
     {
       std::cout << "WARNING! Card file '" << file_name << "' does not exist:"
                 << " look for the file in directory $" << envpath << "='" << getenv( envpath ) << "'" << std::endl;
+      /*
+#ifdef __HIPCC__
+      const std::string file_name2 = std::experimental::filesystem::path( getenv( envpath ) ) / std::experimental::filesystem::path( file_name ).filename();
+#else
       const std::string file_name2 = std::filesystem::path( getenv( envpath ) ) / std::filesystem::path( file_name ).filename();
+#endif
+      */
+      const std::string file_name2 = std::string( getenv( envpath ) ) + "/" + file_name; // bypass std::filesystem #803
       param_card.open( file_name2.c_str(), std::ifstream::in );
       if( param_card.good() )
       {
diff --git a/epochX/cudacpp/gg_tt.sa/CODEGEN_cudacpp_gg_tt_log.txt b/epochX/cudacpp/gg_tt.sa/CODEGEN_cudacpp_gg_tt_log.txt
index 0db09949ad..9d7cc87630 100644
--- a/epochX/cudacpp/gg_tt.sa/CODEGEN_cudacpp_gg_tt_log.txt
+++ b/epochX/cudacpp/gg_tt.sa/CODEGEN_cudacpp_gg_tt_log.txt
@@ -52,7 +52,7 @@ Note that you can still compile and run aMC@NLO with the built-in PDFs
 
 Using default text editor "vi". Set another one in ./input/mg5_configuration.txt
 Using default eps viewer "evince". Set another one in ./input/mg5_configuration.txt
-No valid web browser found. Please set in ./input/mg5_configuration.txt
+Using default web browser "firefox". Set another one in ./input/mg5_configuration.txt
 import /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_cudacpp_gg_tt.mg
 The import format was not given, so we guess it as command
 set stdout_level DEBUG
@@ -62,7 +62,7 @@ generate g g > t t~
 No model currently active, so we import the Standard Model
 INFO: load particles 
 INFO: load vertices 
-DEBUG: model prefixing  takes 0.005459308624267578 
+DEBUG: model prefixing  takes 0.005457878112792969 
 INFO: Restrict model sm with file models/sm/restrict_default.dat . 
 DEBUG: Simplifying conditional expressions 
 DEBUG: remove interactions: u s w+ at order: QED=1 
@@ -180,7 +180,7 @@ Generated helas calls for 1 subprocesses (3 diagrams) in 0.006 s
 ALOHA: aloha starts to compute helicity amplitudes
 ALOHA: aloha creates VVV1 set of routines with options: P0
 ALOHA: aloha creates FFV1 routines
-ALOHA: aloha creates 2 routines in  0.146 s
+ALOHA: aloha creates 2 routines in  0.142 s
 <class 'aloha.create_aloha.AbstractRoutine'> VVV1
 <class 'aloha.create_aloha.AbstractRoutine'> FFV1
 <class 'aloha.create_aloha.AbstractRoutine'> FFV1
@@ -195,7 +195,7 @@ INFO: Created files Parameters_sm.h and Parameters_sm.cc in directory
 INFO: /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_cudacpp_gg_tt/src/. and /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_cudacpp_gg_tt/src/. 
 quit
 
-real	0m0.545s
-user	0m0.487s
-sys	0m0.049s
+real	0m0.546s
+user	0m0.467s
+sys	0m0.054s
 Code generation completed in 1 seconds
diff --git a/epochX/cudacpp/gg_tt.sa/COPYRIGHT b/epochX/cudacpp/gg_tt.sa/COPYRIGHT
index a134b5fef9..9036d9260a 100644
--- a/epochX/cudacpp/gg_tt.sa/COPYRIGHT
+++ b/epochX/cudacpp/gg_tt.sa/COPYRIGHT
@@ -15,6 +15,7 @@ The full development team currently includes the following authors :
   Stephan Hageboeck (CERN)
   Olivier Mattelaer (Universite Catholique de Louvain, original author)
   Stefan Roiser (CERN, original author)
+  Jorgen Teig (CERN)
   Andrea Valassi (CERN, original author)
   Zenny Wettersten (CERN)
 See https://github.com/madgraph5/madgraph4gpu for more details. For the full
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/Bridge.h b/epochX/cudacpp/gg_tt.sa/SubProcesses/Bridge.h
index bf8b5e024d..f9ed70dfde 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/Bridge.h
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/Bridge.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Roiser (Nov 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef BRIDGE_H
 #define BRIDGE_H 1
@@ -14,16 +14,23 @@
 #include "MemoryAccessMomenta.h"  // for MemoryAccessMomenta::neppM
 #include "MemoryBuffers.h"        // for HostBufferMomenta, DeviceBufferMomenta etc
 
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
+
+#include <sys/stat.h> // bypass std::filesystem #803
+
 #include <algorithm>
 #include <cassert>
 #include <cmath>
 #include <cstring>
-#include <filesystem>
 #include <iostream>
 #include <memory>
 #include <type_traits>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +90,7 @@ namespace mg5amcCpu
     Bridge& operator=( const Bridge& ) = delete;
     Bridge& operator=( Bridge&& ) = delete;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     /**
      * Set the gpublocks and gputhreads for the gpusequence - throws if evnt != gpublocks*gputhreads
      * (this is needed for BridgeKernel tests rather than for actual production use in Fortran)
@@ -150,7 +157,7 @@ namespace mg5amcCpu
     unsigned int m_nevt; // number of events
     int m_nGoodHel;      // the number of good helicities (-1 initially when they have not yet been calculated)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     int m_gputhreads; // number of gpu threads (default set from number of events, can be modified)
     int m_gpublocks;  // number of gpu blocks (default set from number of events, can be modified)
     DeviceBuffer<FORTRANFPTYPE, sizePerEventMomenta> m_devMomentaF;
@@ -187,12 +194,12 @@ namespace mg5amcCpu
   // Forward declare transposition methods
   //
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
   template<typename Tin, typename Tout>
   __global__ void dev_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt );
 
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
 
   template<typename Tin, typename Tout>
   void hst_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt );
@@ -209,7 +216,7 @@ namespace mg5amcCpu
   Bridge<FORTRANFPTYPE>::Bridge( unsigned int nevtF, unsigned int nparF, unsigned int np4F )
     : m_nevt( nevtF )
     , m_nGoodHel( -1 )
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     , m_gputhreads( 256 )                  // default number of gpu threads
     , m_gpublocks( m_nevt / m_gputhreads ) // this ensures m_nevt <= m_gpublocks*m_gputhreads
     , m_devMomentaF( m_nevt )
@@ -233,7 +240,7 @@ namespace mg5amcCpu
   {
     if( nparF != CPPProcess::npar ) throw std::runtime_error( "Bridge constructor: npar mismatch" );
     if( np4F != CPPProcess::np4 ) throw std::runtime_error( "Bridge constructor: np4 mismatch" );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( ( m_nevt < s_gputhreadsmin ) || ( m_nevt % s_gputhreadsmin != 0 ) )
       throw std::runtime_error( "Bridge constructor: nevt should be a multiple of " + std::to_string( s_gputhreadsmin ) );
     while( m_nevt != m_gpublocks * m_gputhreads )
@@ -249,20 +256,28 @@ namespace mg5amcCpu
 #else
     std::cout << "WARNING! Instantiate host Bridge (nevt=" << m_nevt << ")" << std::endl;
     m_pmek.reset( new MatrixElementKernelHost( m_hstMomentaC, m_hstGs, m_hstRndHel, m_hstRndCol, m_hstMEs, m_hstSelHel, m_hstSelCol, m_nevt ) );
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
     // Create a process object, read param card and set parameters
     // FIXME: the process instance can happily go out of scope because it is only needed to read parameters?
     // FIXME: the CPPProcess should really be a singleton? what if fbridgecreate is called from several Fortran threads?
     CPPProcess process( /*verbose=*/false );
     std::string paramCard = "../../Cards/param_card.dat";
-    if( !std::filesystem::exists( paramCard ) )
-    {
-      paramCard = "../" + paramCard;
-    }
+    /*
+#ifdef __HIPCC__
+    if( !std::experimental::filesystem::exists( paramCard ) ) paramCard = "../" + paramCard;
+#else
+    if( !std::filesystem::exists( paramCard ) ) paramCard = "../" + paramCard;
+#endif
+    */
+    //struct stat dummybuffer; // bypass std::filesystem #803
+    //if( !( stat( paramCard.c_str(), &dummyBuffer ) == 0 ) ) paramCard = "../" + paramCard; //
+    auto fileExists = []( std::string& fileName )
+    { struct stat buffer; return stat( fileName.c_str(), &buffer ) == 0; };
+    if( !fileExists( paramCard ) ) paramCard = "../" + paramCard; // bypass std::filesystem #803
     process.initProc( paramCard );
   }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::set_gpugrid( const int gpublocks, const int gputhreads )
   {
@@ -276,7 +291,7 @@ namespace mg5amcCpu
   }
 #endif
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::gpu_sequence( const FORTRANFPTYPE* momenta,
                                             const FORTRANFPTYPE* gs,
@@ -291,14 +306,14 @@ namespace mg5amcCpu
     constexpr int neppM = MemoryAccessMomenta::neppM;
     if constexpr( neppM == 1 && std::is_same_v<FORTRANFPTYPE, fptype> )
     {
-      checkCuda( cudaMemcpy( m_devMomentaC.data(), momenta, m_devMomentaC.bytes(), cudaMemcpyHostToDevice ) );
+      gpuMemcpy( m_devMomentaC.data(), momenta, m_devMomentaC.bytes(), gpuMemcpyHostToDevice );
     }
     else
     {
-      checkCuda( cudaMemcpy( m_devMomentaF.data(), momenta, m_devMomentaF.bytes(), cudaMemcpyHostToDevice ) );
+      gpuMemcpy( m_devMomentaF.data(), momenta, m_devMomentaF.bytes(), gpuMemcpyHostToDevice );
       const int thrPerEvt = CPPProcess::npar * CPPProcess::np4; // AV: transpose alg does 1 element per thread (NOT 1 event per thread)
       //const int thrPerEvt = 1; // AV: try new alg with 1 event per thread... this seems slower
-      dev_transposeMomentaF2C<<<m_gpublocks * thrPerEvt, m_gputhreads>>>( m_devMomentaF.data(), m_devMomentaC.data(), m_nevt );
+      gpuLaunchKernel( dev_transposeMomentaF2C, m_gpublocks * thrPerEvt, m_gputhreads, m_devMomentaF.data(), m_devMomentaC.data(), m_nevt );
     }
     if constexpr( std::is_same_v<FORTRANFPTYPE, fptype> )
     {
@@ -341,7 +356,7 @@ namespace mg5amcCpu
   }
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::cpu_sequence( const FORTRANFPTYPE* momenta,
                                             const FORTRANFPTYPE* gs,
@@ -396,7 +411,7 @@ namespace mg5amcCpu
   // - C++ array: momenta[npagM][npar][np4][neppM] with nevt=npagM*neppM (AOSOA)
   //
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename Tin, typename Tout>
   __global__ void dev_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt )
   {
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/BridgeKernels.cc b/epochX/cudacpp/gg_tt.sa/SubProcesses/BridgeKernels.cc
index d58066c9c1..eaf4037a24 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/BridgeKernels.cc
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/BridgeKernels.cc
@@ -1,17 +1,18 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "BridgeKernels.h"
 
+#include "GpuAbstraction.h"
 #include "MemoryAccessMomenta.h"
 
 #include <sstream>
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -45,7 +46,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu
 {
 
@@ -96,7 +97,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/BridgeKernels.h b/epochX/cudacpp/gg_tt.sa/SubProcesses/BridgeKernels.h
index 15eb4bff4d..3efef8ce97 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/BridgeKernels.h
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/BridgeKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef BRIDGEKERNELS_H
 #define BRIDGEKERNELS_H 1
@@ -12,7 +12,7 @@
 #include "MatrixElementKernels.h"
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -49,7 +49,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A Bridge wrapper class encapsulating matrix element calculations on a CPU host
   class BridgeKernelHost final : public BridgeKernelBase
   {
@@ -89,7 +89,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A Bridge wrapper class encapsulating matrix element calculations on a GPU device
   class BridgeKernelDevice : public BridgeKernelBase
   {
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/CommonRandomNumberKernel.cc b/epochX/cudacpp/gg_tt.sa/SubProcesses/CommonRandomNumberKernel.cc
index 985b39f576..010bc4cbd0 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/CommonRandomNumberKernel.cc
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/CommonRandomNumberKernel.cc
@@ -1,15 +1,16 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "CommonRandomNumbers.h"
+#include "GpuAbstraction.h"
 #include "MemoryBuffers.h"
 #include "RandomNumberKernels.h"
 
 #include <cassert>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/CrossSectionKernels.cc b/epochX/cudacpp/gg_tt.sa/SubProcesses/CrossSectionKernels.cc
index 0b355a3c8d..c15b39844d 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/CrossSectionKernels.cc
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/CrossSectionKernels.cc
@@ -1,10 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "CrossSectionKernels.h"
 
+#include "GpuAbstraction.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessWeights.h"
 #include "MemoryBuffers.h"
@@ -77,7 +78,7 @@ debug_me_is_abnormal( const fptype& me, size_t ievtALL )
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -185,7 +186,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/CrossSectionKernels.h b/epochX/cudacpp/gg_tt.sa/SubProcesses/CrossSectionKernels.h
index 7933ca4bbf..4d9659e04e 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/CrossSectionKernels.h
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/CrossSectionKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef CROSSSECTIONKERNELS_H
 #define CROSSSECTIONKERNELS_H 1
@@ -13,7 +13,7 @@
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -96,7 +96,7 @@ namespace mg5amcCpu
   //--------------------------------------------------------------------------
 
   /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating the calculation of event statistics on a GPU device
   class CrossSectionKernelDevice : public CrossSectionKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/CudaRuntime.h b/epochX/cudacpp/gg_tt.sa/SubProcesses/CudaRuntime.h
deleted file mode 100644
index 64ce52f4b3..0000000000
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/CudaRuntime.h
+++ /dev/null
@@ -1,85 +0,0 @@
-// Copyright (C) 2020-2023 CERN and UCLouvain.
-// Licensed under the GNU Lesser General Public License (version 3 or later).
-// Created by: S. Roiser (Jul 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
-
-#ifndef MG5AMC_CUDARUNTIME_H
-#define MG5AMC_CUDARUNTIME_H 1
-
-// MG5AMC on GPU uses the CUDA runtime API, not the lower level CUDA driver API
-// See https://docs.nvidia.com/cuda/cuda-runtime-api/driver-vs-runtime-api.html#driver-vs-runtime-api
-
-#include <cassert>
-#include <iostream>
-
-//--------------------------------------------------------------------------
-
-// See https://stackoverflow.com/a/14038590
-#ifdef __CUDACC__ /* clang-format off */
-#define checkCuda( code ) { assertCuda( code, __FILE__, __LINE__ ); }
-inline void assertCuda( cudaError_t code, const char* file, int line, bool abort = true )
-{
-  if( code != cudaSuccess )
-  {
-    printf( "ERROR! assertCuda: '%s' (%d) in %s:%d\n", cudaGetErrorString( code ), code, file, line );
-    if( abort ) assert( code == cudaSuccess );
-  }
-}
-#endif /* clang-format on */
-
-//--------------------------------------------------------------------------
-
-#ifdef __CUDACC__
-namespace mg5amcGpu
-{
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  // *** FIXME! This will all need to be designed differently when going to multi-GPU nodes! ***
-  struct CudaRuntime final
-  {
-    CudaRuntime( const bool debug = true )
-      : m_debug( debug ) { setUp( m_debug ); }
-    ~CudaRuntime() { tearDown( m_debug ); }
-    CudaRuntime( const CudaRuntime& ) = delete;
-    CudaRuntime( CudaRuntime&& ) = delete;
-    CudaRuntime& operator=( const CudaRuntime& ) = delete;
-    CudaRuntime& operator=( CudaRuntime&& ) = delete;
-    bool m_debug;
-
-    // Set up CUDA application
-    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
-    // Calling cudaSetDevice on startup is useful to properly book-keep the time spent in CUDA initialization
-    static void setUp( const bool debug = true )
-    {
-      // ** NB: it is useful to call cudaSetDevice, or cudaFree, to properly book-keep the time spent in CUDA initialization
-      // ** NB: otherwise, the first CUDA operation (eg a cudaMemcpyToSymbol in CPPProcess ctor) appears to take much longer!
-      /*
-      // [We initially added cudaFree(0) to "ease profile analysis" only because it shows up as a big recognizable block!]
-      // No explicit initialization is needed: https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#initialization
-      // It is not clear what cudaFree(0) does at all: https://stackoverflow.com/questions/69967813/
-      if ( debug ) std::cout << "__CudaRuntime: calling cudaFree(0)" << std::endl;
-      checkCuda( cudaFree( 0 ) ); // SLOW!
-      */
-      // Replace cudaFree(0) by cudaSetDevice(0), even if it is not really needed either
-      // (but see https://developer.nvidia.com/blog/cuda-pro-tip-always-set-current-device-avoid-multithreading-bugs)
-      if( debug ) std::cout << "__CudaRuntime: calling cudaSetDevice(0)" << std::endl;
-      checkCuda( cudaSetDevice( 0 ) ); // SLOW!
-    }
-
-    // Tear down CUDA application (call cudaDeviceReset)
-    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
-    // Calling cudaDeviceReset on shutdown is only needed for checking memory leaks in cuda-memcheck
-    // See https://docs.nvidia.com/cuda/cuda-memcheck/index.html#leak-checking
-    static void tearDown( const bool debug = true )
-    {
-      if( debug ) std::cout << "__CudaRuntime: calling cudaDeviceReset()" << std::endl;
-      checkCuda( cudaDeviceReset() );
-    }
-  };
-
-}
-#endif
-
-//--------------------------------------------------------------------------
-
-#endif // MG5AMC_CUDARUNTIME_H
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/CurandRandomNumberKernel.cc b/epochX/cudacpp/gg_tt.sa/SubProcesses/CurandRandomNumberKernel.cc
index eb56333b03..08a16f6f2c 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/CurandRandomNumberKernel.cc
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/CurandRandomNumberKernel.cc
@@ -1,9 +1,9 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "MemoryBuffers.h"
 #include "RandomNumberKernels.h"
 
@@ -22,7 +22,7 @@ inline void assertCurand( curandStatus_t code, const char *file, int line, bool
 }
 #endif /* clang-format on */
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -36,7 +36,7 @@ namespace mg5amcCpu
   {
     if( m_isOnDevice )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if( !m_rnarray.isOnDevice() )
         throw std::runtime_error( "CurandRandomNumberKernel on device with a host random number array" );
 #else
@@ -114,7 +114,7 @@ namespace mg5amcCpu
     /*
     printf( "\nCurandRandomNumberKernel::generateRnarray size = %d\n", (int)m_rnarray.size() );
     fptype* data = m_rnarray.data();
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( m_rnarray.isOnDevice() )
     {
       data = new fptype[m_rnarray.size()]();
@@ -123,7 +123,7 @@ namespace mg5amcCpu
 #endif
     for( int i = 0; i < ( (int)m_rnarray.size() / 4 ); i++ )
       printf( "[%4d] %f %f %f %f\n", i * 4, data[i * 4], data[i * 4 + 2], data[i * 4 + 2], data[i * 4 + 3] );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( m_rnarray.isOnDevice() ) delete[] data;
 #endif
     */
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/EventStatistics.h b/epochX/cudacpp/gg_tt.sa/SubProcesses/EventStatistics.h
index 48b51e0a49..b425a5bade 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/EventStatistics.h
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/EventStatistics.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef EventStatistics_H
 #define EventStatistics_H 1
@@ -16,7 +16,7 @@
 #include <limits>
 #include <string>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/GpuAbstraction.h b/epochX/cudacpp/gg_tt.sa/SubProcesses/GpuAbstraction.h
new file mode 100644
index 0000000000..9c467b1e04
--- /dev/null
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/GpuAbstraction.h
@@ -0,0 +1,69 @@
+// Copyright (C) 2020-2023 CERN and UCLouvain.
+// Licensed under the GNU Lesser General Public License (version 3 or later).
+// Created by: J. Teig (Jul 2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+
+#ifndef MG5AMC_GPUABSTRACTION_H
+#define MG5AMC_GPUABSTRACTION_H 1
+
+#include <cassert>
+
+//--------------------------------------------------------------------------
+
+#ifdef __CUDACC__
+
+#define gpuError_t cudaError_t
+#define gpuPeekAtLastError cudaPeekAtLastError
+#define gpuGetErrorString cudaGetErrorString
+#define gpuSuccess cudaSuccess
+
+#define gpuMallocHost( ptr, size ) checkGpu( cudaMallocHost( ptr, size ) )
+#define gpuMalloc( ptr, size ) checkGpu( cudaMalloc( ptr, size ) )
+
+#define gpuMemcpy( dstData, srcData, srcBytes, func ) checkGpu( cudaMemcpy( dstData, srcData, srcBytes, func ) )
+#define gpuMemcpyHostToDevice cudaMemcpyHostToDevice
+#define gpuMemcpyDeviceToHost cudaMemcpyDeviceToHost
+#define gpuMemcpyToSymbol( type1, type2, size ) checkGpu( cudaMemcpyToSymbol( type1, type2, size ) )
+
+#define gpuFree( ptr ) checkGpu( cudaFree( ptr ) )
+#define gpuFreeHost( ptr ) checkGpu( cudaFreeHost( ptr ) )
+
+#define gpuSetDevice cudaSetDevice
+#define gpuDeviceSynchronize cudaDeviceSynchronize
+#define gpuDeviceReset cudaDeviceReset
+
+#define gpuLaunchKernel( kernel, blocks, threads, ... ) kernel<<<blocks, threads>>>( __VA_ARGS__ )
+#define gpuLaunchKernelSharedMem( kernel, blocks, threads, sharedMem, ... ) kernel<<<blocks, threads, sharedMem>>>( __VA_ARGS__ )
+
+//--------------------------------------------------------------------------
+
+#elif defined __HIPCC__
+
+#define gpuError_t hipError_t
+#define gpuPeekAtLastError hipPeekAtLastError
+#define gpuGetErrorString hipGetErrorString
+#define gpuSuccess hipSuccess
+
+#define gpuMallocHost( ptr, size ) checkGpu( hipHostMalloc( ptr, size ) ) // HostMalloc better
+#define gpuMalloc( ptr, size ) checkGpu( hipMalloc( ptr, size ) )
+
+#define gpuMemcpy( dstData, srcData, srcBytes, func ) checkGpu( hipMemcpy( dstData, srcData, srcBytes, func ) )
+#define gpuMemcpyHostToDevice hipMemcpyHostToDevice
+#define gpuMemcpyDeviceToHost hipMemcpyDeviceToHost
+#define gpuMemcpyToSymbol( type1, type2, size ) checkGpu( hipMemcpyToSymbol( type1, type2, size ) )
+
+#define gpuFree( ptr ) checkGpu( hipFree( ptr ) )
+#define gpuFreeHost( ptr ) checkGpu( hipHostFree( ptr ) )
+
+#define gpuSetDevice hipSetDevice
+#define gpuDeviceSynchronize hipDeviceSynchronize
+#define gpuDeviceReset hipDeviceReset
+
+#define gpuLaunchKernel( kernel, blocks, threads, ... ) kernel<<<blocks, threads>>>( __VA_ARGS__ )
+#define gpuLaunchKernelSharedMem( kernel, blocks, threads, sharedMem, ... ) kernel<<<blocks, threads, sharedMem>>>( __VA_ARGS__ )
+
+//--------------------------------------------------------------------------
+
+#endif
+
+#endif // MG5AMC_GPUABSTRACTION_H
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/GpuRuntime.h b/epochX/cudacpp/gg_tt.sa/SubProcesses/GpuRuntime.h
new file mode 100644
index 0000000000..93579ef08b
--- /dev/null
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/GpuRuntime.h
@@ -0,0 +1,85 @@
+// Copyright (C) 2020-2023 CERN and UCLouvain.
+// Licensed under the GNU Lesser General Public License (version 3 or later).
+// Created by: J. Teig (Jun 2023, based on earlier work by S. Roiser) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+
+#ifndef MG5AMC_GPURUNTIME_H
+#define MG5AMC_GPURUNTIME_H 1
+
+// MG5AMC on GPU uses the CUDA runtime API, not the lower level CUDA driver API
+// See https://docs.nvidia.com/cuda/cuda-runtime-api/driver-vs-runtime-api.html#driver-vs-runtime-api
+
+#include "GpuAbstraction.h"
+
+#include <iostream>
+
+//--------------------------------------------------------------------------
+
+// See https://stackoverflow.com/a/14038590
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
+#define checkGpu( code ) { assertGpu( code, __FILE__, __LINE__ ); }
+inline void assertGpu( gpuError_t code, const char* file, int line, bool abort = true )
+{
+  if( code != gpuSuccess )
+  {
+    printf( "ERROR! assertGpu: '%s' (%d) in %s:%d\n", gpuGetErrorString( code ), code, file, line );
+    if( abort ) assert( code == gpuSuccess );
+  }
+}
+#endif /* clang-format on */
+
+//--------------------------------------------------------------------------
+
+#ifdef MGONGPUCPP_GPUIMPL
+namespace mg5amcGpu
+{
+  // Instantiate a GpuRuntime at the beginnining of the application's main to
+  // invoke gpuSetDevice(0) in the constructor and book a gpuDeviceReset() call in the destructor
+  // *** FIXME! This will all need to be designed differently when going to multi-GPU nodes! ***
+  struct GpuRuntime final
+  {
+    GpuRuntime( const bool debug = true )
+      : m_debug( debug ) { setUp( m_debug ); }
+    ~GpuRuntime() { tearDown( m_debug ); }
+    GpuRuntime( const GpuRuntime& ) = delete;
+    GpuRuntime( GpuRuntime&& ) = delete;
+    GpuRuntime& operator=( const GpuRuntime& ) = delete;
+    GpuRuntime& operator=( GpuRuntime&& ) = delete;
+    bool m_debug;
+
+    // Set up CUDA application
+    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
+    // Calling cudaSetDevice on startup is useful to properly book-keep the time spent in CUDA initialization
+    static void setUp( const bool debug = true )
+    {
+      // ** NB: it is useful to call cudaSetDevice, or cudaFree, to properly book-keep the time spent in CUDA initialization
+      // ** NB: otherwise, the first CUDA operation (eg a cudaMemcpyToSymbol in CPPProcess ctor) appears to take much longer!
+      /*
+      // [We initially added cudaFree(0) to "ease profile analysis" only because it shows up as a big recognizable block!]
+      // No explicit initialization is needed: https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#initialization
+      // It is not clear what cudaFree(0) does at all: https://stackoverflow.com/questions/69967813/
+      if ( debug ) std::cout << "__CudaRuntime: calling cudaFree(0)" << std::endl;
+      checkCuda( cudaFree( 0 ) ); // SLOW!
+      */
+      // Replace cudaFree(0) by cudaSetDevice(0), even if it is not really needed either
+      // (but see https://developer.nvidia.com/blog/cuda-pro-tip-always-set-current-device-avoid-multithreading-bugs)
+      if( debug ) std::cout << "__GpuRuntime: calling GpuSetDevice(0)" << std::endl;
+      checkGpu( gpuSetDevice( 0 ) ); // SLOW!
+    }
+
+    // Tear down CUDA application (call cudaDeviceReset)
+    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
+    // Calling cudaDeviceReset on shutdown is only needed for checking memory leaks in cuda-memcheck
+    // See https://docs.nvidia.com/cuda/cuda-memcheck/index.html#leak-checking
+    static void tearDown( const bool debug = true )
+    {
+      if( debug ) std::cout << "__GpuRuntime: calling GpuDeviceReset()" << std::endl;
+      checkGpu( gpuDeviceReset() );
+    }
+  };
+}
+#endif
+
+//--------------------------------------------------------------------------
+
+#endif // MG5AMC_GPURUNTIME_H
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/MadgraphTest.h b/epochX/cudacpp/gg_tt.sa/SubProcesses/MadgraphTest.h
index ef40624c88..6054185300 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/MadgraphTest.h
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/MadgraphTest.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Hageboeck (Dec 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MADGRAPHTEST_H_
 #define MADGRAPHTEST_H_ 1
@@ -14,7 +14,11 @@
 
 #include <array>
 #include <cmath>
-#include <filesystem>
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
 #include <fstream>
 #include <iomanip>
 #include <memory>
@@ -22,7 +26,7 @@
 #include <string>
 #include <vector>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using mg5amcGpu::CPPProcess;
 #else
 using mg5amcCpu::CPPProcess;
@@ -201,7 +205,7 @@ class MadgraphTest : public testing::TestWithParam<TestDriverBase*>
 
 // Since we link both the CPU-only and GPU tests into the same executable, we prevent
 // a multiply defined symbol by only compiling this in the non-CUDA phase:
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
 /// Compare momenta and matrix elements.
 /// This uses an implementation of TestDriverBase to run a madgraph workflow,
@@ -219,7 +223,14 @@ TEST_P( MadgraphTest, CompareMomentaAndME )
   const char* dumpEventsC = getenv( "CUDACPP_RUNTEST_DUMPEVENTS" );
   const bool dumpEvents = ( dumpEventsC != 0 ) && ( std::string( dumpEventsC ) != "" );
   const std::string refFileName = testDriver->getRefFileName();
+  /*
+#ifdef __HIPCC__
+  const std::string dumpFileName = std::experimental::filesystem::path( refFileName ).filename();
+#else
   const std::string dumpFileName = std::filesystem::path( refFileName ).filename();
+#endif
+  */
+  const std::string dumpFileName = refFileName; // bypass std::filesystem #803
   std::ofstream dumpFile;
   if( dumpEvents )
   {
@@ -307,6 +318,6 @@ TEST_P( MadgraphTest, CompareMomentaAndME )
   }
 }
 
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
 
 #endif /* MADGRAPHTEST_H_ */
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/MatrixElementKernels.cc b/epochX/cudacpp/gg_tt.sa/SubProcesses/MatrixElementKernels.cc
index 74b5239ebf..81699dfea9 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/MatrixElementKernels.cc
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/MatrixElementKernels.cc
@@ -1,12 +1,12 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "MatrixElementKernels.h"
 
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h" // Includes the abstraction for Nvidia/AMD compilation
 #include "MemoryAccessMomenta.h"
 #include "MemoryBuffers.h"
 
@@ -14,7 +14,7 @@
 
 //============================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu
 {
 
@@ -150,7 +150,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
@@ -209,13 +209,13 @@ namespace mg5amcGpu
     PinnedHostBufferHelicityMask hstIsGoodHel( ncomb );
     DeviceBufferHelicityMask devIsGoodHel( ncomb );
     // ... 0d1. Compute good helicity mask on the device
-    computeDependentCouplings<<<m_gpublocks, m_gputhreads>>>( m_gs.data(), m_couplings.data() );
+    gpuLaunchKernel( computeDependentCouplings, m_gpublocks, m_gputhreads, m_gs.data(), m_couplings.data() );
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-    sigmaKin_getGoodHel<<<m_gpublocks, m_gputhreads>>>( m_momenta.data(), m_couplings.data(), m_matrixElements.data(), m_numerators.data(), m_denominators.data(), devIsGoodHel.data() );
+    gpuLaunchKernel( sigmaKin_getGoodHel, m_gpublocks, m_gputhreads, m_momenta.data(), m_couplings.data(), m_matrixElements.data(), m_numerators.data(), m_denominators.data(), devIsGoodHel.data() );
 #else
-    sigmaKin_getGoodHel<<<m_gpublocks, m_gputhreads>>>( m_momenta.data(), m_couplings.data(), m_matrixElements.data(), devIsGoodHel.data() );
+    gpuLaunchKernel( sigmaKin_getGoodHel, m_gpublocks, m_gputhreads, m_momenta.data(), m_couplings.data(), m_matrixElements.data(), devIsGoodHel.data() );
 #endif
-    checkCuda( cudaPeekAtLastError() );
+    checkGpu( gpuPeekAtLastError() );
     // ... 0d2. Copy back good helicity mask to the host
     copyHostFromDevice( hstIsGoodHel, devIsGoodHel );
     // ... 0d3. Copy back good helicity list to constant memory on the device
@@ -226,19 +226,19 @@ namespace mg5amcGpu
 
   void MatrixElementKernelDevice::computeMatrixElements( const unsigned int channelId )
   {
-    computeDependentCouplings<<<m_gpublocks, m_gputhreads>>>( m_gs.data(), m_couplings.data() );
+    gpuLaunchKernel( computeDependentCouplings, m_gpublocks, m_gputhreads, m_gs.data(), m_couplings.data() );
 #ifndef MGONGPU_NSIGHT_DEBUG
     constexpr unsigned int sharedMemSize = 0;
 #else
     constexpr unsigned int sharedMemSize = ntpbMAX * sizeof( float );
 #endif
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-    sigmaKin<<<m_gpublocks, m_gputhreads, sharedMemSize>>>( m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), channelId, m_numerators.data(), m_denominators.data(), m_selhel.data(), m_selcol.data() );
+    gpuLaunchKernelSharedMem( sigmaKin, m_gpublocks, m_gputhreads, sharedMemSize, m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), channelId, m_numerators.data(), m_denominators.data(), m_selhel.data(), m_selcol.data() );
 #else
-    sigmaKin<<<m_gpublocks, m_gputhreads, sharedMemSize>>>( m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), m_selhel.data(), m_selcol.data() );
+    gpuLaunchKernelSharedMem( sigmaKin, m_gpublocks, m_gputhreads, sharedMemSize, m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), m_selhel.data(), m_selcol.data() );
 #endif
-    checkCuda( cudaPeekAtLastError() );
-    checkCuda( cudaDeviceSynchronize() );
+    checkGpu( gpuPeekAtLastError() );
+    checkGpu( gpuDeviceSynchronize() );
   }
 
   //--------------------------------------------------------------------------
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/MatrixElementKernels.h b/epochX/cudacpp/gg_tt.sa/SubProcesses/MatrixElementKernels.h
index 23e84757a2..72bd8f195b 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/MatrixElementKernels.h
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/MatrixElementKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MATRIXELEMENTKERNELS_H
 #define MATRIXELEMENTKERNELS_H 1
@@ -10,7 +10,7 @@
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -81,7 +81,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating matrix element calculations on a CPU host
   class MatrixElementKernelHost final : public MatrixElementKernelBase, public NumberOfEvents
   {
@@ -130,7 +130,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating matrix element calculations on a GPU device
   class MatrixElementKernelDevice : public MatrixElementKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessAmplitudes.h b/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessAmplitudes.h
index 573b3bbbc9..ffb76e93de 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessAmplitudes.h
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessAmplitudes.h
@@ -15,7 +15,7 @@
 #define MGONGPU_TRIVIAL_AMPLITUDES 1
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessCouplings.h b/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessCouplings.h
index 35a3af42e0..3afdf3e554 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessCouplings.h
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessCouplings.h
@@ -15,7 +15,7 @@
 #include "MemoryBuffers.h"       // for HostBufferCouplings::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessCouplingsFixed.h b/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessCouplingsFixed.h
index dc0d93afff..ffcdf4dbef 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessCouplingsFixed.h
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessCouplingsFixed.h
@@ -14,7 +14,7 @@
 //#include "MemoryAccessHelpers.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessDenominators.h b/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessDenominators.h
index 3bce635718..66f2d32a6b 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessDenominators.h
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessDenominators.h
@@ -10,7 +10,7 @@
 #include "MemoryAccessGs.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessGs.h b/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessGs.h
index 31311aa375..4c726b30f3 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessGs.h
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessGs.h
@@ -13,7 +13,7 @@
 #include "MemoryBuffers.h" // for HostBufferMatrixElements::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessHelpers.h b/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessHelpers.h
index c82a6c7635..db73e4e064 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessHelpers.h
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessHelpers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessHelpers_H
 #define MemoryAccessHelpers_H 1
@@ -105,7 +105,7 @@ class KernelAccessHelper : public MemoryAccessHelper<T>
     }
     else
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
       //printf( "kernelAccessRecord: ievt=%d threadId=%d\n", ievt, threadIdx.x );
       return T::ieventAccessRecord( buffer, ievt ); // NB fptype and fptype_sv coincide for CUDA
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessMatrixElements.h b/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessMatrixElements.h
index f32e6fea5b..3741011971 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessMatrixElements.h
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessMatrixElements.h
@@ -13,7 +13,7 @@
 #include "MemoryBuffers.h" // for HostBufferMatrixElements::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessMomenta.h b/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessMomenta.h
index 29266de32c..3be229d392 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessMomenta.h
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessMomenta.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessMomenta_H
 #define MemoryAccessMomenta_H 1
@@ -13,7 +13,7 @@
 #include "MemoryAccessVectors.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -30,7 +30,7 @@ namespace mg5amcCpu
 
     // Number of Events Per Page in the momenta AOSOA memory buffer layout
     // (these are all best kept as a compile-time constants: see issue #23)
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
     // -----------------------------------------------------------------------------------------------
     // --- GPUs: neppM is best set to a power of 2 times the number of fptype's in a 32-byte cacheline
     // --- This is relevant to ensure coalesced access to momenta in global memory
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessNumerators.h b/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessNumerators.h
index b152183b28..18991f4fa6 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessNumerators.h
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessNumerators.h
@@ -10,7 +10,7 @@
 #include "MemoryAccessGs.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessRandomNumbers.h b/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessRandomNumbers.h
index e2988d39f3..40cb089135 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessRandomNumbers.h
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessRandomNumbers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessRandomNumbers_H
 #define MemoryAccessRandomNumbers_H 1
@@ -11,7 +11,7 @@
 #include "CPPProcess.h"
 #include "MemoryAccessHelpers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using mg5amcGpu::CPPProcess;
 #else
 using mg5amcCpu::CPPProcess;
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessVectors.h b/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessVectors.h
index e9b197368e..08faccff0f 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessVectors.h
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessVectors.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessVectors_H
 #define MemoryAccessVectors_H 1
@@ -10,7 +10,7 @@
 
 #include "mgOnGpuVectors.h"
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu // this is only needed for CPU SIMD vectorization
 {
 
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessWavefunctions.h b/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessWavefunctions.h
index 5428aaf933..33bef4559e 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessWavefunctions.h
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryAccessWavefunctions.h
@@ -15,7 +15,7 @@
 #define MGONGPU_TRIVIAL_WAVEFUNCTIONS 1
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryBuffers.h b/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryBuffers.h
index 3093e6ed18..7756a71621 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryBuffers.h
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/MemoryBuffers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021, based on earlier work by S. Hageboeck) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryBuffers_H
 #define MemoryBuffers_H 1
@@ -11,12 +11,12 @@
 #include "mgOnGpuCxtypes.h"
 
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "Parameters_sm.h"
 
 #include <sstream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -87,7 +87,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   constexpr bool HostBufferALIGNED = false;   // ismisaligned=false
   constexpr bool HostBufferMISALIGNED = true; // ismisaligned=true
 
@@ -119,7 +119,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA pinned host buffer
   template<typename T>
   class PinnedHostBufferBase : public BufferBase<T>
@@ -128,18 +128,18 @@ namespace mg5amcCpu
     PinnedHostBufferBase( const size_t size )
       : BufferBase<T>( size, false )
     {
-      checkCuda( cudaMallocHost( &( this->m_data ), this->bytes() ) );
+      gpuMallocHost( &( this->m_data ), this->bytes() );
     }
     virtual ~PinnedHostBufferBase()
     {
-      checkCuda( cudaFreeHost( this->m_data ) );
+      gpuFreeHost( this->m_data );
     }
   };
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA device buffer
   template<typename T>
   class DeviceBufferBase : public BufferBase<T>
@@ -148,18 +148,18 @@ namespace mg5amcCpu
     DeviceBufferBase( const size_t size )
       : BufferBase<T>( size, true )
     {
-      checkCuda( cudaMalloc( &( this->m_data ), this->bytes() ) );
+      gpuMalloc( &( this->m_data ), this->bytes() );
     }
     virtual ~DeviceBufferBase()
     {
-      checkCuda( cudaFree( this->m_data ) );
+      gpuFree( this->m_data );
     }
   };
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for a given number of events
   template<typename T, size_t sizePerEvent, bool ismisaligned>
   class HostBuffer : public HostBufferBase<T, ismisaligned>, virtual private NumberOfEvents
@@ -175,7 +175,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA pinned host buffer for a given number of events
   template<typename T, size_t sizePerEvent>
   class PinnedHostBuffer : public PinnedHostBufferBase<T>, virtual private NumberOfEvents
@@ -191,7 +191,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA device buffer for a given number of events
   template<typename T, size_t sizePerEvent>
   class DeviceBuffer : public DeviceBufferBase<T>, virtual private NumberOfEvents
@@ -213,7 +213,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for momenta random numbers
   constexpr size_t sizePerEventRndNumMomenta = MemoryBuffers::np4 * MemoryBuffers::nparf;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for momenta random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumMomenta, HostBufferALIGNED> HostBufferRndNumMomenta;
 #else
@@ -232,7 +232,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer with ONE fptype per event
   constexpr size_t sizePerEventOneFp = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer with ONE fptype per event
   typedef HostBuffer<fptype, sizePerEventOneFp, HostBufferALIGNED> HostBufferOneFp;
 #else
@@ -257,7 +257,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for Gs
   constexpr size_t sizePerEventGs = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventGs, HostBufferALIGNED> HostBufferGs;
 #else
@@ -276,7 +276,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for numerators
   constexpr size_t sizePerEventNumerators = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventNumerators, HostBufferALIGNED> HostBufferNumerators;
 #else
@@ -296,7 +296,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for denominators
   constexpr size_t sizePerEventDenominators = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventDenominators, HostBufferALIGNED> HostBufferDenominators;
 #else
@@ -315,7 +315,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for random numbers
   constexpr size_t sizePerEventCouplings = MemoryBuffers::ndcoup * MemoryBuffers::nx2;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventCouplings, HostBufferALIGNED> HostBufferCouplings;
 #else
@@ -333,7 +333,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for momenta
   constexpr size_t sizePerEventMomenta = MemoryBuffers::np4 * MemoryBuffers::npar;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for momenta
   typedef HostBuffer<fptype, sizePerEventMomenta, HostBufferALIGNED> HostBufferMomenta;
   //typedef HostBuffer<fptype, sizePerEventMomenta, HostBufferMISALIGNED> HostBufferMomenta; // TEST MISALIGNMENT!
@@ -352,7 +352,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for sampling weights
   constexpr size_t sizePerEventWeights = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for sampling weights
   typedef HostBuffer<fptype, sizePerEventWeights, HostBufferALIGNED> HostBufferWeights;
 #else
@@ -370,7 +370,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for matrix elements
   constexpr size_t sizePerEventMatrixElements = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for matrix elements
   typedef HostBuffer<fptype, sizePerEventMatrixElements, HostBufferALIGNED> HostBufferMatrixElements;
 #else
@@ -385,7 +385,7 @@ namespace mg5amcCpu
   // A base class encapsulating a memory buffer for the helicity mask
   typedef BufferBase<bool> BufferHelicityMask;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for the helicity mask
   typedef HostBufferBase<bool, HostBufferALIGNED> HostBufferHelicityMask;
 #else
@@ -403,7 +403,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for wavefunctions
   constexpr size_t sizePerEventWavefunctions = MemoryBuffers::nw6 * MemoryBuffers::nx2;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for wavefunctions
   typedef HostBuffer<fptype, sizePerEventWavefunctions, HostBufferALIGNED> HostBufferWavefunctions;
 #else
@@ -421,7 +421,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for helicity random numbers
   constexpr size_t sizePerEventRndNumHelicity = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for helicity random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumHelicity, HostBufferALIGNED> HostBufferRndNumHelicity;
 #else
@@ -439,7 +439,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for color random numbers
   constexpr size_t sizePerEventRndNumColor = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for color random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumColor, HostBufferALIGNED> HostBufferRndNumColor;
 #else
@@ -457,7 +457,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for helicity selection
   constexpr size_t sizePerEventSelectedHelicity = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for helicity selection
   typedef HostBuffer<int, sizePerEventSelectedHelicity, HostBufferALIGNED> HostBufferSelectedHelicity;
 #else
@@ -475,7 +475,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for color selection
   constexpr size_t sizePerEventSelectedColor = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for color selection
   typedef HostBuffer<int, sizePerEventSelectedColor, HostBufferALIGNED> HostBufferSelectedColor;
 #else
@@ -487,7 +487,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<class Tdst, class Tsrc>
   void copyDeviceFromHost( Tdst& dst, const Tsrc& src ) // keep the same order of arguments as in memcpy
   {
@@ -504,13 +504,13 @@ namespace mg5amcCpu
       throw std::runtime_error( sstr.str() );
     }
     // NB (PR #45): cudaMemcpy involves an intermediate memcpy to pinned memory if host array is a not a pinned host array
-    checkCuda( cudaMemcpy( dst.data(), src.data(), src.bytes(), cudaMemcpyHostToDevice ) );
+    gpuMemcpy( dst.data(), src.data(), src.bytes(), gpuMemcpyHostToDevice );
   }
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<class Tdst, class Tsrc>
   void copyHostFromDevice( Tdst& dst, const Tsrc& src ) // keep the same order of arguments as in memcpy
   {
@@ -527,7 +527,7 @@ namespace mg5amcCpu
       throw std::runtime_error( sstr.str() );
     }
     // NB (PR #45): cudaMemcpy involves an intermediate memcpy to pinned memory if host array is a not a pinned host array
-    checkCuda( cudaMemcpy( dst.data(), src.data(), src.bytes(), cudaMemcpyDeviceToHost ) );
+    gpuMemcpy( dst.data(), src.data(), src.bytes(), gpuMemcpyDeviceToHost );
   }
 #endif
 
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/P1_Sigma_sm_gg_ttx/CPPProcess.cc b/epochX/cudacpp/gg_tt.sa/SubProcesses/P1_Sigma_sm_gg_ttx/CPPProcess.cc
index d390883453..c51f01c456 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/P1_Sigma_sm_gg_ttx/CPPProcess.cc
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/P1_Sigma_sm_gg_ttx/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_sm.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -46,7 +45,7 @@
 // Class member functions for calculating the matrix elements for
 // Process: g g > t t~ WEIGHTED<=2 @1
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -80,7 +79,7 @@ namespace mg5amcCpu
   __device__ const fptype cIPD[2] = { (fptype)Parameters_sm::mdl_MT, (fptype)Parameters_sm::mdl_WT };
   __device__ const fptype* cIPC = nullptr; // unused as nicoup=0
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ fptype cIPD[2];
   __device__ __constant__ fptype* cIPC = nullptr; // unused as nicoup=0
 #else
@@ -90,7 +89,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -118,13 +117,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -151,7 +150,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -187,7 +186,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -200,8 +199,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -299,7 +300,7 @@ namespace mg5amcCpu
         { 16, -2 },
         { -2, 16 } }; // 2-D array[2][2]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -356,7 +357,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -415,7 +416,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -462,8 +463,8 @@ namespace mg5amcCpu
       { 1, 1, -1, -1 },
       { 1, 1, 1, 1 },
       { 1, 1, 1, -1 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -503,9 +504,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     const fptype tIPD[2] = { (fptype)m_pars->mdl_MT, (fptype)m_pars->mdl_WT };
     //const cxtype tIPC[0] = { ... }; // nicoup=0
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) ) );
-    //checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ) ); // nicoup=0
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) );
+    //gpuMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
 #else
     memcpy( cIPD, tIPD, 2 * sizeof( fptype ) );
     //memcpy( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
@@ -541,7 +542,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -570,6 +571,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -583,6 +588,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -606,12 +613,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -632,7 +639,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -758,9 +765,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -784,7 +791,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -804,7 +811,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 256 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -818,9 +825,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -848,7 +858,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -1058,7 +1068,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/P1_Sigma_sm_gg_ttx/CPPProcess.h b/epochX/cudacpp/gg_tt.sa/SubProcesses/P1_Sigma_sm_gg_ttx/CPPProcess.h
index 3ebd92c038..4a88a07226 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/P1_Sigma_sm_gg_ttx/CPPProcess.h
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/P1_Sigma_sm_gg_ttx/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -107,7 +107,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -120,7 +120,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -150,7 +150,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/P1_Sigma_sm_gg_ttx/CudaRuntime.h b/epochX/cudacpp/gg_tt.sa/SubProcesses/P1_Sigma_sm_gg_ttx/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/P1_Sigma_sm_gg_ttx/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/P1_Sigma_sm_gg_ttx/GpuAbstraction.h b/epochX/cudacpp/gg_tt.sa/SubProcesses/P1_Sigma_sm_gg_ttx/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/P1_Sigma_sm_gg_ttx/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/P1_Sigma_sm_gg_ttx/GpuRuntime.h b/epochX/cudacpp/gg_tt.sa/SubProcesses/P1_Sigma_sm_gg_ttx/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/P1_Sigma_sm_gg_ttx/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/P1_Sigma_sm_gg_ttx/check_sa.cc b/epochX/cudacpp/gg_tt.sa/SubProcesses/P1_Sigma_sm_gg_ttx/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/P1_Sigma_sm_gg_ttx/check_sa.cc
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/P1_Sigma_sm_gg_ttx/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/RamboSamplingKernels.cc b/epochX/cudacpp/gg_tt.sa/SubProcesses/RamboSamplingKernels.cc
index da68aa9255..79abbcc4f8 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/RamboSamplingKernels.cc
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/RamboSamplingKernels.cc
@@ -1,11 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "RamboSamplingKernels.h"
 
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "MemoryAccessMomenta.h"
 #include "MemoryAccessRandomNumbers.h"
 #include "MemoryAccessWeights.h"
@@ -14,7 +14,7 @@
 
 #include <sstream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -92,7 +92,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingKernelDevice::RamboSamplingKernelDevice( const fptype energy,               // input: energy
                                                         const BufferRndNumMomenta& rndmom, // input: random numbers in [0,1]
                                                         BufferMomenta& momenta,            // output: momenta
@@ -135,7 +135,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   getMomentaInitialDevice( const fptype energy,
                            fptype* momenta )
@@ -147,17 +147,17 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   void
   RamboSamplingKernelDevice::getMomentaInitial()
   {
-    getMomentaInitialDevice<<<m_gpublocks, m_gputhreads>>>( m_energy, m_momenta.data() );
+    gpuLaunchKernel( getMomentaInitialDevice, m_gpublocks, m_gputhreads, m_energy, m_momenta.data() );
   }
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   getMomentaFinalDevice( const fptype energy,
                          const fptype* rndmom,
@@ -171,11 +171,11 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   void
   RamboSamplingKernelDevice::getMomentaFinal()
   {
-    getMomentaFinalDevice<<<m_gpublocks, m_gputhreads>>>( m_energy, m_rndmom.data(), m_momenta.data(), m_weights.data() );
+    gpuLaunchKernel( getMomentaFinalDevice, m_gpublocks, m_gputhreads, m_energy, m_rndmom.data(), m_momenta.data(), m_weights.data() );
   }
 #endif
 
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/RamboSamplingKernels.h b/epochX/cudacpp/gg_tt.sa/SubProcesses/RamboSamplingKernels.h
index 184089efd7..7c214cd74b 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/RamboSamplingKernels.h
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/RamboSamplingKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef RAMBOSAMPLINGKERNELS_H
 #define RAMBOSAMPLINGKERNELS_H 1
@@ -10,7 +10,7 @@
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -93,7 +93,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating RAMBO phase space sampling on a GPU device
   class RamboSamplingKernelDevice final : public SamplingKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/RandomNumberKernels.h b/epochX/cudacpp/gg_tt.sa/SubProcesses/RandomNumberKernels.h
index 188a72c2c9..21d63beeac 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/RandomNumberKernels.h
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/RandomNumberKernels.h
@@ -1,14 +1,14 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef RANDOMNUMBERKERNELS_H
 #define RANDOMNUMBERKERNELS_H 1
 
 #include "mgOnGpuConfig.h"
 
-// NB This must come AFTER mgOnGpuConfig.h which contains our definition of __global__ when __CUDACC__ is not defined
+// NB This must come AFTER mgOnGpuConfig.h which contains our definition of __global__ when MGONGPUCPP_GPUIMPL is not defined
 #ifndef MGONGPU_HAS_NO_CURAND
 //#include "curand.h"
 struct curandGenerator_st; // forward definition from curand.h
@@ -16,7 +16,7 @@ struct curandGenerator_st; // forward definition from curand.h
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/cudacpp.mk b/epochX/cudacpp/gg_tt.sa/SubProcesses/cudacpp.mk
index 509307506b..df74dfc284 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/cudacpp.mk
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/cudacpp.mk
@@ -1,7 +1,7 @@
 # Copyright (C) 2020-2023 CERN and UCLouvain.
 # Licensed under the GNU Lesser General Public License (version 3 or later).
 # Created by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-# Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+# Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #=== Determine the name of this makefile (https://ftp.gnu.org/old-gnu/Manuals/make-3.80/html_node/make_17.html)
 #=== NB: use ':=' to ensure that the value of CUDACPP_MAKEFILE is not modified further down after including make_opts
@@ -42,10 +42,10 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure common compiler flags for C++ and CUDA
+#=== Configure common compiler flags for C++ and CUDA/HIP
 
 INCFLAGS = -I.
-OPTFLAGS = -O3 # this ends up in CUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
+OPTFLAGS = -O3 # this ends up in GPUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
 
 # Dependency on src directory
 MG5AMC_COMMONLIB = mg5amc_common
@@ -121,24 +121,45 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure the CUDA compiler
+#=== Configure the GPU compiler (CUDA or HIP)
+
+# FIXME! (AV 24.01.2024)
+# In the current implementation (without separate builds for C++ and CUDA/HIP), we first check for cudacc and hipcc in CUDA_HOME and HIP_HOME.
+# If CUDA_HOME or HIP_HOME are not set, try to determine them from the path to cudacc and hipcc.
+# While convoluted, this is currently necessary to allow disabling CUDA/HIP builds by setting CUDA_HOME or HIP_HOME to invalid paths.
+# This will (probably?) be fixed when separate C++ and CUDA/HIP builds are implemented (PR #775).
 
 # If CXX is not a single word (example "clang++ --gcc-toolchain...") then disable CUDA builds (issue #505)
-# This is because it is impossible to pass this to "CUFLAGS += -ccbin <host-compiler>" below
+# This is because it is impossible to pass this to "GPUFLAGS += -ccbin <host-compiler>" below
 ifneq ($(words $(subst ccache ,,$(CXX))),1) # allow at most "CXX=ccache <host-compiler>" from outside
   $(warning CUDA builds are not supported for multi-word CXX "$(CXX)")
   override CUDA_HOME=disabled
 endif
 
-# If CUDA_HOME is not set, try to set it from the location of nvcc
+# If CUDA_HOME is not set, try to set it from the path to nvcc
 ifndef CUDA_HOME
-  CUDA_HOME = $(patsubst %bin/nvcc,%,$(shell which nvcc 2>/dev/null))
+  CUDA_HOME = $(patsubst %/bin/nvcc,%,$(shell which nvcc 2>/dev/null))
   $(warning CUDA_HOME was not set: using "$(CUDA_HOME)")
 endif
 
-# Set NVCC as $(CUDA_HOME)/bin/nvcc if it exists
+# If HIP_HOME is not set, try to set it from the path to hipcc
+ifndef HIP_HOME
+  HIP_HOME = $(patsubst %/bin/hipcc,%,$(shell which hipcc 2>/dev/null))
+  $(warning HIP_HOME was not set: using "$(HIP_HOME)")
+endif
+
+# FIXME! (AV 24.01.2024)
+# In the current implementation (without separate builds for C++ and CUDA/HIP),
+# builds are performed for HIP only if CUDA is not found in the path.
+# If both CUDA and HIP are installed, HIP builds can be triggered by unsetting CUDA_HOME.
+# This will be fixed when separate C++ and CUDA/HIP builds are implemented (PR #775).
+
+#--- Option 1: CUDA exists -> use CUDA
+
+# Set GPUCC as $(CUDA_HOME)/bin/nvcc if it exists
 ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
-  NVCC = $(CUDA_HOME)/bin/nvcc
+
+  GPUCC = $(CUDA_HOME)/bin/nvcc
   USE_NVTX ?=-DUSE_NVTX
   # See https://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/index.html
   # See https://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/
@@ -148,7 +169,7 @@ ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
   # Examples: use 60 for P100 (Piz Daint), 80 for A100 (Juwels Booster, NVidia raplab/Curiosity).
   MADGRAPH_CUDA_ARCHITECTURE ?= 70
   ###CUARCHFLAGS = -gencode arch=compute_$(MADGRAPH_CUDA_ARCHITECTURE),code=compute_$(MADGRAPH_CUDA_ARCHITECTURE) -gencode arch=compute_$(MADGRAPH_CUDA_ARCHITECTURE),code=sm_$(MADGRAPH_CUDA_ARCHITECTURE) # Older implementation (AV): go back to this one for multi-GPU support #533
-  ###CUARCHFLAGS = --gpu-architecture=compute_$(MADGRAPH_CUDA_ARCHITECTURE) --gpu-code=sm_$(MADGRAPH_CUDA_ARCHITECTURE),compute_$(MADGRAPH_CUDA_ARCHITECTURE) # Newer implementation (SH): cannot use this as-is for multi-GPU support #533
+  ###CUARCHFLAGS = --gpu-architecture=compute_$(MADGRAPH_CUDA_ARCHITECTURE) --gpu-code=sm_$(MADGRAPH_CUDA_ARCHITECTURE),compute_$(MADGRAPH_CUDA_ARCHITECTURE)  # Newer implementation (SH): cannot use this as-is for multi-GPU support #533
   comma:=,
   CUARCHFLAGS = $(foreach arch,$(subst $(comma), ,$(MADGRAPH_CUDA_ARCHITECTURE)),-gencode arch=compute_$(arch),code=compute_$(arch) -gencode arch=compute_$(arch),code=sm_$(arch))
   CUINC = -I$(CUDA_HOME)/include/
@@ -158,41 +179,79 @@ ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
     CURANDLIBFLAGS = -L$(CUDA_HOME)/lib64/ -lcurand # NB: -lcuda is not needed here!
   endif
   CUOPTFLAGS = -lineinfo
-  CUFLAGS = $(foreach opt, $(OPTFLAGS), -Xcompiler $(opt)) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
-  ###CUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
-  ###NVCC_VERSION = $(shell $(NVCC) --version | grep 'Cuda compilation tools' | cut -d' ' -f5 | cut -d, -f1)
-  CUFLAGS += -std=c++17 # need CUDA >= 11.2 (see #333): this is enforced in mgOnGpuConfig.h
+  ###GPUFLAGS = $(OPTFLAGS) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
+  GPUFLAGS = $(foreach opt, $(OPTFLAGS), -Xcompiler $(opt)) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
+  ###GPUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
+  ###GPUCC_VERSION = $(shell $(GPUCC) --version | grep 'Cuda compilation tools' | cut -d' ' -f5 | cut -d, -f1)
+  GPUFLAGS += -std=c++17 # need CUDA >= 11.2 (see #333): this is enforced in mgOnGpuConfig.h
   # Without -maxrregcount: baseline throughput: 6.5E8 (16384 32 12) up to 7.3E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 160 # improves throughput: 6.9E8 (16384 32 12) up to 7.7E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 128 # improves throughput: 7.3E8 (16384 32 12) up to 7.6E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 96 # degrades throughput: 4.1E8 (16384 32 12) up to 4.5E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 64 # degrades throughput: 1.7E8 (16384 32 12) flat at 1.7E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 160 # improves throughput: 6.9E8 (16384 32 12) up to 7.7E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 128 # improves throughput: 7.3E8 (16384 32 12) up to 7.6E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 96 # degrades throughput: 4.1E8 (16384 32 12) up to 4.5E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 64 # degrades throughput: 1.7E8 (16384 32 12) flat at 1.7E8 (65536 128 12)
+  CUBUILDRULEFLAGS = -Xcompiler -fPIC -c
+  CCBUILDRULEFLAGS = -Xcompiler -fPIC -c -x cu
+  CUDATESTFLAGS = -lcuda
+
+  # Set the host C++ compiler for GPUCC via "-ccbin <host-compiler>"
+  # (NB issue #505: this must be a single word, "clang++ --gcc-toolchain..." is not supported)
+  GPUFLAGS += -ccbin $(shell which $(subst ccache ,,$(CXX)))
+
+  # Allow newer (unsupported) C++ compilers with older versions of CUDA if ALLOW_UNSUPPORTED_COMPILER_IN_CUDA is set (#504)
+  ifneq ($(origin ALLOW_UNSUPPORTED_COMPILER_IN_CUDA),undefined)
+  GPUFLAGS += -allow-unsupported-compiler
+  endif
+
 else ifneq ($(origin REQUIRE_CUDA),undefined)
+
   # If REQUIRE_CUDA is set but no cuda is found, stop here (e.g. for CI tests on GPU #443)
-  $(error No cuda installation found (set CUDA_HOME or make nvcc visible in PATH))
+  $(error No cuda installation found (set CUDA_HOME or make GPUCC visible in PATH))
+
+#--- Option 2: CUDA does not exist, HIP exists -> use HIP
+
+# Set GPUCC as $(HIP_HOME)/bin/hipcc if it exists
+else ifneq ($(wildcard $(HIP_HOME)/bin/hipcc),)
+
+  GPUCC = $(HIP_HOME)/bin/hipcc
+  #USE_NVTX ?=-DUSE_NVTX # should maybe find something equivalent to this in HIP?
+  HIPARCHFLAGS = -target x86_64-linux-gnu --offload-arch=gfx90a
+  HIPINC = -I$(HIP_HOME)/include/
+  # Note: -DHIP_FAST_MATH is equivalent to -use_fast_math in HIP 
+  # (but only for single precision line 208: https://rocm-developer-tools.github.io/HIP/hcc__detail_2math__functions_8h_source.html)
+  # Note: CUOPTFLAGS should not be used for HIP, it had been added here but was then removed (#808)
+  GPUFLAGS = $(OPTFLAGS) $(INCFLAGS) $(HIPINC) $(HIPARCHFLAGS) -DHIP_FAST_MATH -DHIP_PLATFORM=amd -fPIC
+  ###GPUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
+  GPUFLAGS += -std=c++17
+  ###GPUFLAGS+= --maxrregcount 255 # (AV: is this option valid on HIP and meaningful on AMD GPUs?)
+  CUBUILDRULEFLAGS = -fPIC -c
+  CCBUILDRULEFLAGS = -fPIC -c -x hip
+
+else ifneq ($(origin REQUIRE_HIP),undefined)
+
+  # If REQUIRE_HIP is set but no HIP is found, stop here (e.g. for CI tests on GPU #443)
+  $(error No hip installation found (set HIP_HOME or make GPUCC visible in PATH))
+
+#--- Option 3: CUDA does not exist, HIP does not exist -> switch off both CUDA and HIP
+
 else
-  # No cuda. Switch cuda compilation off and go to common random numbers in C++
+
+  # No cudacc and no hipcc: switch CUDA and HIP compilation off and go to common random numbers in C++
   $(warning CUDA_HOME is not set or is invalid: export CUDA_HOME to compile with cuda)
-  override NVCC=
+  $(warning HIP_HOME is not set or is invalid: export HIP_HOME to compile with hip)
+  override GPUCC=
   override USE_NVTX=
   override CUINC=
   override CURANDLIBFLAGS=
-endif
-export NVCC
-export CUFLAGS
 
-# Set the host C++ compiler for nvcc via "-ccbin <host-compiler>"
-# (NB issue #505: this must be a single word, "clang++ --gcc-toolchain..." is not supported)
-CUFLAGS += -ccbin $(shell which $(subst ccache ,,$(CXX)))
-
-# Allow newer (unsupported) C++ compilers with older versions of CUDA if ALLOW_UNSUPPORTED_COMPILER_IN_CUDA is set (#504)
-ifneq ($(origin ALLOW_UNSUPPORTED_COMPILER_IN_CUDA),undefined)
-CUFLAGS += -allow-unsupported-compiler
 endif
 
+# Export GPUCC (so that it can also be used in cudacpp_src.mk?)
+export GPUCC
+export GPUFLAGS
+
 #-------------------------------------------------------------------------------
 
-#=== Configure ccache for C++ and CUDA builds
+#=== Configure ccache for C++ and CUDA/HIP builds
 
 # Enable ccache if USECCACHE=1
 ifeq ($(USECCACHE)$(shell echo $(CXX) | grep ccache),1)
@@ -201,15 +260,15 @@ endif
 #ifeq ($(USECCACHE)$(shell echo $(AR) | grep ccache),1)
 #  override AR:=ccache $(AR)
 #endif
-ifneq ($(NVCC),)
-  ifeq ($(USECCACHE)$(shell echo $(NVCC) | grep ccache),1)
-    override NVCC:=ccache $(NVCC)
+ifneq ($(GPUCC),)
+  ifeq ($(USECCACHE)$(shell echo $(GPUCC) | grep ccache),1)
+    override GPUCC:=ccache $(GPUCC)
   endif
 endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure PowerPC-specific compiler flags for C++ and CUDA
+#=== Configure PowerPC-specific compiler flags for C++ and CUDA/HIP
 
 # PowerPC-specific CXX compiler flags (being reviewed)
 ifeq ($(UNAME_P),ppc64le)
@@ -225,9 +284,9 @@ else
   ######CXXFLAGS+= -fno-semantic-interposition # no benefit (neither alone, nor combined with -flto)
 endif
 
-# PowerPC-specific CUDA compiler flags (to be reviewed!)
+# PowerPC-specific CUDA/HIP compiler flags (to be reviewed!)
 ifeq ($(UNAME_P),ppc64le)
-  CUFLAGS+= -Xcompiler -mno-float128
+  GPUFLAGS+= -Xcompiler -mno-float128
 endif
 
 #-------------------------------------------------------------------------------
@@ -235,9 +294,11 @@ endif
 #=== Configure defaults and check if user-defined choices exist for OMPFLAGS, AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
 
 # Set the default OMPFLAGS choice
-ifneq ($(shell $(CXX) --version | egrep '^Intel'),)
+ifneq ($(findstring hipcc,$(GPUCC)),)
+override OMPFLAGS = # disable OpenMP MT when using hipcc #802
+else ifneq ($(shell $(CXX) --version | egrep '^Intel'),)
 override OMPFLAGS = -fopenmp
-###override OMPFLAGS = # disable OpenMP MT on Intel (was ok without nvcc but not ok with nvcc before #578)
+###override OMPFLAGS = # disable OpenMP MT on Intel (was ok without GPUCC but not ok with GPUCC before #578)
 else ifneq ($(shell $(CXX) --version | egrep '^(clang)'),)
 override OMPFLAGS = -fopenmp
 ###override OMPFLAGS = # disable OpenMP MT on clang (was not ok without or with nvcc before #578)
@@ -293,7 +354,10 @@ endif
 
 # Set the default RNDGEN (random number generator) choice
 ifeq ($(RNDGEN),)
-  ifeq ($(NVCC),)
+  ifeq ($(GPUCC),)
+    override RNDGEN = hasNoCurand
+  # Edgecase for HIP compilation
+  else ifeq ($(findstring hipcc,$(GPUCC)),hipcc)
     override RNDGEN = hasNoCurand
   else ifeq ($(RNDGEN),)
     override RNDGEN = hasCurand
@@ -310,7 +374,7 @@ export OMPFLAGS
 
 #-------------------------------------------------------------------------------
 
-#=== Set the CUDA/C++ compiler flags appropriate to user-defined choices of AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
+#=== Set the CUDA/HIP/C++ compiler flags appropriate to user-defined choices of AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
 
 # Set the build flags appropriate to OMPFLAGS
 $(info OMPFLAGS=$(OMPFLAGS))
@@ -368,13 +432,13 @@ CXXFLAGS+= $(AVXFLAGS)
 $(info FPTYPE=$(FPTYPE))
 ifeq ($(FPTYPE),d)
   CXXFLAGS += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
-  CUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
+  GPUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
 else ifeq ($(FPTYPE),f)
   CXXFLAGS += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
-  CUFLAGS  += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
+  GPUFLAGS  += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
 else ifeq ($(FPTYPE),m)
   CXXFLAGS += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
-  CUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
+  GPUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
 else
   $(error Unknown FPTYPE='$(FPTYPE)': only 'd', 'f' and 'm' are supported)
 endif
@@ -383,7 +447,7 @@ endif
 $(info HELINL=$(HELINL))
 ifeq ($(HELINL),1)
   CXXFLAGS += -DMGONGPU_INLINE_HELAMPS
-  CUFLAGS  += -DMGONGPU_INLINE_HELAMPS
+  GPUFLAGS  += -DMGONGPU_INLINE_HELAMPS
 else ifneq ($(HELINL),0)
   $(error Unknown HELINL='$(HELINL)': only '0' and '1' are supported)
 endif
@@ -392,7 +456,7 @@ endif
 $(info HRDCOD=$(HRDCOD))
 ifeq ($(HRDCOD),1)
   CXXFLAGS += -DMGONGPU_HARDCODE_PARAM
-  CUFLAGS  += -DMGONGPU_HARDCODE_PARAM
+  GPUFLAGS  += -DMGONGPU_HARDCODE_PARAM
 else ifneq ($(HRDCOD),0)
   $(error Unknown HRDCOD='$(HRDCOD)': only '0' and '1' are supported)
 endif
@@ -444,11 +508,11 @@ ifeq ($(UNAME_S),Darwin)
   override CULIBFLAGSRPATH2 =
 else
   # RPATH to cuda/cpp libs when linking executables
-  override CXXLIBFLAGSRPATH = -Wl,-rpath,$(LIBDIRRPATH)
-  override CULIBFLAGSRPATH = -Xlinker -rpath,$(LIBDIRRPATH)
+  override CXXLIBFLAGSRPATH = -Wl,-rpath=$(LIBDIRRPATH)
+  override CULIBFLAGSRPATH = -Xlinker -rpath=$(LIBDIRRPATH)
   # RPATH to common lib when linking cuda/cpp libs
-  override CXXLIBFLAGSRPATH2 = -Wl,-rpath,'$$ORIGIN'
-  override CULIBFLAGSRPATH2 = -Xlinker -rpath,'$$ORIGIN'
+  override CXXLIBFLAGSRPATH2 = -Wl,-rpath='$$ORIGIN'
+  override CULIBFLAGSRPATH2 = -Xlinker -rpath='$$ORIGIN'
 endif
 
 # Setting LD_LIBRARY_PATH or DYLD_LIBRARY_PATH in the RUNTIME is no longer necessary (neither on Linux nor on Mac)
@@ -461,7 +525,7 @@ override RUNTIME =
 cxx_main=$(BUILDDIR)/check.exe
 fcxx_main=$(BUILDDIR)/fcheck.exe
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 cu_main=$(BUILDDIR)/gcheck.exe
 fcu_main=$(BUILDDIR)/fgcheck.exe
 else
@@ -492,14 +556,15 @@ $(BUILDDIR)/.build.$(TAG):
 	@touch $(BUILDDIR)/.build.$(TAG)
 
 # Generic target and build rules: objects from CUDA compilation
-ifneq ($(NVCC),)
+# NB: CCBUILDRULEFLAGS includes "-x cu" for nvcc and "-x hip" for hipcc (#810)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/%.o : %.cu *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $(CUBUILDRULEFLAGS) $< -o $@
 
 $(BUILDDIR)/%_cu.o : %.cc *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c -x cu $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $(CCBUILDRULEFLAGS) $< -o $@
 endif
 
 # Generic target and build rules: objects from C++ compilation
@@ -509,11 +574,14 @@ $(BUILDDIR)/%.o : %.cc *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	$(CXX) $(CPPFLAGS) $(CXXFLAGS) -fPIC -c $< -o $@
 
 # Apply special build flags only to CrossSectionKernel.cc and gCrossSectionKernel.cu (no fast math, see #117 and #516)
+# Added edgecase for HIP compilation
 ifeq ($(shell $(CXX) --version | grep ^nvc++),)
 $(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS := $(filter-out -ffast-math,$(CXXFLAGS))
 $(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS += -fno-fast-math
-ifneq ($(NVCC),)
-$(BUILDDIR)/gCrossSectionKernels.o: CUFLAGS += -Xcompiler -fno-fast-math
+ifeq ($(findstring nvcc,$(GPUCC)),nvcc)
+  $(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -Xcompiler -fno-fast-math
+else
+  $(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -fno-fast-math
 endif
 endif
 
@@ -530,10 +598,10 @@ ifeq ($(RNDGEN),hasCurand)
 $(BUILDDIR)/CurandRandomNumberKernel.o: CXXFLAGS += $(CUINC)
 endif
 
-# Avoid "warning: builtin __has_trivial_... is deprecated; use __is_trivially_... instead" in nvcc with icx2023 (#592)
+# Avoid "warning: builtin __has_trivial_... is deprecated; use __is_trivially_... instead" in GPUCC with icx2023 (#592)
 ifneq ($(shell $(CXX) --version | egrep '^(Intel)'),)
-ifneq ($(NVCC),)
-CUFLAGS += -Xcompiler -Wno-deprecated-builtins
+ifneq ($(GPUCC),)
+GPUFLAGS += -Wno-deprecated-builtins
 endif
 endif
 
@@ -541,8 +609,8 @@ endif
 # This patch does remove the warning, but I prefer to keep it disabled for the moment...
 ###ifneq ($(shell $(CXX) --version | egrep '^(clang|Apple clang|Intel)'),)
 ###$(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS += -Wno-overriding-t-option
-###ifneq ($(NVCC),)
-###$(BUILDDIR)/gCrossSectionKernels.o: CUFLAGS += -Xcompiler -Wno-overriding-t-option
+###ifneq ($(GPUCC),)
+###$(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -Xcompiler -Wno-overriding-t-option
 ###endif
 ###endif
 
@@ -569,7 +637,7 @@ MG5AMC_CXXLIB = mg5amc_$(processid_short)_cpp
 cxx_objects_lib=$(BUILDDIR)/CPPProcess.o $(BUILDDIR)/MatrixElementKernels.o $(BUILDDIR)/BridgeKernels.o $(BUILDDIR)/CrossSectionKernels.o
 cxx_objects_exe=$(BUILDDIR)/CommonRandomNumberKernel.o $(BUILDDIR)/RamboSamplingKernels.o
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 MG5AMC_CULIB = mg5amc_$(processid_short)_cuda
 cu_objects_lib=$(BUILDDIR)/gCPPProcess.o $(BUILDDIR)/gMatrixElementKernels.o $(BUILDDIR)/gBridgeKernels.o $(BUILDDIR)/gCrossSectionKernels.o
 cu_objects_exe=$(BUILDDIR)/gCommonRandomNumberKernel.o $(BUILDDIR)/gRamboSamplingKernels.o
@@ -581,11 +649,17 @@ $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: cxx_objects_lib += $(BUILDDIR)/fbridge.o
 $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib)
 	$(CXX) -shared -o $@ $(cxx_objects_lib) $(CXXLIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: $(BUILDDIR)/fbridge_cu.o
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: cu_objects_lib += $(BUILDDIR)/fbridge_cu.o
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cu_objects_lib)
-	$(NVCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+# Bypass std::filesystem completely to ease portability on LUMI #803
+#ifneq ($(findstring hipcc,$(GPUCC)),)
+#	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB) -lstdc++fs
+#else
+#	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+#endif
 endif
 
 #-------------------------------------------------------------------------------
@@ -602,16 +676,16 @@ $(cxx_main): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PAT
 $(cxx_main): $(BUILDDIR)/check_sa.o $(LIBDIR)/lib$(MG5AMC_CXXLIB).so $(cxx_objects_exe) $(BUILDDIR)/CurandRandomNumberKernel.o
 	$(CXX) -o $@ $(BUILDDIR)/check_sa.o $(OMPFLAGS) -ldl -pthread $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe) $(BUILDDIR)/CurandRandomNumberKernel.o $(CURANDLIBFLAGS)
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(cu_main): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(cu_main): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(cu_main): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(cu_main): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 else ifneq ($(shell $(CXX) --version | grep ^nvc++),) # support nvc++ #531
 $(cu_main): LIBFLAGS += -L$(patsubst %bin/nvc++,%lib,$(subst ccache ,,$(CXX))) -lnvhpcatm -lnvcpumath -lnvc
 endif
 $(cu_main): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(cu_main): $(BUILDDIR)/gcheck_sa.o $(LIBDIR)/lib$(MG5AMC_CULIB).so $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o
-	$(NVCC) -o $@ $(BUILDDIR)/gcheck_sa.o $(CUARCHFLAGS) $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o $(CURANDLIBFLAGS)
+	$(GPUCC) -o $@ $(BUILDDIR)/gcheck_sa.o $(CUARCHFLAGS) $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o $(CURANDLIBFLAGS)
 endif
 
 #-------------------------------------------------------------------------------
@@ -635,19 +709,27 @@ $(fcxx_main): LIBFLAGS += -L$(shell dirname $(shell $(FC) --print-file-name libg
 endif
 $(fcxx_main): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(fcxx_main): $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler.o $(LIBDIR)/lib$(MG5AMC_CXXLIB).so $(cxx_objects_exe)
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(BUILDDIR)/fcheck_sa.o $(OMPFLAGS) $(BUILDDIR)/fsampler.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe) -lstdc++
+else
 	$(CXX) -o $@ $(BUILDDIR)/fcheck_sa.o $(OMPFLAGS) $(BUILDDIR)/fsampler.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe)
+endif
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(fcu_main): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(fcu_main): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(fcu_main): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(fcu_main): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 endif
 ifeq ($(UNAME_S),Darwin)
 $(fcu_main): LIBFLAGS += -L$(shell dirname $(shell $(FC) --print-file-name libgfortran.dylib)) # add path to libgfortran on Mac #375
 endif
 $(fcu_main): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(fcu_main): $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBDIR)/lib$(MG5AMC_CULIB).so $(cu_objects_exe)
-	$(NVCC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe)
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) -lstdc++ -L$(shell dirname $(shell $(GPUCC) -print-prog-name=clang))/../../lib -lamdhip64
+else
+	$(GPUCC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe)
+endif
 endif
 
 #-------------------------------------------------------------------------------
@@ -659,7 +741,7 @@ $(BUILDDIR)/testxxx.o: testxxx_cc_ref.txt
 $(testmain): $(BUILDDIR)/testxxx.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/testxxx.o # Comment out this line to skip the C++ test of xxx functions
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/testxxx_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/testxxx_cu.o: INCFLAGS += $(GTESTINC)
 $(BUILDDIR)/testxxx_cu.o: testxxx_cc_ref.txt
@@ -672,7 +754,7 @@ $(BUILDDIR)/testmisc.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/testmisc.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/testmisc.o # Comment out this line to skip the C++ miscellaneous tests
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/testmisc_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/testmisc_cu.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/testmisc_cu.o
@@ -684,12 +766,12 @@ $(BUILDDIR)/runTest.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/runTest.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/runTest.o
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/runTest_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/runTest_cu.o: INCFLAGS += $(GTESTINC)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(testmain): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(testmain): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(testmain): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(testmain): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 else ifneq ($(shell $(CXX) --version | grep ^nvc++),) # support nvc++ #531
 $(testmain): LIBFLAGS += -L$(patsubst %bin/nvc++,%lib,$(subst ccache ,,$(CXX))) -lnvhpcatm -lnvcpumath -lnvc
 endif
@@ -713,14 +795,23 @@ $(testmain): LIBFLAGS += -lgomp
 endif
 endif
 
-ifeq ($(NVCC),) # link only runTest.o
+# Bypass std::filesystem completely to ease portability on LUMI #803
+#ifneq ($(findstring hipcc,$(GPUCC)),)
+#$(testmain): LIBFLAGS += -lstdc++fs
+#endif
+
+ifeq ($(GPUCC),) # link only runTest.o
 $(testmain): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(testmain): $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib) $(cxx_objects_exe) $(GTESTLIBS)
 	$(CXX) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) -ldl -pthread $(LIBFLAGS)
 else # link both runTest.o and runTest_cu.o
 $(testmain): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(testmain): $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) $(GTESTLIBS)
-	$(NVCC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) -lcuda
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) $(CUDATESTFLAGS) -lstdc++ -lpthread  -L$(shell dirname $(shell $(GPUCC) -print-prog-name=clang))/../../lib -lamdhip64
+else
+	$(GPUCC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) $(CUDATESTFLAGS)
+endif
 endif
 
 # Use target gtestlibs to build only googletest
@@ -829,9 +920,9 @@ ifeq ($(USECCACHE),1)
 	ccache --version | head -1
 endif
 	@echo ""
-	@echo NVCC=$(NVCC)
-ifneq ($(NVCC),)
-	$(NVCC) --version
+	@echo GPUCC=$(GPUCC)
+ifneq ($(GPUCC),)
+	$(GPUCC) --version
 endif
 	@echo ""
 	@echo CXX=$(CXX)
@@ -850,7 +941,7 @@ endif
 
 # Target: check (run the C++ test executable)
 # [NB THIS IS WHAT IS USED IN THE GITHUB CI!]
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 check: runTest cmpFcheck cmpFGcheck
 else
 check: runTest cmpFcheck
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/fbridge.cc b/epochX/cudacpp/gg_tt.sa/SubProcesses/fbridge.cc
index 2d2b36d560..22ce3f5115 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/fbridge.cc
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/fbridge.cc
@@ -1,11 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Roiser (Oct 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "Bridge.h"
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 
 extern "C"
 {
@@ -22,7 +22,7 @@ extern "C"
    * Using the same Fortran MadEvent code, linking to the hetrerogeneous library would allow access to both CPU and GPU implementations.
    * The specific heterogeneous configuration (how many GPUs, how many threads on each CPU, etc) could be loaded in CUDA/C++ from a data file.
    */
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -46,8 +46,8 @@ extern "C"
    */
   void fbridgecreate_( CppObjectInFortran** ppbridge, const int* pnevtF, const int* pnparF, const int* pnp4F )
   {
-#ifdef __CUDACC__
-    CudaRuntime::setUp();
+#ifdef MGONGPUCPP_GPUIMPL
+    GpuRuntime::setUp();
 #endif
     // (NB: CPPProcess::initProc no longer needs to be executed here because it is called in the Bridge constructor)
     // FIXME: disable OMP in Bridge when called from Fortran
@@ -65,8 +65,8 @@ extern "C"
     Bridge<FORTRANFPTYPE>* pbridge = dynamic_cast<Bridge<FORTRANFPTYPE>*>( *ppbridge );
     if( pbridge == 0 ) throw std::runtime_error( "fbridgedelete_: invalid Bridge address" );
     delete pbridge;
-#ifdef __CUDACC__
-    CudaRuntime::tearDown();
+#ifdef MGONGPUCPP_GPUIMPL
+    GpuRuntime::tearDown();
 #endif
   }
 
@@ -96,7 +96,7 @@ extern "C"
   {
     Bridge<FORTRANFPTYPE>* pbridge = dynamic_cast<Bridge<FORTRANFPTYPE>*>( *ppbridge );
     if( pbridge == 0 ) throw std::runtime_error( "fbridgesequence_: invalid Bridge address" );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Use the device/GPU implementation in the CUDA library
     // (there is also a host implementation in this library)
     pbridge->gpu_sequence( momenta, gs, rndhel, rndcol, *pchannelId, mes, selhel, selcol );
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/fsampler.cc b/epochX/cudacpp/gg_tt.sa/SubProcesses/fsampler.cc
index 2fb445372d..3743934f41 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/fsampler.cc
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/fsampler.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Feb 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "mgOnGpuConfig.h"
 
@@ -13,7 +13,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -40,7 +40,7 @@ namespace mg5amcCpu
   private:
     const int m_nevt; // The number of events in each iteration
     int m_iiter;      // The iteration counter (for random number seeding)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     HostBufferRndNumMomenta m_hstRndmom; // Memory buffers for random numbers
     HostBufferMomenta m_hstMomenta;      // Memory buffers for momenta
     HostBufferWeights m_hstWeights;      // Memory buffers for sampling weights
@@ -105,7 +105,7 @@ namespace mg5amcCpu
 
 extern "C"
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/runTest.cc b/epochX/cudacpp/gg_tt.sa/SubProcesses/runTest.cc
index d4a760a71b..de327f2321 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/runTest.cc
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/runTest.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Hageboeck (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*xxx to run only testxxx.cc tests
 //----------------------------------------------------------------------------
@@ -18,7 +18,7 @@
 #include "RandomNumberKernels.h"
 #include "epoch_process_id.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using namespace mg5amcGpu;
 #else
 using namespace mg5amcCpu;
@@ -35,7 +35,7 @@ struct CUDA_CPU_TestBase : public TestDriverBase
     : TestDriverBase( npar, refFileName ) {}
 };
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 struct CPUTest : public CUDA_CPU_TestBase
 {
   // Struct data members (process, and memory structures for random numbers, momenta, matrix elements and weights on host and device)
@@ -119,7 +119,7 @@ struct CPUTest : public CUDA_CPU_TestBase
 };
 #endif
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 struct CUDATest : public CUDA_CPU_TestBase
 {
   // Reset the device when our test goes out of scope. Note that this should happen after
@@ -128,7 +128,7 @@ struct CUDATest : public CUDA_CPU_TestBase
   {
     ~DeviceReset()
     {
-      checkCuda( cudaDeviceReset() ); // this is needed by cuda-memcheck --leak-check full
+      checkGpu( gpuDeviceReset() ); // this is needed by cuda-memcheck --leak-check full
     }
   } deviceResetter;
 
@@ -256,7 +256,7 @@ INSTANTIATE_TEST_SUITE_P( prefix, \
                           test_suite_name, \
                           testing::Values( new CUDATest( MG_EPOCH_REFERENCE_FILE_NAME ) ) );
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 MG_INSTANTIATE_TEST_SUITE_GPU( XTESTID_GPU( MG_EPOCH_PROCESS_ID ), MadgraphTest );
 #else
 MG_INSTANTIATE_TEST_SUITE_CPU( XTESTID_CPU( MG_EPOCH_PROCESS_ID ), MadgraphTest );
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/testmisc.cc b/epochX/cudacpp/gg_tt.sa/SubProcesses/testmisc.cc
index 895d6eeb56..ba9e59a8a3 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/testmisc.cc
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/testmisc.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*misc to run only testmisc.cc tests
 //----------------------------------------------------------------------------
@@ -17,7 +17,7 @@
 #include <sstream>
 #include <typeinfo>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #define TESTID( s ) s##_GPU_MISC
 #else
 #define TESTID( s ) s##_CPU_MISC
@@ -26,7 +26,7 @@
 #define XTESTID( s ) TESTID( s )
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -59,7 +59,7 @@ namespace mg5amcCpu
 
 TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testmisc )
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_tt.sa/SubProcesses/testxxx.cc b/epochX/cudacpp/gg_tt.sa/SubProcesses/testxxx.cc
index 3361fe5aa9..e5167de00c 100644
--- a/epochX/cudacpp/gg_tt.sa/SubProcesses/testxxx.cc
+++ b/epochX/cudacpp/gg_tt.sa/SubProcesses/testxxx.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Apr 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*xxx to run only testxxx.cc tests
 //----------------------------------------------------------------------------
@@ -24,7 +24,7 @@
 #include <iomanip>
 #include <iostream>
 #include <vector>
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #define TESTID( s ) s##_GPU_XXX
 #else
 #define TESTID( s ) s##_CPU_XXX
@@ -32,7 +32,7 @@
 
 #define XTESTID( s ) TESTID( s )
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -42,7 +42,7 @@ namespace mg5amcCpu
   int FPEhandlerIevt = -1;
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU): '" << FPEhandlerMessage << "' ievt=" << FPEhandlerIevt << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU neppV=" << neppV << "): '" << FPEhandlerMessage << "' ievt=" << FPEhandlerIevt << std::endl;
@@ -53,7 +53,7 @@ namespace mg5amcCpu
 
 TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -77,7 +77,7 @@ TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
   assert( nevt % neppM == 0 ); // nevt must be a multiple of neppM
   assert( nevt % neppV == 0 ); // nevt must be a multiple of neppV
   // Fill in the input momenta
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   mg5amcGpu::PinnedHostBufferMomenta hstMomenta( nevt ); // AOSOA[npagM][npar=4][np4=4][neppM]
 #else
   mg5amcCpu::HostBufferMomenta hstMomenta( nevt ); // AOSOA[npagM][npar=4][np4=4][neppM]
@@ -322,7 +322,7 @@ TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
   {
     for( int ievt = 0; ievt < nevt; ievt++ )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       using namespace mg5amcGpu;
 #else
       using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_tt.sa/src/HelAmps_sm.h b/epochX/cudacpp/gg_tt.sa/src/HelAmps_sm.h
index 55f43bb43a..add8fce575 100644
--- a/epochX/cudacpp/gg_tt.sa/src/HelAmps_sm.h
+++ b/epochX/cudacpp/gg_tt.sa/src/HelAmps_sm.h
@@ -5,7 +5,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -28,7 +28,7 @@
 //#include <iomanip>
 //#include <iostream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt.sa/src/Parameters_sm.cc b/epochX/cudacpp/gg_tt.sa/src/Parameters_sm.cc
index a9bc93ff98..c5dd6e7e4c 100644
--- a/epochX/cudacpp/gg_tt.sa/src/Parameters_sm.cc
+++ b/epochX/cudacpp/gg_tt.sa/src/Parameters_sm.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -17,7 +17,7 @@
 #include <iomanip>
 #include <iostream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using namespace mg5amcGpu;
 #else
 using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_tt.sa/src/Parameters_sm.h b/epochX/cudacpp/gg_tt.sa/src/Parameters_sm.h
index 932f123fea..5f2f4391b9 100644
--- a/epochX/cudacpp/gg_tt.sa/src/Parameters_sm.h
+++ b/epochX/cudacpp/gg_tt.sa/src/Parameters_sm.h
@@ -27,7 +27,7 @@
 #include "read_slha.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -93,7 +93,7 @@ namespace mg5amcCpu
 #include <limits>
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -217,7 +217,7 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -236,7 +236,7 @@ namespace mg5amcCpu
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wunused-variable"  // e.g. <<warning: unused variable ‘mdl_G__exp__2’ [-Wunused-variable]>>
 #pragma GCC diagnostic ignored "-Wunused-parameter" // e.g. <<warning: unused parameter ‘G’ [-Wunused-parameter]>>
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #pragma nv_diagnostic push
 #pragma nv_diag_suppress 177 // e.g. <<warning #177-D: variable "mdl_G__exp__2" was declared but never referenced>>
 #endif
@@ -263,7 +263,7 @@ namespace mg5amcCpu
       // End SM implementation - no special handling of vectors of floats as in EFT (#439)
       return out;
     }
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #pragma GCC diagnostic pop
 #pragma nv_diagnostic pop
 #endif
diff --git a/epochX/cudacpp/gg_tt.sa/src/cudacpp_src.mk b/epochX/cudacpp/gg_tt.sa/src/cudacpp_src.mk
index d4cc628aec..b2b9da5288 100644
--- a/epochX/cudacpp/gg_tt.sa/src/cudacpp_src.mk
+++ b/epochX/cudacpp/gg_tt.sa/src/cudacpp_src.mk
@@ -19,7 +19,7 @@ SHELL := /bin/bash
 #=== Configure common compiler flags for CUDA and C++
 
 INCFLAGS = -I.
-OPTFLAGS = -O3 # this ends up in CUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
+OPTFLAGS = -O3 # this ends up in GPUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
 
 #-------------------------------------------------------------------------------
 
@@ -45,13 +45,13 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure the CUDA compiler (note: NVCC is already exported including ccache)
+#=== Configure the CUDA compiler (note: GPUCC is already exported including ccache)
 
-###$(info NVCC=$(NVCC))
+###$(info GPUCC=$(GPUCC))
 
 #-------------------------------------------------------------------------------
 
-#=== Configure ccache for C++ builds (note: NVCC is already exported including ccache)
+#=== Configure ccache for C++ builds (note: GPUCC is already exported including ccache)
 
 # Enable ccache if USECCACHE=1
 ifeq ($(USECCACHE)$(shell echo $(CXX) | grep ccache),1)
@@ -92,6 +92,13 @@ endif
 ###$(info OMPFLAGS=$(OMPFLAGS))
 CXXFLAGS += $(OMPFLAGS)
 
+# Add correct flags for nvcc (-x cu) and hipcc (-x hip) for GPU code (see #810)
+ifeq ($(findstring nvcc,$(GPUCC)),nvcc)
+  GPUFLAGS += -Xcompiler -fPIC -c -x cu
+else ifeq ($(findstring hipcc,$(GPUCC)),hipcc)
+  GPUFLAGS += -fPIC -c -x hip
+endif
+
 # Set the build flags appropriate to each AVX choice (example: "make AVX=none")
 # [NB MGONGPU_PVW512 is needed because "-mprefer-vector-width=256" is not exposed in a macro]
 # [See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96476]
@@ -253,20 +260,20 @@ $(BUILDDIR)/%.o : %.cc *.h $(BUILDDIR)/.build.$(TAG)
 # Generic target and build rules: objects from CUDA compilation
 $(BUILDDIR)/%_cu.o : %.cc *.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c -x cu $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $< -o $@
 
 #-------------------------------------------------------------------------------
 
 cxx_objects=$(addprefix $(BUILDDIR)/, Parameters_sm.o read_slha.o)
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 cu_objects=$(addprefix $(BUILDDIR)/, Parameters_sm_cu.o)
 endif
 
 # Target (and build rules): common (src) library
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so : $(cxx_objects) $(cu_objects)
 	@if [ ! -d $(LIBDIR) ]; then echo "mkdir -p $(LIBDIR)"; mkdir -p $(LIBDIR); fi
-	$(NVCC) -shared -o $@ $(cxx_objects) $(cu_objects) $(LDFLAGS)
+	$(GPUCC) -shared -o $@ $(cxx_objects) $(cu_objects) $(LDFLAGS)
 else
 $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so : $(cxx_objects)
 	@if [ ! -d $(LIBDIR) ]; then echo "mkdir -p $(LIBDIR)"; mkdir -p $(LIBDIR); fi
diff --git a/epochX/cudacpp/gg_tt.sa/src/mgOnGpuConfig.h b/epochX/cudacpp/gg_tt.sa/src/mgOnGpuConfig.h
index b247654dcf..475749ca7c 100644
--- a/epochX/cudacpp/gg_tt.sa/src/mgOnGpuConfig.h
+++ b/epochX/cudacpp/gg_tt.sa/src/mgOnGpuConfig.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jul 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUCONFIG_H
 #define MGONGPUCONFIG_H 1
@@ -10,12 +10,26 @@
 // There are two different code bases for standalone_cudacpp (without multichannel) and madevent+cudacpp (with multichannel)
 #undef MGONGPU_SUPPORTS_MULTICHANNEL
 
+// Is this a GPU (CUDA, HIP) or CPU implementation?
+#ifdef __CUDACC__
+#define MGONGPUCPP_GPUIMPL cuda
+#elif defined __HIPCC__
+#define MGONGPUCPP_GPUIMPL hip
+#include "hip/hip_runtime.h" // needed for blockDim, blockIdx, threadIdx: better in mgOnGpuConfig.h than in GpuAbstraction.h
+#else
+#undef MGONGPUCPP_GPUIMPL
+#endif
+
 // ** NB1 Throughputs (e.g. 6.8E8) are events/sec for "./gcheck.exe -p 65536 128 12"
 // ** NB2 Baseline on b7g47n0004 fluctuates (probably depends on load on other VMs)
 
 // Choose if curand is supported for generating random numbers
+// For HIP, by default, do not use curand (common random numbers will be used instead)
 // For both CUDA and C++, by default, do not inline, but allow this macro to be set from outside with e.g. -DMGONGPU_HAS_NO_CURAND
-// (there exist CUDA installations, e.g. using the HPC package, which do not include curand - see PR #784)
+// (there exist CUDA installations, e.g. using the HPC package, which do not include curand - see PR #784 and #785)
+#if defined __HIPCC__
+#define MGONGPU_HAS_NO_CURAND 1
+#else
 //#ifdef __CUDACC__
 //#undef MGONGPU_HAS_NO_CURAND // default
 ////#define MGONGPU_HAS_NO_CURAND 1
@@ -23,6 +37,7 @@
 //#undef MGONGPU_HAS_NO_CURAND // default
 ////#define MGONGPU_HAS_NO_CURAND 1
 //#endif
+#endif
 
 // Choose floating point precision (for everything but color algebra #537)
 // If one of these macros has been set from outside with e.g. -DMGONGPU_FPTYPE_FLOAT, nothing happens (issue #167)
@@ -54,23 +69,31 @@
 //#undef MGONGPU_HARDCODE_PARAM // default
 ////#define MGONGPU_HARDCODE_PARAM 1
 
-// Complex type in c++: std::complex or cxsmpl (CHOOSE ONLY ONE)
-#ifndef __CUDACC__
-//#define MGONGPU_CPPCXTYPE_STDCOMPLEX 1 // ~8 percent slower on float, same on double (5.1E6/double, 9.4E6/float)
-#define MGONGPU_CPPCXTYPE_CXSMPL 1 // new default (5.1E6/double, 10.2E6/float)
-#endif
-
-// Complex type in cuda: thrust or cucomplex or cxsmpl (CHOOSE ONLY ONE)
+// Complex type in CUDA: thrust or cucomplex or cxsmpl (CHOOSE ONLY ONE)
+// (NB THIS IS MGONGPU_*CU*CXTYPE_xxx)
 #ifdef __CUDACC__
 #define MGONGPU_CUCXTYPE_THRUST 1 // default (~1.15E9/double, ~3.2E9/float)
 //#define MGONGPU_CUCXTYPE_CUCOMPLEX 1 // ~10 percent slower (1.03E9/double, ~2.8E9/float)
 //#define MGONGPU_CUCXTYPE_CXSMPL 1 // ~10 percent slower (1.00E9/double, ~2.9E9/float)
+
+// Complex type in HIP: cxsmpl (ONLY ONE OPTION POSSIBLE? #810)
+// (NB THIS IS MGONGPU_*HIP*CXTYPE_xxx)
+#elif defined __HIPCC__
+#define MGONGPU_HIPCXTYPE_CXSMPL 1 // default for HIP
+
+// Complex type in C++: std::complex or cxsmpl (CHOOSE ONLY ONE)
+// (NB THIS IS MGONGPU_*CPP*CXTYPE_xxx)
+#else
+//#define MGONGPU_CPPCXTYPE_STDCOMPLEX 1 // ~8 percent slower on float, same on double (5.1E6/double, 9.4E6/float)
+#define MGONGPU_CPPCXTYPE_CXSMPL 1 // new default (5.1E6/double, 10.2E6/float)
 #endif
 
-// Cuda nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
+// CUDA nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
 #ifdef __CUDACC__
-#undef MGONGPU_NSIGHT_DEBUG // default
+#undef MGONGPU_NSIGHT_DEBUG // default in CUDA
 //#define MGONGPU_NSIGHT_DEBUG 1
+#else
+#undef MGONGPU_NSIGHT_DEBUG // only option in HIP or C++
 #endif
 
 // SANITY CHECKS (floating point precision for everything but color algebra #537)
@@ -86,17 +109,21 @@
 #error You cannot use double precision for color algebra and single precision elsewhere
 #endif
 
-// SANITY CHECKS (c++ complex number implementation)
-#ifndef __CUDACC__
-#if defined MGONGPU_CPPCXTYPE_STDCOMPLEX and defined MGONGPU_CPPCXTYPE_CXSMPL
-#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CPPCXTYPE_STDCOMPLEX or MGONGPU_CPPCXTYPE_CXSMPL
+// SANITY CHECKS (CUDA complex number implementation)
+#ifdef __CUDACC__
+#if defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CUCOMPLEX
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX for CUDA
+#elif defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CXSMPL for CUDA
+#elif defined MGONGPU_CUCXTYPE_CUCOMPLEX and defined MGONGPU_CUCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE OF MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL for CUDA
 #endif
 #endif
 
-// SANITY CHECKS (cuda complex number implementation)
-#ifdef __CUDACC__
-#if defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CUCOMPLEX and defined MGONGPU_CUCXTYPE_CXSMPL
-#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL
+// SANITY CHECKS (C++ complex number implementation)
+#ifndef MGONGPUCPP_GPUIMPL
+#if defined MGONGPU_CPPCXTYPE_STDCOMPLEX and defined MGONGPU_CPPCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CPPCXTYPE_STDCOMPLEX or MGONGPU_CPPCXTYPE_CXSMPL for C++
 #endif
 #endif
 
@@ -134,7 +161,7 @@ namespace mgOnGpu
   // Alignment requirement for using reinterpret_cast with SIMD vectorized code
   // (using reinterpret_cast with non aligned memory may lead to segmentation faults!)
   // Only needed for C++ code but can be enforced also in NVCC builds of C++ code using CUDA>=11.2 and C++17 (#318, #319, #333)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   constexpr int cppAlign = 64; // alignment requirement for SIMD vectorization (64-byte i.e. 512-bit)
 #endif
 
@@ -145,7 +172,7 @@ using mgOnGpu::fptype;
 using mgOnGpu::fptype2;
 
 // C++ SIMD vectorization width (this will be used to set neppV)
-#ifdef __CUDACC__ // CUDA implementation has no SIMD
+#ifdef MGONGPUCPP_GPUIMPL // CUDA and HIP implementations have no SIMD
 #undef MGONGPU_CPPSIMD
 #elif defined __AVX512VL__ && defined MGONGPU_PVW512 // C++ "512z" AVX512 with 512 width (512-bit ie 64-byte): 8 (DOUBLE) or 16 (FLOAT)
 #ifdef MGONGPU_FPTYPE_DOUBLE
@@ -175,9 +202,9 @@ using mgOnGpu::fptype2;
 #undef MGONGPU_CPPSIMD
 #endif
 
-// Cuda nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
+// CUDA nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
 // Arguments (not used so far): text is __FUNCTION__, code is 0 (start) or 1 (end)
-#if defined __CUDACC__ && defined MGONGPU_NSIGHT_DEBUG /* clang-format off */
+#if defined __CUDA__ && defined MGONGPU_NSIGHT_DEBUG /* clang-format off */
 #define mgDebugDeclare() __shared__ float mgDebugCounter[mgOnGpu::ntpbMAX];
 #define mgDebugInitialise() { mgDebugCounter[threadIdx.x] = 0; }
 #define mgDebug( code, text ) { mgDebugCounter[threadIdx.x] += 1; }
@@ -189,8 +216,8 @@ using mgOnGpu::fptype2;
 #define mgDebugFinalise() { /*noop*/ }
 #endif /* clang-format on */
 
-// Define empty CUDA declaration specifiers for C++
-#ifndef __CUDACC__
+// Define empty CUDA/HIP declaration specifiers for C++
+#ifndef MGONGPUCPP_GPUIMPL
 #define __global__
 #define __host__
 #define __device__
diff --git a/epochX/cudacpp/gg_tt.sa/src/mgOnGpuCxtypes.h b/epochX/cudacpp/gg_tt.sa/src/mgOnGpuCxtypes.h
index ca9a9f00c0..7ede1dbfae 100644
--- a/epochX/cudacpp/gg_tt.sa/src/mgOnGpuCxtypes.h
+++ b/epochX/cudacpp/gg_tt.sa/src/mgOnGpuCxtypes.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022, based on earlier work by D. Smith) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUCXTYPES_H
 #define MGONGPUCXTYPES_H 1
@@ -19,7 +19,7 @@
 #include <complex>
 
 // Complex type in cuda: thrust or cucomplex or cxsmpl
-#ifdef __CUDACC__
+#ifdef __CUDACC__ // this must be __CUDAC__ (not MGONGPUCPP_GPUIMPL)
 #if defined MGONGPU_CUCXTYPE_THRUST
 #pragma clang diagnostic push
 #pragma clang diagnostic ignored "-Wtautological-compare" // for icpx2021/clang13 (https://stackoverflow.com/a/15864661)
@@ -30,8 +30,13 @@
 #elif not defined MGONGPU_CUCXTYPE_CXSMPL
 #error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL
 #endif
+// Complex type in HIP: cxsmpl
+#elif defined __HIPCC__
+#if not defined MGONGPU_HIPCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_HIPCXTYPE_CXSMPL
+#endif
 #else
-// Complex type in c++: std::complex or cxsmpl
+// Complex type in c++ or HIP: std::complex or cxsmpl
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
 #include <cmath>
 #elif not defined MGONGPU_CPPCXTYPE_CXSMPL
@@ -82,7 +87,7 @@ namespace mgOnGpu /* clang-format off */
 using mgOnGpu::cxsmpl;
 
 // Printout to stream for user defined types
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -92,7 +97,7 @@ namespace mg5amcCpu
   inline __host__ std::ostream&
   operator<<( std::ostream& out, const cxsmpl<FP>& c )
   {
-    out << std::complex( c.real(), c.imag() );
+    out << std::complex<FP>( c.real(), c.imag() );
     return out;
   }
 
@@ -215,14 +220,14 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
   // --- Type definitions (complex type: cxtype)
-#ifdef __CUDACC__ // cuda
+#ifdef __CUDACC__ // this must be __CUDAC__ (not MGONGPUCPP_GPUIMPL)
 #if defined MGONGPU_CUCXTYPE_THRUST
   typedef thrust::complex<fptype> cxtype;
 #elif defined MGONGPU_CUCXTYPE_CUCOMPLEX
@@ -255,13 +260,13 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
-#if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+#if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
 
   //------------------------------
   // CUDA or C++ - using cxsmpl
@@ -303,11 +308,11 @@ namespace mg5amcCpu
     return cxmake( c.real(), c.imag() );
   }
 
-#endif // #if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+#endif // #if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
 
   //==========================================================================
 
-#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_THRUST // cuda + thrust
+#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_THRUST // cuda + thrust (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // CUDA - using thrust::complex
@@ -347,7 +352,7 @@ namespace mg5amcCpu
 
   //==========================================================================
 
-#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_CUCOMPLEX // cuda + cucomplex
+#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_CUCOMPLEX // cuda + cucomplex (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // CUDA - using cuComplex
@@ -566,7 +571,7 @@ namespace mg5amcCpu
 
   //==========================================================================
 
-#if not defined __CUDACC__ and defined MGONGPU_CPPCXTYPE_STDCOMPLEX // c++ + stdcomplex
+#if not defined __CUDACC__ and defined MGONGPU_CPPCXTYPE_STDCOMPLEX // c++/hip + stdcomplex (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // C++ - using std::complex
@@ -633,7 +638,7 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt.sa/src/mgOnGpuFptypes.h b/epochX/cudacpp/gg_tt.sa/src/mgOnGpuFptypes.h
index 905c97d700..fa3a02664b 100644
--- a/epochX/cudacpp/gg_tt.sa/src/mgOnGpuFptypes.h
+++ b/epochX/cudacpp/gg_tt.sa/src/mgOnGpuFptypes.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUFPTYPES_H
 #define MGONGPUFPTYPES_H 1
@@ -12,7 +12,7 @@
 #include <cmath>
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL // cuda
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -20,7 +20,7 @@ namespace mg5amcCpu
 {
   //==========================================================================
 
-#ifdef __CUDACC__ // cuda
+#ifdef MGONGPUCPP_GPUIMPL // cuda
 
   //------------------------------
   // Floating point types - Cuda
@@ -64,11 +64,11 @@ namespace mg5amcCpu
 #endif
   }
 
-#endif // #ifdef __CUDACC__
+#endif // #ifdef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
   //------------------------------
   // Floating point types - C++
@@ -92,7 +92,7 @@ namespace mg5amcCpu
     return std::sqrt( f );
   }
 
-#endif // #ifndef __CUDACC__
+#endif // #ifndef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
diff --git a/epochX/cudacpp/gg_tt.sa/src/mgOnGpuVectors.h b/epochX/cudacpp/gg_tt.sa/src/mgOnGpuVectors.h
index e1299ba81e..cdae04326b 100644
--- a/epochX/cudacpp/gg_tt.sa/src/mgOnGpuVectors.h
+++ b/epochX/cudacpp/gg_tt.sa/src/mgOnGpuVectors.h
@@ -32,7 +32,7 @@
 #endif
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -131,7 +131,7 @@ namespace mg5amcCpu
 #endif
 #endif
 
-#else // i.e #ifndef MGONGPU_CPPSIMD (this includes #ifdef __CUDACC__)
+#else // i.e #ifndef MGONGPU_CPPSIMD (this includes #ifdef MGONGPUCPP_GPUIMPL)
 
   const int neppV = 1;
 
@@ -153,13 +153,13 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
   // Printout to stream for user defined types
 
@@ -805,11 +805,11 @@ namespace mg5amcCpu
 
 #endif // #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
 
-#endif // #ifndef __CUDACC__
+#endif // #ifndef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
   //------------------------------
   // Vector types - CUDA
@@ -853,12 +853,12 @@ namespace mg5amcCpu
     return mask;
   }
 
-#endif // #ifdef __CUDACC__
+#endif // #ifdef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
   // Scalar-or-vector types: scalar in CUDA, vector or scalar in C++
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   typedef bool bool_sv;
   typedef fptype fptype_sv;
   typedef fptype2 fptype2_sv;
@@ -879,7 +879,7 @@ namespace mg5amcCpu
 #endif
 
   // Scalar-or-vector zeros: scalar in CUDA, vector or scalar in C++
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   inline __host__ __device__ cxtype cxzero_sv(){ return cxtype( 0, 0 ); }
 #elif defined MGONGPU_CPPSIMD
   inline cxtype_v cxzero_sv() { return cxtype_v(); } // RRRR=0000 IIII=0000
diff --git a/epochX/cudacpp/gg_tt.sa/src/rambo.h b/epochX/cudacpp/gg_tt.sa/src/rambo.h
index e02ea52496..cd7e1008ea 100644
--- a/epochX/cudacpp/gg_tt.sa/src/rambo.h
+++ b/epochX/cudacpp/gg_tt.sa/src/rambo.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -18,7 +18,7 @@
 #include <iostream>
 
 // Simplified rambo version for 2 to N (with N>=2) processes with massless particles
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +83,7 @@ namespace mg5amcCpu
       static bool first = true;
       if( first )
       {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
         if constexpr( M_ACCESS::isOnDevice() ) // avoid
         {
           const int ievt0 = 0;
@@ -166,7 +166,7 @@ namespace mg5amcCpu
     wt = po2log;
     if( nparf != 2 ) wt = ( 2. * nparf - 4. ) * log( energy ) + z[nparf - 1];
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     // issue warnings if weight is too small or too large
     static int iwarn[5] = { 0, 0, 0, 0, 0 };
     if( wt < -180. )
diff --git a/epochX/cudacpp/gg_tt.sa/src/read_slha.cc b/epochX/cudacpp/gg_tt.sa/src/read_slha.cc
index 055b19a779..f8e46f2e66 100644
--- a/epochX/cudacpp/gg_tt.sa/src/read_slha.cc
+++ b/epochX/cudacpp/gg_tt.sa/src/read_slha.cc
@@ -11,7 +11,11 @@
 
 #include <algorithm>
 #include <cstdlib>
-#include <filesystem>
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
 #include <fstream>
 #include <iostream>
 
@@ -60,7 +64,14 @@ SLHAReader::read_slha_file( std::string file_name, bool verbose )
     {
       std::cout << "WARNING! Card file '" << file_name << "' does not exist:"
                 << " look for the file in directory $" << envpath << "='" << getenv( envpath ) << "'" << std::endl;
+      /*
+#ifdef __HIPCC__
+      const std::string file_name2 = std::experimental::filesystem::path( getenv( envpath ) ) / std::experimental::filesystem::path( file_name ).filename();
+#else
       const std::string file_name2 = std::filesystem::path( getenv( envpath ) ) / std::filesystem::path( file_name ).filename();
+#endif
+      */
+      const std::string file_name2 = std::string( getenv( envpath ) ) + "/" + file_name; // bypass std::filesystem #803
       param_card.open( file_name2.c_str(), std::ifstream::in );
       if( param_card.good() )
       {
diff --git a/epochX/cudacpp/gg_tt01g.mad/CODEGEN_mad_gg_tt01g_log.txt b/epochX/cudacpp/gg_tt01g.mad/CODEGEN_mad_gg_tt01g_log.txt
index b3d319e039..f5287cc1ca 100644
--- a/epochX/cudacpp/gg_tt01g.mad/CODEGEN_mad_gg_tt01g_log.txt
+++ b/epochX/cudacpp/gg_tt01g.mad/CODEGEN_mad_gg_tt01g_log.txt
@@ -52,7 +52,7 @@ Note that you can still compile and run aMC@NLO with the built-in PDFs
 
 Using default text editor "vi". Set another one in ./input/mg5_configuration.txt
 Using default eps viewer "evince". Set another one in ./input/mg5_configuration.txt
-No valid web browser found. Please set in ./input/mg5_configuration.txt
+Using default web browser "firefox". Set another one in ./input/mg5_configuration.txt
 import /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_mad_gg_tt01g.mg
 The import format was not given, so we guess it as command
 set stdout_level DEBUG
@@ -62,7 +62,7 @@ generate g g > t t~
 No model currently active, so we import the Standard Model
 INFO: load particles 
 INFO: load vertices 
-DEBUG: model prefixing  takes 0.005671977996826172 
+DEBUG: model prefixing  takes 0.005505561828613281 
 INFO: Restrict model sm with file models/sm/restrict_default.dat . 
 DEBUG: Simplifying conditional expressions 
 DEBUG: remove interactions: u s w+ at order: QED=1 
@@ -163,7 +163,7 @@ INFO: Please specify coupling orders to bypass this step.
 INFO: Trying coupling order WEIGHTED<=3: WEIGTHED IS QCD+2*QED 
 INFO: Trying process: g g > t t~ g WEIGHTED<=3 @2  
 INFO: Process has 16 diagrams 
-1 processes with 16 diagrams generated in 0.020 s
+1 processes with 16 diagrams generated in 0.019 s
 Total: 2 processes with 19 diagrams
 output madevent ../TMPOUT/CODEGEN_mad_gg_tt01g --hel_recycling=False --vector_size=32 --me_exporter=standalone_cudacpp
 Load PLUGIN.CUDACPP_OUTPUT
@@ -185,7 +185,7 @@ INFO: Generating Helas calls for process: g g > t t~ WEIGHTED<=2 @1
 INFO: Processing color information for process: g g > t t~ @1 
 INFO: Creating files in directory P2_gg_ttxg 
 DEBUG:  kwargs[prefix] = 0 [model_handling.py at line 1058] 
-DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7f4feb5031c0> [export_v4.py at line 6262] 
+DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7fe54cc4a1f0> [export_v4.py at line 6262] 
 INFO: Creating files in directory . 
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.h
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.cc
@@ -202,7 +202,7 @@ INFO: Generating Feynman diagrams for Process: g g > t t~ g WEIGHTED<=3 @2
 INFO: Finding symmetric diagrams for subprocess group gg_ttxg 
 INFO: Creating files in directory P1_gg_ttx 
 DEBUG:  kwargs[prefix] = 0 [model_handling.py at line 1058] 
-DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7f4feb492fd0> [export_v4.py at line 6262] 
+DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7fe54cc4a070> [export_v4.py at line 6262] 
 INFO: Creating files in directory . 
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.h
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.cc
@@ -217,8 +217,8 @@ INFO: Created files CPPProcess.h and CPPProcess.cc in directory ./.
 DEBUG:  vector, subproc_group,self.opt['vector_size'] =  32 True 32 [export_v4.py at line 1872] 
 INFO: Generating Feynman diagrams for Process: g g > t t~ WEIGHTED<=2 @1 
 INFO: Finding symmetric diagrams for subprocess group gg_ttx 
-Generated helas calls for 2 subprocesses (19 diagrams) in 0.044 s
-Wrote files for 46 helas calls in 0.247 s
+Generated helas calls for 2 subprocesses (19 diagrams) in 0.042 s
+Wrote files for 46 helas calls in 0.243 s
 ALOHA: aloha starts to compute helicity amplitudes
 ALOHA: aloha creates VVV1 routines
 ALOHA: aloha creates FFV1 routines
@@ -233,7 +233,7 @@ ALOHA: aloha creates FFV1 routines
 ALOHA: aloha creates VVVV1 set of routines with options: P0
 ALOHA: aloha creates VVVV3 set of routines with options: P0
 ALOHA: aloha creates VVVV4 set of routines with options: P0
-ALOHA: aloha creates 10 routines in  0.316 s
+ALOHA: aloha creates 10 routines in  0.311 s
 <class 'aloha.create_aloha.AbstractRoutine'> VVV1
 <class 'aloha.create_aloha.AbstractRoutine'> VVV1
 <class 'aloha.create_aloha.AbstractRoutine'> FFV1
@@ -283,10 +283,10 @@ Type "launch" to generate events from this process, or see
 Run "open index.html" to see more information about this process.
 quit
 
-real	0m2.334s
-user	0m2.083s
-sys	0m0.238s
-Code generation completed in 2 seconds
+real	0m2.369s
+user	0m2.050s
+sys	0m0.252s
+Code generation completed in 3 seconds
 ************************************************************
 *                                                          *
 *                      W E L C O M E to                    *
@@ -312,7 +312,7 @@ INFO: load configuration from /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/mg5amc
 INFO: load configuration from /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_mad_gg_tt01g/Cards/me5_configuration.txt  
 Using default text editor "vi". Set another one in ./input/mg5_configuration.txt
 Using default eps viewer "evince". Set another one in ./input/mg5_configuration.txt
-No valid web browser found. Please set in ./input/mg5_configuration.txt
+Using default web browser "firefox". Set another one in ./input/mg5_configuration.txt
 treatcards run
 quit
 INFO:  
@@ -342,7 +342,7 @@ INFO: load configuration from /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/mg5amc
 INFO: load configuration from /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_mad_gg_tt01g/Cards/me5_configuration.txt  
 Using default text editor "vi". Set another one in ./input/mg5_configuration.txt
 Using default eps viewer "evince". Set another one in ./input/mg5_configuration.txt
-No valid web browser found. Please set in ./input/mg5_configuration.txt
+Using default web browser "firefox". Set another one in ./input/mg5_configuration.txt
 treatcards param
 quit
 INFO:  
diff --git a/epochX/cudacpp/gg_tt01g.mad/COPYRIGHT b/epochX/cudacpp/gg_tt01g.mad/COPYRIGHT
index a134b5fef9..9036d9260a 100644
--- a/epochX/cudacpp/gg_tt01g.mad/COPYRIGHT
+++ b/epochX/cudacpp/gg_tt01g.mad/COPYRIGHT
@@ -15,6 +15,7 @@ The full development team currently includes the following authors :
   Stephan Hageboeck (CERN)
   Olivier Mattelaer (Universite Catholique de Louvain, original author)
   Stefan Roiser (CERN, original author)
+  Jorgen Teig (CERN)
   Andrea Valassi (CERN, original author)
   Zenny Wettersten (CERN)
 See https://github.com/madgraph5/madgraph4gpu for more details. For the full
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/Bridge.h b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/Bridge.h
index bf8b5e024d..f9ed70dfde 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/Bridge.h
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/Bridge.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Roiser (Nov 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef BRIDGE_H
 #define BRIDGE_H 1
@@ -14,16 +14,23 @@
 #include "MemoryAccessMomenta.h"  // for MemoryAccessMomenta::neppM
 #include "MemoryBuffers.h"        // for HostBufferMomenta, DeviceBufferMomenta etc
 
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
+
+#include <sys/stat.h> // bypass std::filesystem #803
+
 #include <algorithm>
 #include <cassert>
 #include <cmath>
 #include <cstring>
-#include <filesystem>
 #include <iostream>
 #include <memory>
 #include <type_traits>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +90,7 @@ namespace mg5amcCpu
     Bridge& operator=( const Bridge& ) = delete;
     Bridge& operator=( Bridge&& ) = delete;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     /**
      * Set the gpublocks and gputhreads for the gpusequence - throws if evnt != gpublocks*gputhreads
      * (this is needed for BridgeKernel tests rather than for actual production use in Fortran)
@@ -150,7 +157,7 @@ namespace mg5amcCpu
     unsigned int m_nevt; // number of events
     int m_nGoodHel;      // the number of good helicities (-1 initially when they have not yet been calculated)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     int m_gputhreads; // number of gpu threads (default set from number of events, can be modified)
     int m_gpublocks;  // number of gpu blocks (default set from number of events, can be modified)
     DeviceBuffer<FORTRANFPTYPE, sizePerEventMomenta> m_devMomentaF;
@@ -187,12 +194,12 @@ namespace mg5amcCpu
   // Forward declare transposition methods
   //
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
   template<typename Tin, typename Tout>
   __global__ void dev_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt );
 
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
 
   template<typename Tin, typename Tout>
   void hst_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt );
@@ -209,7 +216,7 @@ namespace mg5amcCpu
   Bridge<FORTRANFPTYPE>::Bridge( unsigned int nevtF, unsigned int nparF, unsigned int np4F )
     : m_nevt( nevtF )
     , m_nGoodHel( -1 )
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     , m_gputhreads( 256 )                  // default number of gpu threads
     , m_gpublocks( m_nevt / m_gputhreads ) // this ensures m_nevt <= m_gpublocks*m_gputhreads
     , m_devMomentaF( m_nevt )
@@ -233,7 +240,7 @@ namespace mg5amcCpu
   {
     if( nparF != CPPProcess::npar ) throw std::runtime_error( "Bridge constructor: npar mismatch" );
     if( np4F != CPPProcess::np4 ) throw std::runtime_error( "Bridge constructor: np4 mismatch" );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( ( m_nevt < s_gputhreadsmin ) || ( m_nevt % s_gputhreadsmin != 0 ) )
       throw std::runtime_error( "Bridge constructor: nevt should be a multiple of " + std::to_string( s_gputhreadsmin ) );
     while( m_nevt != m_gpublocks * m_gputhreads )
@@ -249,20 +256,28 @@ namespace mg5amcCpu
 #else
     std::cout << "WARNING! Instantiate host Bridge (nevt=" << m_nevt << ")" << std::endl;
     m_pmek.reset( new MatrixElementKernelHost( m_hstMomentaC, m_hstGs, m_hstRndHel, m_hstRndCol, m_hstMEs, m_hstSelHel, m_hstSelCol, m_nevt ) );
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
     // Create a process object, read param card and set parameters
     // FIXME: the process instance can happily go out of scope because it is only needed to read parameters?
     // FIXME: the CPPProcess should really be a singleton? what if fbridgecreate is called from several Fortran threads?
     CPPProcess process( /*verbose=*/false );
     std::string paramCard = "../../Cards/param_card.dat";
-    if( !std::filesystem::exists( paramCard ) )
-    {
-      paramCard = "../" + paramCard;
-    }
+    /*
+#ifdef __HIPCC__
+    if( !std::experimental::filesystem::exists( paramCard ) ) paramCard = "../" + paramCard;
+#else
+    if( !std::filesystem::exists( paramCard ) ) paramCard = "../" + paramCard;
+#endif
+    */
+    //struct stat dummybuffer; // bypass std::filesystem #803
+    //if( !( stat( paramCard.c_str(), &dummyBuffer ) == 0 ) ) paramCard = "../" + paramCard; //
+    auto fileExists = []( std::string& fileName )
+    { struct stat buffer; return stat( fileName.c_str(), &buffer ) == 0; };
+    if( !fileExists( paramCard ) ) paramCard = "../" + paramCard; // bypass std::filesystem #803
     process.initProc( paramCard );
   }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::set_gpugrid( const int gpublocks, const int gputhreads )
   {
@@ -276,7 +291,7 @@ namespace mg5amcCpu
   }
 #endif
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::gpu_sequence( const FORTRANFPTYPE* momenta,
                                             const FORTRANFPTYPE* gs,
@@ -291,14 +306,14 @@ namespace mg5amcCpu
     constexpr int neppM = MemoryAccessMomenta::neppM;
     if constexpr( neppM == 1 && std::is_same_v<FORTRANFPTYPE, fptype> )
     {
-      checkCuda( cudaMemcpy( m_devMomentaC.data(), momenta, m_devMomentaC.bytes(), cudaMemcpyHostToDevice ) );
+      gpuMemcpy( m_devMomentaC.data(), momenta, m_devMomentaC.bytes(), gpuMemcpyHostToDevice );
     }
     else
     {
-      checkCuda( cudaMemcpy( m_devMomentaF.data(), momenta, m_devMomentaF.bytes(), cudaMemcpyHostToDevice ) );
+      gpuMemcpy( m_devMomentaF.data(), momenta, m_devMomentaF.bytes(), gpuMemcpyHostToDevice );
       const int thrPerEvt = CPPProcess::npar * CPPProcess::np4; // AV: transpose alg does 1 element per thread (NOT 1 event per thread)
       //const int thrPerEvt = 1; // AV: try new alg with 1 event per thread... this seems slower
-      dev_transposeMomentaF2C<<<m_gpublocks * thrPerEvt, m_gputhreads>>>( m_devMomentaF.data(), m_devMomentaC.data(), m_nevt );
+      gpuLaunchKernel( dev_transposeMomentaF2C, m_gpublocks * thrPerEvt, m_gputhreads, m_devMomentaF.data(), m_devMomentaC.data(), m_nevt );
     }
     if constexpr( std::is_same_v<FORTRANFPTYPE, fptype> )
     {
@@ -341,7 +356,7 @@ namespace mg5amcCpu
   }
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::cpu_sequence( const FORTRANFPTYPE* momenta,
                                             const FORTRANFPTYPE* gs,
@@ -396,7 +411,7 @@ namespace mg5amcCpu
   // - C++ array: momenta[npagM][npar][np4][neppM] with nevt=npagM*neppM (AOSOA)
   //
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename Tin, typename Tout>
   __global__ void dev_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt )
   {
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/BridgeKernels.cc b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/BridgeKernels.cc
index d58066c9c1..eaf4037a24 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/BridgeKernels.cc
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/BridgeKernels.cc
@@ -1,17 +1,18 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "BridgeKernels.h"
 
+#include "GpuAbstraction.h"
 #include "MemoryAccessMomenta.h"
 
 #include <sstream>
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -45,7 +46,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu
 {
 
@@ -96,7 +97,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/BridgeKernels.h b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/BridgeKernels.h
index 15eb4bff4d..3efef8ce97 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/BridgeKernels.h
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/BridgeKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef BRIDGEKERNELS_H
 #define BRIDGEKERNELS_H 1
@@ -12,7 +12,7 @@
 #include "MatrixElementKernels.h"
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -49,7 +49,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A Bridge wrapper class encapsulating matrix element calculations on a CPU host
   class BridgeKernelHost final : public BridgeKernelBase
   {
@@ -89,7 +89,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A Bridge wrapper class encapsulating matrix element calculations on a GPU device
   class BridgeKernelDevice : public BridgeKernelBase
   {
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/CommonRandomNumberKernel.cc b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/CommonRandomNumberKernel.cc
index 985b39f576..010bc4cbd0 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/CommonRandomNumberKernel.cc
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/CommonRandomNumberKernel.cc
@@ -1,15 +1,16 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "CommonRandomNumbers.h"
+#include "GpuAbstraction.h"
 #include "MemoryBuffers.h"
 #include "RandomNumberKernels.h"
 
 #include <cassert>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/CrossSectionKernels.cc b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/CrossSectionKernels.cc
index 0b355a3c8d..c15b39844d 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/CrossSectionKernels.cc
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/CrossSectionKernels.cc
@@ -1,10 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "CrossSectionKernels.h"
 
+#include "GpuAbstraction.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessWeights.h"
 #include "MemoryBuffers.h"
@@ -77,7 +78,7 @@ debug_me_is_abnormal( const fptype& me, size_t ievtALL )
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -185,7 +186,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/CrossSectionKernels.h b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/CrossSectionKernels.h
index 7933ca4bbf..4d9659e04e 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/CrossSectionKernels.h
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/CrossSectionKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef CROSSSECTIONKERNELS_H
 #define CROSSSECTIONKERNELS_H 1
@@ -13,7 +13,7 @@
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -96,7 +96,7 @@ namespace mg5amcCpu
   //--------------------------------------------------------------------------
 
   /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating the calculation of event statistics on a GPU device
   class CrossSectionKernelDevice : public CrossSectionKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/CudaRuntime.h b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/CudaRuntime.h
deleted file mode 100644
index 64ce52f4b3..0000000000
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/CudaRuntime.h
+++ /dev/null
@@ -1,85 +0,0 @@
-// Copyright (C) 2020-2023 CERN and UCLouvain.
-// Licensed under the GNU Lesser General Public License (version 3 or later).
-// Created by: S. Roiser (Jul 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
-
-#ifndef MG5AMC_CUDARUNTIME_H
-#define MG5AMC_CUDARUNTIME_H 1
-
-// MG5AMC on GPU uses the CUDA runtime API, not the lower level CUDA driver API
-// See https://docs.nvidia.com/cuda/cuda-runtime-api/driver-vs-runtime-api.html#driver-vs-runtime-api
-
-#include <cassert>
-#include <iostream>
-
-//--------------------------------------------------------------------------
-
-// See https://stackoverflow.com/a/14038590
-#ifdef __CUDACC__ /* clang-format off */
-#define checkCuda( code ) { assertCuda( code, __FILE__, __LINE__ ); }
-inline void assertCuda( cudaError_t code, const char* file, int line, bool abort = true )
-{
-  if( code != cudaSuccess )
-  {
-    printf( "ERROR! assertCuda: '%s' (%d) in %s:%d\n", cudaGetErrorString( code ), code, file, line );
-    if( abort ) assert( code == cudaSuccess );
-  }
-}
-#endif /* clang-format on */
-
-//--------------------------------------------------------------------------
-
-#ifdef __CUDACC__
-namespace mg5amcGpu
-{
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  // *** FIXME! This will all need to be designed differently when going to multi-GPU nodes! ***
-  struct CudaRuntime final
-  {
-    CudaRuntime( const bool debug = true )
-      : m_debug( debug ) { setUp( m_debug ); }
-    ~CudaRuntime() { tearDown( m_debug ); }
-    CudaRuntime( const CudaRuntime& ) = delete;
-    CudaRuntime( CudaRuntime&& ) = delete;
-    CudaRuntime& operator=( const CudaRuntime& ) = delete;
-    CudaRuntime& operator=( CudaRuntime&& ) = delete;
-    bool m_debug;
-
-    // Set up CUDA application
-    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
-    // Calling cudaSetDevice on startup is useful to properly book-keep the time spent in CUDA initialization
-    static void setUp( const bool debug = true )
-    {
-      // ** NB: it is useful to call cudaSetDevice, or cudaFree, to properly book-keep the time spent in CUDA initialization
-      // ** NB: otherwise, the first CUDA operation (eg a cudaMemcpyToSymbol in CPPProcess ctor) appears to take much longer!
-      /*
-      // [We initially added cudaFree(0) to "ease profile analysis" only because it shows up as a big recognizable block!]
-      // No explicit initialization is needed: https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#initialization
-      // It is not clear what cudaFree(0) does at all: https://stackoverflow.com/questions/69967813/
-      if ( debug ) std::cout << "__CudaRuntime: calling cudaFree(0)" << std::endl;
-      checkCuda( cudaFree( 0 ) ); // SLOW!
-      */
-      // Replace cudaFree(0) by cudaSetDevice(0), even if it is not really needed either
-      // (but see https://developer.nvidia.com/blog/cuda-pro-tip-always-set-current-device-avoid-multithreading-bugs)
-      if( debug ) std::cout << "__CudaRuntime: calling cudaSetDevice(0)" << std::endl;
-      checkCuda( cudaSetDevice( 0 ) ); // SLOW!
-    }
-
-    // Tear down CUDA application (call cudaDeviceReset)
-    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
-    // Calling cudaDeviceReset on shutdown is only needed for checking memory leaks in cuda-memcheck
-    // See https://docs.nvidia.com/cuda/cuda-memcheck/index.html#leak-checking
-    static void tearDown( const bool debug = true )
-    {
-      if( debug ) std::cout << "__CudaRuntime: calling cudaDeviceReset()" << std::endl;
-      checkCuda( cudaDeviceReset() );
-    }
-  };
-
-}
-#endif
-
-//--------------------------------------------------------------------------
-
-#endif // MG5AMC_CUDARUNTIME_H
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/CurandRandomNumberKernel.cc b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/CurandRandomNumberKernel.cc
index eb56333b03..08a16f6f2c 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/CurandRandomNumberKernel.cc
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/CurandRandomNumberKernel.cc
@@ -1,9 +1,9 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "MemoryBuffers.h"
 #include "RandomNumberKernels.h"
 
@@ -22,7 +22,7 @@ inline void assertCurand( curandStatus_t code, const char *file, int line, bool
 }
 #endif /* clang-format on */
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -36,7 +36,7 @@ namespace mg5amcCpu
   {
     if( m_isOnDevice )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if( !m_rnarray.isOnDevice() )
         throw std::runtime_error( "CurandRandomNumberKernel on device with a host random number array" );
 #else
@@ -114,7 +114,7 @@ namespace mg5amcCpu
     /*
     printf( "\nCurandRandomNumberKernel::generateRnarray size = %d\n", (int)m_rnarray.size() );
     fptype* data = m_rnarray.data();
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( m_rnarray.isOnDevice() )
     {
       data = new fptype[m_rnarray.size()]();
@@ -123,7 +123,7 @@ namespace mg5amcCpu
 #endif
     for( int i = 0; i < ( (int)m_rnarray.size() / 4 ); i++ )
       printf( "[%4d] %f %f %f %f\n", i * 4, data[i * 4], data[i * 4 + 2], data[i * 4 + 2], data[i * 4 + 3] );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( m_rnarray.isOnDevice() ) delete[] data;
 #endif
     */
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/EventStatistics.h b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/EventStatistics.h
index 48b51e0a49..b425a5bade 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/EventStatistics.h
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/EventStatistics.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef EventStatistics_H
 #define EventStatistics_H 1
@@ -16,7 +16,7 @@
 #include <limits>
 #include <string>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/GpuAbstraction.h b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/GpuAbstraction.h
new file mode 100644
index 0000000000..9c467b1e04
--- /dev/null
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/GpuAbstraction.h
@@ -0,0 +1,69 @@
+// Copyright (C) 2020-2023 CERN and UCLouvain.
+// Licensed under the GNU Lesser General Public License (version 3 or later).
+// Created by: J. Teig (Jul 2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+
+#ifndef MG5AMC_GPUABSTRACTION_H
+#define MG5AMC_GPUABSTRACTION_H 1
+
+#include <cassert>
+
+//--------------------------------------------------------------------------
+
+#ifdef __CUDACC__
+
+#define gpuError_t cudaError_t
+#define gpuPeekAtLastError cudaPeekAtLastError
+#define gpuGetErrorString cudaGetErrorString
+#define gpuSuccess cudaSuccess
+
+#define gpuMallocHost( ptr, size ) checkGpu( cudaMallocHost( ptr, size ) )
+#define gpuMalloc( ptr, size ) checkGpu( cudaMalloc( ptr, size ) )
+
+#define gpuMemcpy( dstData, srcData, srcBytes, func ) checkGpu( cudaMemcpy( dstData, srcData, srcBytes, func ) )
+#define gpuMemcpyHostToDevice cudaMemcpyHostToDevice
+#define gpuMemcpyDeviceToHost cudaMemcpyDeviceToHost
+#define gpuMemcpyToSymbol( type1, type2, size ) checkGpu( cudaMemcpyToSymbol( type1, type2, size ) )
+
+#define gpuFree( ptr ) checkGpu( cudaFree( ptr ) )
+#define gpuFreeHost( ptr ) checkGpu( cudaFreeHost( ptr ) )
+
+#define gpuSetDevice cudaSetDevice
+#define gpuDeviceSynchronize cudaDeviceSynchronize
+#define gpuDeviceReset cudaDeviceReset
+
+#define gpuLaunchKernel( kernel, blocks, threads, ... ) kernel<<<blocks, threads>>>( __VA_ARGS__ )
+#define gpuLaunchKernelSharedMem( kernel, blocks, threads, sharedMem, ... ) kernel<<<blocks, threads, sharedMem>>>( __VA_ARGS__ )
+
+//--------------------------------------------------------------------------
+
+#elif defined __HIPCC__
+
+#define gpuError_t hipError_t
+#define gpuPeekAtLastError hipPeekAtLastError
+#define gpuGetErrorString hipGetErrorString
+#define gpuSuccess hipSuccess
+
+#define gpuMallocHost( ptr, size ) checkGpu( hipHostMalloc( ptr, size ) ) // HostMalloc better
+#define gpuMalloc( ptr, size ) checkGpu( hipMalloc( ptr, size ) )
+
+#define gpuMemcpy( dstData, srcData, srcBytes, func ) checkGpu( hipMemcpy( dstData, srcData, srcBytes, func ) )
+#define gpuMemcpyHostToDevice hipMemcpyHostToDevice
+#define gpuMemcpyDeviceToHost hipMemcpyDeviceToHost
+#define gpuMemcpyToSymbol( type1, type2, size ) checkGpu( hipMemcpyToSymbol( type1, type2, size ) )
+
+#define gpuFree( ptr ) checkGpu( hipFree( ptr ) )
+#define gpuFreeHost( ptr ) checkGpu( hipHostFree( ptr ) )
+
+#define gpuSetDevice hipSetDevice
+#define gpuDeviceSynchronize hipDeviceSynchronize
+#define gpuDeviceReset hipDeviceReset
+
+#define gpuLaunchKernel( kernel, blocks, threads, ... ) kernel<<<blocks, threads>>>( __VA_ARGS__ )
+#define gpuLaunchKernelSharedMem( kernel, blocks, threads, sharedMem, ... ) kernel<<<blocks, threads, sharedMem>>>( __VA_ARGS__ )
+
+//--------------------------------------------------------------------------
+
+#endif
+
+#endif // MG5AMC_GPUABSTRACTION_H
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/GpuRuntime.h b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/GpuRuntime.h
new file mode 100644
index 0000000000..93579ef08b
--- /dev/null
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/GpuRuntime.h
@@ -0,0 +1,85 @@
+// Copyright (C) 2020-2023 CERN and UCLouvain.
+// Licensed under the GNU Lesser General Public License (version 3 or later).
+// Created by: J. Teig (Jun 2023, based on earlier work by S. Roiser) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+
+#ifndef MG5AMC_GPURUNTIME_H
+#define MG5AMC_GPURUNTIME_H 1
+
+// MG5AMC on GPU uses the CUDA runtime API, not the lower level CUDA driver API
+// See https://docs.nvidia.com/cuda/cuda-runtime-api/driver-vs-runtime-api.html#driver-vs-runtime-api
+
+#include "GpuAbstraction.h"
+
+#include <iostream>
+
+//--------------------------------------------------------------------------
+
+// See https://stackoverflow.com/a/14038590
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
+#define checkGpu( code ) { assertGpu( code, __FILE__, __LINE__ ); }
+inline void assertGpu( gpuError_t code, const char* file, int line, bool abort = true )
+{
+  if( code != gpuSuccess )
+  {
+    printf( "ERROR! assertGpu: '%s' (%d) in %s:%d\n", gpuGetErrorString( code ), code, file, line );
+    if( abort ) assert( code == gpuSuccess );
+  }
+}
+#endif /* clang-format on */
+
+//--------------------------------------------------------------------------
+
+#ifdef MGONGPUCPP_GPUIMPL
+namespace mg5amcGpu
+{
+  // Instantiate a GpuRuntime at the beginnining of the application's main to
+  // invoke gpuSetDevice(0) in the constructor and book a gpuDeviceReset() call in the destructor
+  // *** FIXME! This will all need to be designed differently when going to multi-GPU nodes! ***
+  struct GpuRuntime final
+  {
+    GpuRuntime( const bool debug = true )
+      : m_debug( debug ) { setUp( m_debug ); }
+    ~GpuRuntime() { tearDown( m_debug ); }
+    GpuRuntime( const GpuRuntime& ) = delete;
+    GpuRuntime( GpuRuntime&& ) = delete;
+    GpuRuntime& operator=( const GpuRuntime& ) = delete;
+    GpuRuntime& operator=( GpuRuntime&& ) = delete;
+    bool m_debug;
+
+    // Set up CUDA application
+    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
+    // Calling cudaSetDevice on startup is useful to properly book-keep the time spent in CUDA initialization
+    static void setUp( const bool debug = true )
+    {
+      // ** NB: it is useful to call cudaSetDevice, or cudaFree, to properly book-keep the time spent in CUDA initialization
+      // ** NB: otherwise, the first CUDA operation (eg a cudaMemcpyToSymbol in CPPProcess ctor) appears to take much longer!
+      /*
+      // [We initially added cudaFree(0) to "ease profile analysis" only because it shows up as a big recognizable block!]
+      // No explicit initialization is needed: https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#initialization
+      // It is not clear what cudaFree(0) does at all: https://stackoverflow.com/questions/69967813/
+      if ( debug ) std::cout << "__CudaRuntime: calling cudaFree(0)" << std::endl;
+      checkCuda( cudaFree( 0 ) ); // SLOW!
+      */
+      // Replace cudaFree(0) by cudaSetDevice(0), even if it is not really needed either
+      // (but see https://developer.nvidia.com/blog/cuda-pro-tip-always-set-current-device-avoid-multithreading-bugs)
+      if( debug ) std::cout << "__GpuRuntime: calling GpuSetDevice(0)" << std::endl;
+      checkGpu( gpuSetDevice( 0 ) ); // SLOW!
+    }
+
+    // Tear down CUDA application (call cudaDeviceReset)
+    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
+    // Calling cudaDeviceReset on shutdown is only needed for checking memory leaks in cuda-memcheck
+    // See https://docs.nvidia.com/cuda/cuda-memcheck/index.html#leak-checking
+    static void tearDown( const bool debug = true )
+    {
+      if( debug ) std::cout << "__GpuRuntime: calling GpuDeviceReset()" << std::endl;
+      checkGpu( gpuDeviceReset() );
+    }
+  };
+}
+#endif
+
+//--------------------------------------------------------------------------
+
+#endif // MG5AMC_GPURUNTIME_H
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MadgraphTest.h b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MadgraphTest.h
index ef40624c88..6054185300 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MadgraphTest.h
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MadgraphTest.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Hageboeck (Dec 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MADGRAPHTEST_H_
 #define MADGRAPHTEST_H_ 1
@@ -14,7 +14,11 @@
 
 #include <array>
 #include <cmath>
-#include <filesystem>
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
 #include <fstream>
 #include <iomanip>
 #include <memory>
@@ -22,7 +26,7 @@
 #include <string>
 #include <vector>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using mg5amcGpu::CPPProcess;
 #else
 using mg5amcCpu::CPPProcess;
@@ -201,7 +205,7 @@ class MadgraphTest : public testing::TestWithParam<TestDriverBase*>
 
 // Since we link both the CPU-only and GPU tests into the same executable, we prevent
 // a multiply defined symbol by only compiling this in the non-CUDA phase:
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
 /// Compare momenta and matrix elements.
 /// This uses an implementation of TestDriverBase to run a madgraph workflow,
@@ -219,7 +223,14 @@ TEST_P( MadgraphTest, CompareMomentaAndME )
   const char* dumpEventsC = getenv( "CUDACPP_RUNTEST_DUMPEVENTS" );
   const bool dumpEvents = ( dumpEventsC != 0 ) && ( std::string( dumpEventsC ) != "" );
   const std::string refFileName = testDriver->getRefFileName();
+  /*
+#ifdef __HIPCC__
+  const std::string dumpFileName = std::experimental::filesystem::path( refFileName ).filename();
+#else
   const std::string dumpFileName = std::filesystem::path( refFileName ).filename();
+#endif
+  */
+  const std::string dumpFileName = refFileName; // bypass std::filesystem #803
   std::ofstream dumpFile;
   if( dumpEvents )
   {
@@ -307,6 +318,6 @@ TEST_P( MadgraphTest, CompareMomentaAndME )
   }
 }
 
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
 
 #endif /* MADGRAPHTEST_H_ */
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MatrixElementKernels.cc b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MatrixElementKernels.cc
index 74b5239ebf..81699dfea9 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MatrixElementKernels.cc
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MatrixElementKernels.cc
@@ -1,12 +1,12 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "MatrixElementKernels.h"
 
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h" // Includes the abstraction for Nvidia/AMD compilation
 #include "MemoryAccessMomenta.h"
 #include "MemoryBuffers.h"
 
@@ -14,7 +14,7 @@
 
 //============================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu
 {
 
@@ -150,7 +150,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
@@ -209,13 +209,13 @@ namespace mg5amcGpu
     PinnedHostBufferHelicityMask hstIsGoodHel( ncomb );
     DeviceBufferHelicityMask devIsGoodHel( ncomb );
     // ... 0d1. Compute good helicity mask on the device
-    computeDependentCouplings<<<m_gpublocks, m_gputhreads>>>( m_gs.data(), m_couplings.data() );
+    gpuLaunchKernel( computeDependentCouplings, m_gpublocks, m_gputhreads, m_gs.data(), m_couplings.data() );
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-    sigmaKin_getGoodHel<<<m_gpublocks, m_gputhreads>>>( m_momenta.data(), m_couplings.data(), m_matrixElements.data(), m_numerators.data(), m_denominators.data(), devIsGoodHel.data() );
+    gpuLaunchKernel( sigmaKin_getGoodHel, m_gpublocks, m_gputhreads, m_momenta.data(), m_couplings.data(), m_matrixElements.data(), m_numerators.data(), m_denominators.data(), devIsGoodHel.data() );
 #else
-    sigmaKin_getGoodHel<<<m_gpublocks, m_gputhreads>>>( m_momenta.data(), m_couplings.data(), m_matrixElements.data(), devIsGoodHel.data() );
+    gpuLaunchKernel( sigmaKin_getGoodHel, m_gpublocks, m_gputhreads, m_momenta.data(), m_couplings.data(), m_matrixElements.data(), devIsGoodHel.data() );
 #endif
-    checkCuda( cudaPeekAtLastError() );
+    checkGpu( gpuPeekAtLastError() );
     // ... 0d2. Copy back good helicity mask to the host
     copyHostFromDevice( hstIsGoodHel, devIsGoodHel );
     // ... 0d3. Copy back good helicity list to constant memory on the device
@@ -226,19 +226,19 @@ namespace mg5amcGpu
 
   void MatrixElementKernelDevice::computeMatrixElements( const unsigned int channelId )
   {
-    computeDependentCouplings<<<m_gpublocks, m_gputhreads>>>( m_gs.data(), m_couplings.data() );
+    gpuLaunchKernel( computeDependentCouplings, m_gpublocks, m_gputhreads, m_gs.data(), m_couplings.data() );
 #ifndef MGONGPU_NSIGHT_DEBUG
     constexpr unsigned int sharedMemSize = 0;
 #else
     constexpr unsigned int sharedMemSize = ntpbMAX * sizeof( float );
 #endif
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-    sigmaKin<<<m_gpublocks, m_gputhreads, sharedMemSize>>>( m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), channelId, m_numerators.data(), m_denominators.data(), m_selhel.data(), m_selcol.data() );
+    gpuLaunchKernelSharedMem( sigmaKin, m_gpublocks, m_gputhreads, sharedMemSize, m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), channelId, m_numerators.data(), m_denominators.data(), m_selhel.data(), m_selcol.data() );
 #else
-    sigmaKin<<<m_gpublocks, m_gputhreads, sharedMemSize>>>( m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), m_selhel.data(), m_selcol.data() );
+    gpuLaunchKernelSharedMem( sigmaKin, m_gpublocks, m_gputhreads, sharedMemSize, m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), m_selhel.data(), m_selcol.data() );
 #endif
-    checkCuda( cudaPeekAtLastError() );
-    checkCuda( cudaDeviceSynchronize() );
+    checkGpu( gpuPeekAtLastError() );
+    checkGpu( gpuDeviceSynchronize() );
   }
 
   //--------------------------------------------------------------------------
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MatrixElementKernels.h b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MatrixElementKernels.h
index 23e84757a2..72bd8f195b 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MatrixElementKernels.h
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MatrixElementKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MATRIXELEMENTKERNELS_H
 #define MATRIXELEMENTKERNELS_H 1
@@ -10,7 +10,7 @@
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -81,7 +81,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating matrix element calculations on a CPU host
   class MatrixElementKernelHost final : public MatrixElementKernelBase, public NumberOfEvents
   {
@@ -130,7 +130,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating matrix element calculations on a GPU device
   class MatrixElementKernelDevice : public MatrixElementKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessAmplitudes.h b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessAmplitudes.h
index 573b3bbbc9..ffb76e93de 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessAmplitudes.h
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessAmplitudes.h
@@ -15,7 +15,7 @@
 #define MGONGPU_TRIVIAL_AMPLITUDES 1
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessCouplings.h b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessCouplings.h
index 35a3af42e0..3afdf3e554 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessCouplings.h
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessCouplings.h
@@ -15,7 +15,7 @@
 #include "MemoryBuffers.h"       // for HostBufferCouplings::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessCouplingsFixed.h b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessCouplingsFixed.h
index dc0d93afff..ffcdf4dbef 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessCouplingsFixed.h
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessCouplingsFixed.h
@@ -14,7 +14,7 @@
 //#include "MemoryAccessHelpers.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessDenominators.h b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessDenominators.h
index 3bce635718..66f2d32a6b 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessDenominators.h
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessDenominators.h
@@ -10,7 +10,7 @@
 #include "MemoryAccessGs.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessGs.h b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessGs.h
index 31311aa375..4c726b30f3 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessGs.h
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessGs.h
@@ -13,7 +13,7 @@
 #include "MemoryBuffers.h" // for HostBufferMatrixElements::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessHelpers.h b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessHelpers.h
index c82a6c7635..db73e4e064 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessHelpers.h
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessHelpers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessHelpers_H
 #define MemoryAccessHelpers_H 1
@@ -105,7 +105,7 @@ class KernelAccessHelper : public MemoryAccessHelper<T>
     }
     else
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
       //printf( "kernelAccessRecord: ievt=%d threadId=%d\n", ievt, threadIdx.x );
       return T::ieventAccessRecord( buffer, ievt ); // NB fptype and fptype_sv coincide for CUDA
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessMatrixElements.h b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessMatrixElements.h
index f32e6fea5b..3741011971 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessMatrixElements.h
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessMatrixElements.h
@@ -13,7 +13,7 @@
 #include "MemoryBuffers.h" // for HostBufferMatrixElements::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessMomenta.h b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessMomenta.h
index 29266de32c..3be229d392 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessMomenta.h
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessMomenta.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessMomenta_H
 #define MemoryAccessMomenta_H 1
@@ -13,7 +13,7 @@
 #include "MemoryAccessVectors.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -30,7 +30,7 @@ namespace mg5amcCpu
 
     // Number of Events Per Page in the momenta AOSOA memory buffer layout
     // (these are all best kept as a compile-time constants: see issue #23)
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
     // -----------------------------------------------------------------------------------------------
     // --- GPUs: neppM is best set to a power of 2 times the number of fptype's in a 32-byte cacheline
     // --- This is relevant to ensure coalesced access to momenta in global memory
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessNumerators.h b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessNumerators.h
index b152183b28..18991f4fa6 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessNumerators.h
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessNumerators.h
@@ -10,7 +10,7 @@
 #include "MemoryAccessGs.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessRandomNumbers.h b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessRandomNumbers.h
index e2988d39f3..40cb089135 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessRandomNumbers.h
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessRandomNumbers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessRandomNumbers_H
 #define MemoryAccessRandomNumbers_H 1
@@ -11,7 +11,7 @@
 #include "CPPProcess.h"
 #include "MemoryAccessHelpers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using mg5amcGpu::CPPProcess;
 #else
 using mg5amcCpu::CPPProcess;
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessVectors.h b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessVectors.h
index e9b197368e..08faccff0f 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessVectors.h
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessVectors.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessVectors_H
 #define MemoryAccessVectors_H 1
@@ -10,7 +10,7 @@
 
 #include "mgOnGpuVectors.h"
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu // this is only needed for CPU SIMD vectorization
 {
 
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessWavefunctions.h b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessWavefunctions.h
index 5428aaf933..33bef4559e 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessWavefunctions.h
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryAccessWavefunctions.h
@@ -15,7 +15,7 @@
 #define MGONGPU_TRIVIAL_WAVEFUNCTIONS 1
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryBuffers.h b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryBuffers.h
index 3093e6ed18..7756a71621 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryBuffers.h
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/MemoryBuffers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021, based on earlier work by S. Hageboeck) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryBuffers_H
 #define MemoryBuffers_H 1
@@ -11,12 +11,12 @@
 #include "mgOnGpuCxtypes.h"
 
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "Parameters_sm.h"
 
 #include <sstream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -87,7 +87,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   constexpr bool HostBufferALIGNED = false;   // ismisaligned=false
   constexpr bool HostBufferMISALIGNED = true; // ismisaligned=true
 
@@ -119,7 +119,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA pinned host buffer
   template<typename T>
   class PinnedHostBufferBase : public BufferBase<T>
@@ -128,18 +128,18 @@ namespace mg5amcCpu
     PinnedHostBufferBase( const size_t size )
       : BufferBase<T>( size, false )
     {
-      checkCuda( cudaMallocHost( &( this->m_data ), this->bytes() ) );
+      gpuMallocHost( &( this->m_data ), this->bytes() );
     }
     virtual ~PinnedHostBufferBase()
     {
-      checkCuda( cudaFreeHost( this->m_data ) );
+      gpuFreeHost( this->m_data );
     }
   };
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA device buffer
   template<typename T>
   class DeviceBufferBase : public BufferBase<T>
@@ -148,18 +148,18 @@ namespace mg5amcCpu
     DeviceBufferBase( const size_t size )
       : BufferBase<T>( size, true )
     {
-      checkCuda( cudaMalloc( &( this->m_data ), this->bytes() ) );
+      gpuMalloc( &( this->m_data ), this->bytes() );
     }
     virtual ~DeviceBufferBase()
     {
-      checkCuda( cudaFree( this->m_data ) );
+      gpuFree( this->m_data );
     }
   };
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for a given number of events
   template<typename T, size_t sizePerEvent, bool ismisaligned>
   class HostBuffer : public HostBufferBase<T, ismisaligned>, virtual private NumberOfEvents
@@ -175,7 +175,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA pinned host buffer for a given number of events
   template<typename T, size_t sizePerEvent>
   class PinnedHostBuffer : public PinnedHostBufferBase<T>, virtual private NumberOfEvents
@@ -191,7 +191,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA device buffer for a given number of events
   template<typename T, size_t sizePerEvent>
   class DeviceBuffer : public DeviceBufferBase<T>, virtual private NumberOfEvents
@@ -213,7 +213,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for momenta random numbers
   constexpr size_t sizePerEventRndNumMomenta = MemoryBuffers::np4 * MemoryBuffers::nparf;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for momenta random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumMomenta, HostBufferALIGNED> HostBufferRndNumMomenta;
 #else
@@ -232,7 +232,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer with ONE fptype per event
   constexpr size_t sizePerEventOneFp = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer with ONE fptype per event
   typedef HostBuffer<fptype, sizePerEventOneFp, HostBufferALIGNED> HostBufferOneFp;
 #else
@@ -257,7 +257,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for Gs
   constexpr size_t sizePerEventGs = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventGs, HostBufferALIGNED> HostBufferGs;
 #else
@@ -276,7 +276,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for numerators
   constexpr size_t sizePerEventNumerators = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventNumerators, HostBufferALIGNED> HostBufferNumerators;
 #else
@@ -296,7 +296,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for denominators
   constexpr size_t sizePerEventDenominators = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventDenominators, HostBufferALIGNED> HostBufferDenominators;
 #else
@@ -315,7 +315,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for random numbers
   constexpr size_t sizePerEventCouplings = MemoryBuffers::ndcoup * MemoryBuffers::nx2;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventCouplings, HostBufferALIGNED> HostBufferCouplings;
 #else
@@ -333,7 +333,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for momenta
   constexpr size_t sizePerEventMomenta = MemoryBuffers::np4 * MemoryBuffers::npar;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for momenta
   typedef HostBuffer<fptype, sizePerEventMomenta, HostBufferALIGNED> HostBufferMomenta;
   //typedef HostBuffer<fptype, sizePerEventMomenta, HostBufferMISALIGNED> HostBufferMomenta; // TEST MISALIGNMENT!
@@ -352,7 +352,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for sampling weights
   constexpr size_t sizePerEventWeights = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for sampling weights
   typedef HostBuffer<fptype, sizePerEventWeights, HostBufferALIGNED> HostBufferWeights;
 #else
@@ -370,7 +370,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for matrix elements
   constexpr size_t sizePerEventMatrixElements = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for matrix elements
   typedef HostBuffer<fptype, sizePerEventMatrixElements, HostBufferALIGNED> HostBufferMatrixElements;
 #else
@@ -385,7 +385,7 @@ namespace mg5amcCpu
   // A base class encapsulating a memory buffer for the helicity mask
   typedef BufferBase<bool> BufferHelicityMask;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for the helicity mask
   typedef HostBufferBase<bool, HostBufferALIGNED> HostBufferHelicityMask;
 #else
@@ -403,7 +403,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for wavefunctions
   constexpr size_t sizePerEventWavefunctions = MemoryBuffers::nw6 * MemoryBuffers::nx2;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for wavefunctions
   typedef HostBuffer<fptype, sizePerEventWavefunctions, HostBufferALIGNED> HostBufferWavefunctions;
 #else
@@ -421,7 +421,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for helicity random numbers
   constexpr size_t sizePerEventRndNumHelicity = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for helicity random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumHelicity, HostBufferALIGNED> HostBufferRndNumHelicity;
 #else
@@ -439,7 +439,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for color random numbers
   constexpr size_t sizePerEventRndNumColor = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for color random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumColor, HostBufferALIGNED> HostBufferRndNumColor;
 #else
@@ -457,7 +457,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for helicity selection
   constexpr size_t sizePerEventSelectedHelicity = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for helicity selection
   typedef HostBuffer<int, sizePerEventSelectedHelicity, HostBufferALIGNED> HostBufferSelectedHelicity;
 #else
@@ -475,7 +475,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for color selection
   constexpr size_t sizePerEventSelectedColor = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for color selection
   typedef HostBuffer<int, sizePerEventSelectedColor, HostBufferALIGNED> HostBufferSelectedColor;
 #else
@@ -487,7 +487,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<class Tdst, class Tsrc>
   void copyDeviceFromHost( Tdst& dst, const Tsrc& src ) // keep the same order of arguments as in memcpy
   {
@@ -504,13 +504,13 @@ namespace mg5amcCpu
       throw std::runtime_error( sstr.str() );
     }
     // NB (PR #45): cudaMemcpy involves an intermediate memcpy to pinned memory if host array is a not a pinned host array
-    checkCuda( cudaMemcpy( dst.data(), src.data(), src.bytes(), cudaMemcpyHostToDevice ) );
+    gpuMemcpy( dst.data(), src.data(), src.bytes(), gpuMemcpyHostToDevice );
   }
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<class Tdst, class Tsrc>
   void copyHostFromDevice( Tdst& dst, const Tsrc& src ) // keep the same order of arguments as in memcpy
   {
@@ -527,7 +527,7 @@ namespace mg5amcCpu
       throw std::runtime_error( sstr.str() );
     }
     // NB (PR #45): cudaMemcpy involves an intermediate memcpy to pinned memory if host array is a not a pinned host array
-    checkCuda( cudaMemcpy( dst.data(), src.data(), src.bytes(), cudaMemcpyDeviceToHost ) );
+    gpuMemcpy( dst.data(), src.data(), src.bytes(), gpuMemcpyDeviceToHost );
   }
 #endif
 
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P1_gg_ttx/CPPProcess.cc b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P1_gg_ttx/CPPProcess.cc
index 18052b6676..dbaa56b35c 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P1_gg_ttx/CPPProcess.cc
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P1_gg_ttx/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_sm.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -46,7 +45,7 @@
 // Class member functions for calculating the matrix elements for
 // Process: g g > t t~ WEIGHTED<=2 @1
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -80,7 +79,7 @@ namespace mg5amcCpu
   __device__ const fptype cIPD[2] = { (fptype)Parameters_sm::mdl_MT, (fptype)Parameters_sm::mdl_WT };
   __device__ const fptype* cIPC = nullptr; // unused as nicoup=0
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ fptype cIPD[2];
   __device__ __constant__ fptype* cIPC = nullptr; // unused as nicoup=0
 #else
@@ -90,7 +89,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -118,13 +117,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -151,7 +150,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -187,7 +186,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -200,8 +199,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -302,7 +303,7 @@ namespace mg5amcCpu
         { 16, -2 },
         { -2, 16 } }; // 2-D array[2][2]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -359,7 +360,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -418,7 +419,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -465,8 +466,8 @@ namespace mg5amcCpu
       { 1, 1, -1, -1 },
       { 1, 1, 1, 1 },
       { 1, 1, 1, -1 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -506,9 +507,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     const fptype tIPD[2] = { (fptype)m_pars->mdl_MT, (fptype)m_pars->mdl_WT };
     //const cxtype tIPC[0] = { ... }; // nicoup=0
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) ) );
-    //checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ) ); // nicoup=0
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) );
+    //gpuMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
 #else
     memcpy( cIPD, tIPD, 2 * sizeof( fptype ) );
     //memcpy( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
@@ -544,7 +545,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -573,6 +574,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -586,6 +591,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -609,12 +616,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -635,7 +642,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -761,9 +768,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -787,7 +794,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -807,7 +814,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 256 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -821,9 +828,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -851,7 +861,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -1061,7 +1071,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P1_gg_ttx/CPPProcess.h b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P1_gg_ttx/CPPProcess.h
index 3ebd92c038..4a88a07226 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P1_gg_ttx/CPPProcess.h
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P1_gg_ttx/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -107,7 +107,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -120,7 +120,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -150,7 +150,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P1_gg_ttx/CudaRuntime.h b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P1_gg_ttx/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P1_gg_ttx/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P1_gg_ttx/GpuAbstraction.h b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P1_gg_ttx/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P1_gg_ttx/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P1_gg_ttx/GpuRuntime.h b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P1_gg_ttx/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P1_gg_ttx/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P1_gg_ttx/check_sa.cc b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P1_gg_ttx/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P1_gg_ttx/check_sa.cc
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P1_gg_ttx/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P2_gg_ttxg/CPPProcess.cc b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P2_gg_ttxg/CPPProcess.cc
index bfab81142d..f80a0127b0 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P2_gg_ttxg/CPPProcess.cc
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P2_gg_ttxg/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_sm.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -46,7 +45,7 @@
 // Class member functions for calculating the matrix elements for
 // Process: g g > t t~ g WEIGHTED<=3 @2
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -80,7 +79,7 @@ namespace mg5amcCpu
   __device__ const fptype cIPD[2] = { (fptype)Parameters_sm::mdl_MT, (fptype)Parameters_sm::mdl_WT };
   __device__ const fptype* cIPC = nullptr; // unused as nicoup=0
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ fptype cIPD[2];
   __device__ __constant__ fptype* cIPC = nullptr; // unused as nicoup=0
 #else
@@ -90,7 +89,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -118,13 +117,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -151,7 +150,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -187,7 +186,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -200,8 +199,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -505,7 +506,7 @@ namespace mg5amcCpu
         { 1, -8, 10, 1, 64, -8 },
         { 10, 1, 1, -8, -8, 64 } }; // 2-D array[6][6]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -562,7 +563,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -621,7 +622,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -684,8 +685,8 @@ namespace mg5amcCpu
       { 1, 1, 1, 1, 1 },
       { 1, 1, 1, -1, -1 },
       { 1, 1, 1, -1, 1 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -726,9 +727,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     const fptype tIPD[2] = { (fptype)m_pars->mdl_MT, (fptype)m_pars->mdl_WT };
     //const cxtype tIPC[0] = { ... }; // nicoup=0
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) ) );
-    //checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ) ); // nicoup=0
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) );
+    //gpuMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
 #else
     memcpy( cIPD, tIPD, 2 * sizeof( fptype ) );
     //memcpy( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
@@ -765,7 +766,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -794,6 +795,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -807,6 +812,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -830,12 +837,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -856,7 +863,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -982,9 +989,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -1008,7 +1015,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -1028,7 +1035,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 256 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -1042,9 +1049,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -1072,7 +1082,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -1282,7 +1292,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P2_gg_ttxg/CPPProcess.h b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P2_gg_ttxg/CPPProcess.h
index 3901ddcb20..d4b3c0445c 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P2_gg_ttxg/CPPProcess.h
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P2_gg_ttxg/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -107,7 +107,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -120,7 +120,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -150,7 +150,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P2_gg_ttxg/CudaRuntime.h b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P2_gg_ttxg/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P2_gg_ttxg/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P2_gg_ttxg/GpuAbstraction.h b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P2_gg_ttxg/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P2_gg_ttxg/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P2_gg_ttxg/GpuRuntime.h b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P2_gg_ttxg/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P2_gg_ttxg/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P2_gg_ttxg/check_sa.cc b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P2_gg_ttxg/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P2_gg_ttxg/check_sa.cc
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/P2_gg_ttxg/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/RamboSamplingKernels.cc b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/RamboSamplingKernels.cc
index da68aa9255..79abbcc4f8 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/RamboSamplingKernels.cc
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/RamboSamplingKernels.cc
@@ -1,11 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "RamboSamplingKernels.h"
 
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "MemoryAccessMomenta.h"
 #include "MemoryAccessRandomNumbers.h"
 #include "MemoryAccessWeights.h"
@@ -14,7 +14,7 @@
 
 #include <sstream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -92,7 +92,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingKernelDevice::RamboSamplingKernelDevice( const fptype energy,               // input: energy
                                                         const BufferRndNumMomenta& rndmom, // input: random numbers in [0,1]
                                                         BufferMomenta& momenta,            // output: momenta
@@ -135,7 +135,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   getMomentaInitialDevice( const fptype energy,
                            fptype* momenta )
@@ -147,17 +147,17 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   void
   RamboSamplingKernelDevice::getMomentaInitial()
   {
-    getMomentaInitialDevice<<<m_gpublocks, m_gputhreads>>>( m_energy, m_momenta.data() );
+    gpuLaunchKernel( getMomentaInitialDevice, m_gpublocks, m_gputhreads, m_energy, m_momenta.data() );
   }
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   getMomentaFinalDevice( const fptype energy,
                          const fptype* rndmom,
@@ -171,11 +171,11 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   void
   RamboSamplingKernelDevice::getMomentaFinal()
   {
-    getMomentaFinalDevice<<<m_gpublocks, m_gputhreads>>>( m_energy, m_rndmom.data(), m_momenta.data(), m_weights.data() );
+    gpuLaunchKernel( getMomentaFinalDevice, m_gpublocks, m_gputhreads, m_energy, m_rndmom.data(), m_momenta.data(), m_weights.data() );
   }
 #endif
 
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/RamboSamplingKernels.h b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/RamboSamplingKernels.h
index 184089efd7..7c214cd74b 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/RamboSamplingKernels.h
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/RamboSamplingKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef RAMBOSAMPLINGKERNELS_H
 #define RAMBOSAMPLINGKERNELS_H 1
@@ -10,7 +10,7 @@
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -93,7 +93,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating RAMBO phase space sampling on a GPU device
   class RamboSamplingKernelDevice final : public SamplingKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/RandomNumberKernels.h b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/RandomNumberKernels.h
index 188a72c2c9..21d63beeac 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/RandomNumberKernels.h
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/RandomNumberKernels.h
@@ -1,14 +1,14 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef RANDOMNUMBERKERNELS_H
 #define RANDOMNUMBERKERNELS_H 1
 
 #include "mgOnGpuConfig.h"
 
-// NB This must come AFTER mgOnGpuConfig.h which contains our definition of __global__ when __CUDACC__ is not defined
+// NB This must come AFTER mgOnGpuConfig.h which contains our definition of __global__ when MGONGPUCPP_GPUIMPL is not defined
 #ifndef MGONGPU_HAS_NO_CURAND
 //#include "curand.h"
 struct curandGenerator_st; // forward definition from curand.h
@@ -16,7 +16,7 @@ struct curandGenerator_st; // forward definition from curand.h
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/cudacpp.mk b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/cudacpp.mk
index 509307506b..df74dfc284 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/cudacpp.mk
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/cudacpp.mk
@@ -1,7 +1,7 @@
 # Copyright (C) 2020-2023 CERN and UCLouvain.
 # Licensed under the GNU Lesser General Public License (version 3 or later).
 # Created by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-# Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+# Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #=== Determine the name of this makefile (https://ftp.gnu.org/old-gnu/Manuals/make-3.80/html_node/make_17.html)
 #=== NB: use ':=' to ensure that the value of CUDACPP_MAKEFILE is not modified further down after including make_opts
@@ -42,10 +42,10 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure common compiler flags for C++ and CUDA
+#=== Configure common compiler flags for C++ and CUDA/HIP
 
 INCFLAGS = -I.
-OPTFLAGS = -O3 # this ends up in CUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
+OPTFLAGS = -O3 # this ends up in GPUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
 
 # Dependency on src directory
 MG5AMC_COMMONLIB = mg5amc_common
@@ -121,24 +121,45 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure the CUDA compiler
+#=== Configure the GPU compiler (CUDA or HIP)
+
+# FIXME! (AV 24.01.2024)
+# In the current implementation (without separate builds for C++ and CUDA/HIP), we first check for cudacc and hipcc in CUDA_HOME and HIP_HOME.
+# If CUDA_HOME or HIP_HOME are not set, try to determine them from the path to cudacc and hipcc.
+# While convoluted, this is currently necessary to allow disabling CUDA/HIP builds by setting CUDA_HOME or HIP_HOME to invalid paths.
+# This will (probably?) be fixed when separate C++ and CUDA/HIP builds are implemented (PR #775).
 
 # If CXX is not a single word (example "clang++ --gcc-toolchain...") then disable CUDA builds (issue #505)
-# This is because it is impossible to pass this to "CUFLAGS += -ccbin <host-compiler>" below
+# This is because it is impossible to pass this to "GPUFLAGS += -ccbin <host-compiler>" below
 ifneq ($(words $(subst ccache ,,$(CXX))),1) # allow at most "CXX=ccache <host-compiler>" from outside
   $(warning CUDA builds are not supported for multi-word CXX "$(CXX)")
   override CUDA_HOME=disabled
 endif
 
-# If CUDA_HOME is not set, try to set it from the location of nvcc
+# If CUDA_HOME is not set, try to set it from the path to nvcc
 ifndef CUDA_HOME
-  CUDA_HOME = $(patsubst %bin/nvcc,%,$(shell which nvcc 2>/dev/null))
+  CUDA_HOME = $(patsubst %/bin/nvcc,%,$(shell which nvcc 2>/dev/null))
   $(warning CUDA_HOME was not set: using "$(CUDA_HOME)")
 endif
 
-# Set NVCC as $(CUDA_HOME)/bin/nvcc if it exists
+# If HIP_HOME is not set, try to set it from the path to hipcc
+ifndef HIP_HOME
+  HIP_HOME = $(patsubst %/bin/hipcc,%,$(shell which hipcc 2>/dev/null))
+  $(warning HIP_HOME was not set: using "$(HIP_HOME)")
+endif
+
+# FIXME! (AV 24.01.2024)
+# In the current implementation (without separate builds for C++ and CUDA/HIP),
+# builds are performed for HIP only if CUDA is not found in the path.
+# If both CUDA and HIP are installed, HIP builds can be triggered by unsetting CUDA_HOME.
+# This will be fixed when separate C++ and CUDA/HIP builds are implemented (PR #775).
+
+#--- Option 1: CUDA exists -> use CUDA
+
+# Set GPUCC as $(CUDA_HOME)/bin/nvcc if it exists
 ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
-  NVCC = $(CUDA_HOME)/bin/nvcc
+
+  GPUCC = $(CUDA_HOME)/bin/nvcc
   USE_NVTX ?=-DUSE_NVTX
   # See https://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/index.html
   # See https://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/
@@ -148,7 +169,7 @@ ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
   # Examples: use 60 for P100 (Piz Daint), 80 for A100 (Juwels Booster, NVidia raplab/Curiosity).
   MADGRAPH_CUDA_ARCHITECTURE ?= 70
   ###CUARCHFLAGS = -gencode arch=compute_$(MADGRAPH_CUDA_ARCHITECTURE),code=compute_$(MADGRAPH_CUDA_ARCHITECTURE) -gencode arch=compute_$(MADGRAPH_CUDA_ARCHITECTURE),code=sm_$(MADGRAPH_CUDA_ARCHITECTURE) # Older implementation (AV): go back to this one for multi-GPU support #533
-  ###CUARCHFLAGS = --gpu-architecture=compute_$(MADGRAPH_CUDA_ARCHITECTURE) --gpu-code=sm_$(MADGRAPH_CUDA_ARCHITECTURE),compute_$(MADGRAPH_CUDA_ARCHITECTURE) # Newer implementation (SH): cannot use this as-is for multi-GPU support #533
+  ###CUARCHFLAGS = --gpu-architecture=compute_$(MADGRAPH_CUDA_ARCHITECTURE) --gpu-code=sm_$(MADGRAPH_CUDA_ARCHITECTURE),compute_$(MADGRAPH_CUDA_ARCHITECTURE)  # Newer implementation (SH): cannot use this as-is for multi-GPU support #533
   comma:=,
   CUARCHFLAGS = $(foreach arch,$(subst $(comma), ,$(MADGRAPH_CUDA_ARCHITECTURE)),-gencode arch=compute_$(arch),code=compute_$(arch) -gencode arch=compute_$(arch),code=sm_$(arch))
   CUINC = -I$(CUDA_HOME)/include/
@@ -158,41 +179,79 @@ ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
     CURANDLIBFLAGS = -L$(CUDA_HOME)/lib64/ -lcurand # NB: -lcuda is not needed here!
   endif
   CUOPTFLAGS = -lineinfo
-  CUFLAGS = $(foreach opt, $(OPTFLAGS), -Xcompiler $(opt)) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
-  ###CUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
-  ###NVCC_VERSION = $(shell $(NVCC) --version | grep 'Cuda compilation tools' | cut -d' ' -f5 | cut -d, -f1)
-  CUFLAGS += -std=c++17 # need CUDA >= 11.2 (see #333): this is enforced in mgOnGpuConfig.h
+  ###GPUFLAGS = $(OPTFLAGS) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
+  GPUFLAGS = $(foreach opt, $(OPTFLAGS), -Xcompiler $(opt)) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
+  ###GPUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
+  ###GPUCC_VERSION = $(shell $(GPUCC) --version | grep 'Cuda compilation tools' | cut -d' ' -f5 | cut -d, -f1)
+  GPUFLAGS += -std=c++17 # need CUDA >= 11.2 (see #333): this is enforced in mgOnGpuConfig.h
   # Without -maxrregcount: baseline throughput: 6.5E8 (16384 32 12) up to 7.3E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 160 # improves throughput: 6.9E8 (16384 32 12) up to 7.7E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 128 # improves throughput: 7.3E8 (16384 32 12) up to 7.6E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 96 # degrades throughput: 4.1E8 (16384 32 12) up to 4.5E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 64 # degrades throughput: 1.7E8 (16384 32 12) flat at 1.7E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 160 # improves throughput: 6.9E8 (16384 32 12) up to 7.7E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 128 # improves throughput: 7.3E8 (16384 32 12) up to 7.6E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 96 # degrades throughput: 4.1E8 (16384 32 12) up to 4.5E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 64 # degrades throughput: 1.7E8 (16384 32 12) flat at 1.7E8 (65536 128 12)
+  CUBUILDRULEFLAGS = -Xcompiler -fPIC -c
+  CCBUILDRULEFLAGS = -Xcompiler -fPIC -c -x cu
+  CUDATESTFLAGS = -lcuda
+
+  # Set the host C++ compiler for GPUCC via "-ccbin <host-compiler>"
+  # (NB issue #505: this must be a single word, "clang++ --gcc-toolchain..." is not supported)
+  GPUFLAGS += -ccbin $(shell which $(subst ccache ,,$(CXX)))
+
+  # Allow newer (unsupported) C++ compilers with older versions of CUDA if ALLOW_UNSUPPORTED_COMPILER_IN_CUDA is set (#504)
+  ifneq ($(origin ALLOW_UNSUPPORTED_COMPILER_IN_CUDA),undefined)
+  GPUFLAGS += -allow-unsupported-compiler
+  endif
+
 else ifneq ($(origin REQUIRE_CUDA),undefined)
+
   # If REQUIRE_CUDA is set but no cuda is found, stop here (e.g. for CI tests on GPU #443)
-  $(error No cuda installation found (set CUDA_HOME or make nvcc visible in PATH))
+  $(error No cuda installation found (set CUDA_HOME or make GPUCC visible in PATH))
+
+#--- Option 2: CUDA does not exist, HIP exists -> use HIP
+
+# Set GPUCC as $(HIP_HOME)/bin/hipcc if it exists
+else ifneq ($(wildcard $(HIP_HOME)/bin/hipcc),)
+
+  GPUCC = $(HIP_HOME)/bin/hipcc
+  #USE_NVTX ?=-DUSE_NVTX # should maybe find something equivalent to this in HIP?
+  HIPARCHFLAGS = -target x86_64-linux-gnu --offload-arch=gfx90a
+  HIPINC = -I$(HIP_HOME)/include/
+  # Note: -DHIP_FAST_MATH is equivalent to -use_fast_math in HIP 
+  # (but only for single precision line 208: https://rocm-developer-tools.github.io/HIP/hcc__detail_2math__functions_8h_source.html)
+  # Note: CUOPTFLAGS should not be used for HIP, it had been added here but was then removed (#808)
+  GPUFLAGS = $(OPTFLAGS) $(INCFLAGS) $(HIPINC) $(HIPARCHFLAGS) -DHIP_FAST_MATH -DHIP_PLATFORM=amd -fPIC
+  ###GPUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
+  GPUFLAGS += -std=c++17
+  ###GPUFLAGS+= --maxrregcount 255 # (AV: is this option valid on HIP and meaningful on AMD GPUs?)
+  CUBUILDRULEFLAGS = -fPIC -c
+  CCBUILDRULEFLAGS = -fPIC -c -x hip
+
+else ifneq ($(origin REQUIRE_HIP),undefined)
+
+  # If REQUIRE_HIP is set but no HIP is found, stop here (e.g. for CI tests on GPU #443)
+  $(error No hip installation found (set HIP_HOME or make GPUCC visible in PATH))
+
+#--- Option 3: CUDA does not exist, HIP does not exist -> switch off both CUDA and HIP
+
 else
-  # No cuda. Switch cuda compilation off and go to common random numbers in C++
+
+  # No cudacc and no hipcc: switch CUDA and HIP compilation off and go to common random numbers in C++
   $(warning CUDA_HOME is not set or is invalid: export CUDA_HOME to compile with cuda)
-  override NVCC=
+  $(warning HIP_HOME is not set or is invalid: export HIP_HOME to compile with hip)
+  override GPUCC=
   override USE_NVTX=
   override CUINC=
   override CURANDLIBFLAGS=
-endif
-export NVCC
-export CUFLAGS
 
-# Set the host C++ compiler for nvcc via "-ccbin <host-compiler>"
-# (NB issue #505: this must be a single word, "clang++ --gcc-toolchain..." is not supported)
-CUFLAGS += -ccbin $(shell which $(subst ccache ,,$(CXX)))
-
-# Allow newer (unsupported) C++ compilers with older versions of CUDA if ALLOW_UNSUPPORTED_COMPILER_IN_CUDA is set (#504)
-ifneq ($(origin ALLOW_UNSUPPORTED_COMPILER_IN_CUDA),undefined)
-CUFLAGS += -allow-unsupported-compiler
 endif
 
+# Export GPUCC (so that it can also be used in cudacpp_src.mk?)
+export GPUCC
+export GPUFLAGS
+
 #-------------------------------------------------------------------------------
 
-#=== Configure ccache for C++ and CUDA builds
+#=== Configure ccache for C++ and CUDA/HIP builds
 
 # Enable ccache if USECCACHE=1
 ifeq ($(USECCACHE)$(shell echo $(CXX) | grep ccache),1)
@@ -201,15 +260,15 @@ endif
 #ifeq ($(USECCACHE)$(shell echo $(AR) | grep ccache),1)
 #  override AR:=ccache $(AR)
 #endif
-ifneq ($(NVCC),)
-  ifeq ($(USECCACHE)$(shell echo $(NVCC) | grep ccache),1)
-    override NVCC:=ccache $(NVCC)
+ifneq ($(GPUCC),)
+  ifeq ($(USECCACHE)$(shell echo $(GPUCC) | grep ccache),1)
+    override GPUCC:=ccache $(GPUCC)
   endif
 endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure PowerPC-specific compiler flags for C++ and CUDA
+#=== Configure PowerPC-specific compiler flags for C++ and CUDA/HIP
 
 # PowerPC-specific CXX compiler flags (being reviewed)
 ifeq ($(UNAME_P),ppc64le)
@@ -225,9 +284,9 @@ else
   ######CXXFLAGS+= -fno-semantic-interposition # no benefit (neither alone, nor combined with -flto)
 endif
 
-# PowerPC-specific CUDA compiler flags (to be reviewed!)
+# PowerPC-specific CUDA/HIP compiler flags (to be reviewed!)
 ifeq ($(UNAME_P),ppc64le)
-  CUFLAGS+= -Xcompiler -mno-float128
+  GPUFLAGS+= -Xcompiler -mno-float128
 endif
 
 #-------------------------------------------------------------------------------
@@ -235,9 +294,11 @@ endif
 #=== Configure defaults and check if user-defined choices exist for OMPFLAGS, AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
 
 # Set the default OMPFLAGS choice
-ifneq ($(shell $(CXX) --version | egrep '^Intel'),)
+ifneq ($(findstring hipcc,$(GPUCC)),)
+override OMPFLAGS = # disable OpenMP MT when using hipcc #802
+else ifneq ($(shell $(CXX) --version | egrep '^Intel'),)
 override OMPFLAGS = -fopenmp
-###override OMPFLAGS = # disable OpenMP MT on Intel (was ok without nvcc but not ok with nvcc before #578)
+###override OMPFLAGS = # disable OpenMP MT on Intel (was ok without GPUCC but not ok with GPUCC before #578)
 else ifneq ($(shell $(CXX) --version | egrep '^(clang)'),)
 override OMPFLAGS = -fopenmp
 ###override OMPFLAGS = # disable OpenMP MT on clang (was not ok without or with nvcc before #578)
@@ -293,7 +354,10 @@ endif
 
 # Set the default RNDGEN (random number generator) choice
 ifeq ($(RNDGEN),)
-  ifeq ($(NVCC),)
+  ifeq ($(GPUCC),)
+    override RNDGEN = hasNoCurand
+  # Edgecase for HIP compilation
+  else ifeq ($(findstring hipcc,$(GPUCC)),hipcc)
     override RNDGEN = hasNoCurand
   else ifeq ($(RNDGEN),)
     override RNDGEN = hasCurand
@@ -310,7 +374,7 @@ export OMPFLAGS
 
 #-------------------------------------------------------------------------------
 
-#=== Set the CUDA/C++ compiler flags appropriate to user-defined choices of AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
+#=== Set the CUDA/HIP/C++ compiler flags appropriate to user-defined choices of AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
 
 # Set the build flags appropriate to OMPFLAGS
 $(info OMPFLAGS=$(OMPFLAGS))
@@ -368,13 +432,13 @@ CXXFLAGS+= $(AVXFLAGS)
 $(info FPTYPE=$(FPTYPE))
 ifeq ($(FPTYPE),d)
   CXXFLAGS += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
-  CUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
+  GPUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
 else ifeq ($(FPTYPE),f)
   CXXFLAGS += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
-  CUFLAGS  += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
+  GPUFLAGS  += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
 else ifeq ($(FPTYPE),m)
   CXXFLAGS += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
-  CUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
+  GPUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
 else
   $(error Unknown FPTYPE='$(FPTYPE)': only 'd', 'f' and 'm' are supported)
 endif
@@ -383,7 +447,7 @@ endif
 $(info HELINL=$(HELINL))
 ifeq ($(HELINL),1)
   CXXFLAGS += -DMGONGPU_INLINE_HELAMPS
-  CUFLAGS  += -DMGONGPU_INLINE_HELAMPS
+  GPUFLAGS  += -DMGONGPU_INLINE_HELAMPS
 else ifneq ($(HELINL),0)
   $(error Unknown HELINL='$(HELINL)': only '0' and '1' are supported)
 endif
@@ -392,7 +456,7 @@ endif
 $(info HRDCOD=$(HRDCOD))
 ifeq ($(HRDCOD),1)
   CXXFLAGS += -DMGONGPU_HARDCODE_PARAM
-  CUFLAGS  += -DMGONGPU_HARDCODE_PARAM
+  GPUFLAGS  += -DMGONGPU_HARDCODE_PARAM
 else ifneq ($(HRDCOD),0)
   $(error Unknown HRDCOD='$(HRDCOD)': only '0' and '1' are supported)
 endif
@@ -444,11 +508,11 @@ ifeq ($(UNAME_S),Darwin)
   override CULIBFLAGSRPATH2 =
 else
   # RPATH to cuda/cpp libs when linking executables
-  override CXXLIBFLAGSRPATH = -Wl,-rpath,$(LIBDIRRPATH)
-  override CULIBFLAGSRPATH = -Xlinker -rpath,$(LIBDIRRPATH)
+  override CXXLIBFLAGSRPATH = -Wl,-rpath=$(LIBDIRRPATH)
+  override CULIBFLAGSRPATH = -Xlinker -rpath=$(LIBDIRRPATH)
   # RPATH to common lib when linking cuda/cpp libs
-  override CXXLIBFLAGSRPATH2 = -Wl,-rpath,'$$ORIGIN'
-  override CULIBFLAGSRPATH2 = -Xlinker -rpath,'$$ORIGIN'
+  override CXXLIBFLAGSRPATH2 = -Wl,-rpath='$$ORIGIN'
+  override CULIBFLAGSRPATH2 = -Xlinker -rpath='$$ORIGIN'
 endif
 
 # Setting LD_LIBRARY_PATH or DYLD_LIBRARY_PATH in the RUNTIME is no longer necessary (neither on Linux nor on Mac)
@@ -461,7 +525,7 @@ override RUNTIME =
 cxx_main=$(BUILDDIR)/check.exe
 fcxx_main=$(BUILDDIR)/fcheck.exe
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 cu_main=$(BUILDDIR)/gcheck.exe
 fcu_main=$(BUILDDIR)/fgcheck.exe
 else
@@ -492,14 +556,15 @@ $(BUILDDIR)/.build.$(TAG):
 	@touch $(BUILDDIR)/.build.$(TAG)
 
 # Generic target and build rules: objects from CUDA compilation
-ifneq ($(NVCC),)
+# NB: CCBUILDRULEFLAGS includes "-x cu" for nvcc and "-x hip" for hipcc (#810)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/%.o : %.cu *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $(CUBUILDRULEFLAGS) $< -o $@
 
 $(BUILDDIR)/%_cu.o : %.cc *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c -x cu $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $(CCBUILDRULEFLAGS) $< -o $@
 endif
 
 # Generic target and build rules: objects from C++ compilation
@@ -509,11 +574,14 @@ $(BUILDDIR)/%.o : %.cc *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	$(CXX) $(CPPFLAGS) $(CXXFLAGS) -fPIC -c $< -o $@
 
 # Apply special build flags only to CrossSectionKernel.cc and gCrossSectionKernel.cu (no fast math, see #117 and #516)
+# Added edgecase for HIP compilation
 ifeq ($(shell $(CXX) --version | grep ^nvc++),)
 $(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS := $(filter-out -ffast-math,$(CXXFLAGS))
 $(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS += -fno-fast-math
-ifneq ($(NVCC),)
-$(BUILDDIR)/gCrossSectionKernels.o: CUFLAGS += -Xcompiler -fno-fast-math
+ifeq ($(findstring nvcc,$(GPUCC)),nvcc)
+  $(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -Xcompiler -fno-fast-math
+else
+  $(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -fno-fast-math
 endif
 endif
 
@@ -530,10 +598,10 @@ ifeq ($(RNDGEN),hasCurand)
 $(BUILDDIR)/CurandRandomNumberKernel.o: CXXFLAGS += $(CUINC)
 endif
 
-# Avoid "warning: builtin __has_trivial_... is deprecated; use __is_trivially_... instead" in nvcc with icx2023 (#592)
+# Avoid "warning: builtin __has_trivial_... is deprecated; use __is_trivially_... instead" in GPUCC with icx2023 (#592)
 ifneq ($(shell $(CXX) --version | egrep '^(Intel)'),)
-ifneq ($(NVCC),)
-CUFLAGS += -Xcompiler -Wno-deprecated-builtins
+ifneq ($(GPUCC),)
+GPUFLAGS += -Wno-deprecated-builtins
 endif
 endif
 
@@ -541,8 +609,8 @@ endif
 # This patch does remove the warning, but I prefer to keep it disabled for the moment...
 ###ifneq ($(shell $(CXX) --version | egrep '^(clang|Apple clang|Intel)'),)
 ###$(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS += -Wno-overriding-t-option
-###ifneq ($(NVCC),)
-###$(BUILDDIR)/gCrossSectionKernels.o: CUFLAGS += -Xcompiler -Wno-overriding-t-option
+###ifneq ($(GPUCC),)
+###$(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -Xcompiler -Wno-overriding-t-option
 ###endif
 ###endif
 
@@ -569,7 +637,7 @@ MG5AMC_CXXLIB = mg5amc_$(processid_short)_cpp
 cxx_objects_lib=$(BUILDDIR)/CPPProcess.o $(BUILDDIR)/MatrixElementKernels.o $(BUILDDIR)/BridgeKernels.o $(BUILDDIR)/CrossSectionKernels.o
 cxx_objects_exe=$(BUILDDIR)/CommonRandomNumberKernel.o $(BUILDDIR)/RamboSamplingKernels.o
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 MG5AMC_CULIB = mg5amc_$(processid_short)_cuda
 cu_objects_lib=$(BUILDDIR)/gCPPProcess.o $(BUILDDIR)/gMatrixElementKernels.o $(BUILDDIR)/gBridgeKernels.o $(BUILDDIR)/gCrossSectionKernels.o
 cu_objects_exe=$(BUILDDIR)/gCommonRandomNumberKernel.o $(BUILDDIR)/gRamboSamplingKernels.o
@@ -581,11 +649,17 @@ $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: cxx_objects_lib += $(BUILDDIR)/fbridge.o
 $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib)
 	$(CXX) -shared -o $@ $(cxx_objects_lib) $(CXXLIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: $(BUILDDIR)/fbridge_cu.o
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: cu_objects_lib += $(BUILDDIR)/fbridge_cu.o
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cu_objects_lib)
-	$(NVCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+# Bypass std::filesystem completely to ease portability on LUMI #803
+#ifneq ($(findstring hipcc,$(GPUCC)),)
+#	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB) -lstdc++fs
+#else
+#	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+#endif
 endif
 
 #-------------------------------------------------------------------------------
@@ -602,16 +676,16 @@ $(cxx_main): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PAT
 $(cxx_main): $(BUILDDIR)/check_sa.o $(LIBDIR)/lib$(MG5AMC_CXXLIB).so $(cxx_objects_exe) $(BUILDDIR)/CurandRandomNumberKernel.o
 	$(CXX) -o $@ $(BUILDDIR)/check_sa.o $(OMPFLAGS) -ldl -pthread $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe) $(BUILDDIR)/CurandRandomNumberKernel.o $(CURANDLIBFLAGS)
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(cu_main): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(cu_main): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(cu_main): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(cu_main): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 else ifneq ($(shell $(CXX) --version | grep ^nvc++),) # support nvc++ #531
 $(cu_main): LIBFLAGS += -L$(patsubst %bin/nvc++,%lib,$(subst ccache ,,$(CXX))) -lnvhpcatm -lnvcpumath -lnvc
 endif
 $(cu_main): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(cu_main): $(BUILDDIR)/gcheck_sa.o $(LIBDIR)/lib$(MG5AMC_CULIB).so $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o
-	$(NVCC) -o $@ $(BUILDDIR)/gcheck_sa.o $(CUARCHFLAGS) $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o $(CURANDLIBFLAGS)
+	$(GPUCC) -o $@ $(BUILDDIR)/gcheck_sa.o $(CUARCHFLAGS) $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o $(CURANDLIBFLAGS)
 endif
 
 #-------------------------------------------------------------------------------
@@ -635,19 +709,27 @@ $(fcxx_main): LIBFLAGS += -L$(shell dirname $(shell $(FC) --print-file-name libg
 endif
 $(fcxx_main): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(fcxx_main): $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler.o $(LIBDIR)/lib$(MG5AMC_CXXLIB).so $(cxx_objects_exe)
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(BUILDDIR)/fcheck_sa.o $(OMPFLAGS) $(BUILDDIR)/fsampler.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe) -lstdc++
+else
 	$(CXX) -o $@ $(BUILDDIR)/fcheck_sa.o $(OMPFLAGS) $(BUILDDIR)/fsampler.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe)
+endif
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(fcu_main): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(fcu_main): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(fcu_main): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(fcu_main): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 endif
 ifeq ($(UNAME_S),Darwin)
 $(fcu_main): LIBFLAGS += -L$(shell dirname $(shell $(FC) --print-file-name libgfortran.dylib)) # add path to libgfortran on Mac #375
 endif
 $(fcu_main): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(fcu_main): $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBDIR)/lib$(MG5AMC_CULIB).so $(cu_objects_exe)
-	$(NVCC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe)
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) -lstdc++ -L$(shell dirname $(shell $(GPUCC) -print-prog-name=clang))/../../lib -lamdhip64
+else
+	$(GPUCC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe)
+endif
 endif
 
 #-------------------------------------------------------------------------------
@@ -659,7 +741,7 @@ $(BUILDDIR)/testxxx.o: testxxx_cc_ref.txt
 $(testmain): $(BUILDDIR)/testxxx.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/testxxx.o # Comment out this line to skip the C++ test of xxx functions
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/testxxx_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/testxxx_cu.o: INCFLAGS += $(GTESTINC)
 $(BUILDDIR)/testxxx_cu.o: testxxx_cc_ref.txt
@@ -672,7 +754,7 @@ $(BUILDDIR)/testmisc.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/testmisc.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/testmisc.o # Comment out this line to skip the C++ miscellaneous tests
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/testmisc_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/testmisc_cu.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/testmisc_cu.o
@@ -684,12 +766,12 @@ $(BUILDDIR)/runTest.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/runTest.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/runTest.o
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/runTest_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/runTest_cu.o: INCFLAGS += $(GTESTINC)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(testmain): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(testmain): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(testmain): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(testmain): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 else ifneq ($(shell $(CXX) --version | grep ^nvc++),) # support nvc++ #531
 $(testmain): LIBFLAGS += -L$(patsubst %bin/nvc++,%lib,$(subst ccache ,,$(CXX))) -lnvhpcatm -lnvcpumath -lnvc
 endif
@@ -713,14 +795,23 @@ $(testmain): LIBFLAGS += -lgomp
 endif
 endif
 
-ifeq ($(NVCC),) # link only runTest.o
+# Bypass std::filesystem completely to ease portability on LUMI #803
+#ifneq ($(findstring hipcc,$(GPUCC)),)
+#$(testmain): LIBFLAGS += -lstdc++fs
+#endif
+
+ifeq ($(GPUCC),) # link only runTest.o
 $(testmain): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(testmain): $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib) $(cxx_objects_exe) $(GTESTLIBS)
 	$(CXX) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) -ldl -pthread $(LIBFLAGS)
 else # link both runTest.o and runTest_cu.o
 $(testmain): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(testmain): $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) $(GTESTLIBS)
-	$(NVCC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) -lcuda
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) $(CUDATESTFLAGS) -lstdc++ -lpthread  -L$(shell dirname $(shell $(GPUCC) -print-prog-name=clang))/../../lib -lamdhip64
+else
+	$(GPUCC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) $(CUDATESTFLAGS)
+endif
 endif
 
 # Use target gtestlibs to build only googletest
@@ -829,9 +920,9 @@ ifeq ($(USECCACHE),1)
 	ccache --version | head -1
 endif
 	@echo ""
-	@echo NVCC=$(NVCC)
-ifneq ($(NVCC),)
-	$(NVCC) --version
+	@echo GPUCC=$(GPUCC)
+ifneq ($(GPUCC),)
+	$(GPUCC) --version
 endif
 	@echo ""
 	@echo CXX=$(CXX)
@@ -850,7 +941,7 @@ endif
 
 # Target: check (run the C++ test executable)
 # [NB THIS IS WHAT IS USED IN THE GITHUB CI!]
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 check: runTest cmpFcheck cmpFGcheck
 else
 check: runTest cmpFcheck
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/fbridge.cc b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/fbridge.cc
index 2d2b36d560..22ce3f5115 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/fbridge.cc
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/fbridge.cc
@@ -1,11 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Roiser (Oct 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "Bridge.h"
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 
 extern "C"
 {
@@ -22,7 +22,7 @@ extern "C"
    * Using the same Fortran MadEvent code, linking to the hetrerogeneous library would allow access to both CPU and GPU implementations.
    * The specific heterogeneous configuration (how many GPUs, how many threads on each CPU, etc) could be loaded in CUDA/C++ from a data file.
    */
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -46,8 +46,8 @@ extern "C"
    */
   void fbridgecreate_( CppObjectInFortran** ppbridge, const int* pnevtF, const int* pnparF, const int* pnp4F )
   {
-#ifdef __CUDACC__
-    CudaRuntime::setUp();
+#ifdef MGONGPUCPP_GPUIMPL
+    GpuRuntime::setUp();
 #endif
     // (NB: CPPProcess::initProc no longer needs to be executed here because it is called in the Bridge constructor)
     // FIXME: disable OMP in Bridge when called from Fortran
@@ -65,8 +65,8 @@ extern "C"
     Bridge<FORTRANFPTYPE>* pbridge = dynamic_cast<Bridge<FORTRANFPTYPE>*>( *ppbridge );
     if( pbridge == 0 ) throw std::runtime_error( "fbridgedelete_: invalid Bridge address" );
     delete pbridge;
-#ifdef __CUDACC__
-    CudaRuntime::tearDown();
+#ifdef MGONGPUCPP_GPUIMPL
+    GpuRuntime::tearDown();
 #endif
   }
 
@@ -96,7 +96,7 @@ extern "C"
   {
     Bridge<FORTRANFPTYPE>* pbridge = dynamic_cast<Bridge<FORTRANFPTYPE>*>( *ppbridge );
     if( pbridge == 0 ) throw std::runtime_error( "fbridgesequence_: invalid Bridge address" );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Use the device/GPU implementation in the CUDA library
     // (there is also a host implementation in this library)
     pbridge->gpu_sequence( momenta, gs, rndhel, rndcol, *pchannelId, mes, selhel, selcol );
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/fsampler.cc b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/fsampler.cc
index 2fb445372d..3743934f41 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/fsampler.cc
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/fsampler.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Feb 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "mgOnGpuConfig.h"
 
@@ -13,7 +13,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -40,7 +40,7 @@ namespace mg5amcCpu
   private:
     const int m_nevt; // The number of events in each iteration
     int m_iiter;      // The iteration counter (for random number seeding)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     HostBufferRndNumMomenta m_hstRndmom; // Memory buffers for random numbers
     HostBufferMomenta m_hstMomenta;      // Memory buffers for momenta
     HostBufferWeights m_hstWeights;      // Memory buffers for sampling weights
@@ -105,7 +105,7 @@ namespace mg5amcCpu
 
 extern "C"
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/runTest.cc b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/runTest.cc
index d4a760a71b..de327f2321 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/runTest.cc
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/runTest.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Hageboeck (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*xxx to run only testxxx.cc tests
 //----------------------------------------------------------------------------
@@ -18,7 +18,7 @@
 #include "RandomNumberKernels.h"
 #include "epoch_process_id.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using namespace mg5amcGpu;
 #else
 using namespace mg5amcCpu;
@@ -35,7 +35,7 @@ struct CUDA_CPU_TestBase : public TestDriverBase
     : TestDriverBase( npar, refFileName ) {}
 };
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 struct CPUTest : public CUDA_CPU_TestBase
 {
   // Struct data members (process, and memory structures for random numbers, momenta, matrix elements and weights on host and device)
@@ -119,7 +119,7 @@ struct CPUTest : public CUDA_CPU_TestBase
 };
 #endif
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 struct CUDATest : public CUDA_CPU_TestBase
 {
   // Reset the device when our test goes out of scope. Note that this should happen after
@@ -128,7 +128,7 @@ struct CUDATest : public CUDA_CPU_TestBase
   {
     ~DeviceReset()
     {
-      checkCuda( cudaDeviceReset() ); // this is needed by cuda-memcheck --leak-check full
+      checkGpu( gpuDeviceReset() ); // this is needed by cuda-memcheck --leak-check full
     }
   } deviceResetter;
 
@@ -256,7 +256,7 @@ INSTANTIATE_TEST_SUITE_P( prefix, \
                           test_suite_name, \
                           testing::Values( new CUDATest( MG_EPOCH_REFERENCE_FILE_NAME ) ) );
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 MG_INSTANTIATE_TEST_SUITE_GPU( XTESTID_GPU( MG_EPOCH_PROCESS_ID ), MadgraphTest );
 #else
 MG_INSTANTIATE_TEST_SUITE_CPU( XTESTID_CPU( MG_EPOCH_PROCESS_ID ), MadgraphTest );
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/testmisc.cc b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/testmisc.cc
index 895d6eeb56..ba9e59a8a3 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/testmisc.cc
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/testmisc.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*misc to run only testmisc.cc tests
 //----------------------------------------------------------------------------
@@ -17,7 +17,7 @@
 #include <sstream>
 #include <typeinfo>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #define TESTID( s ) s##_GPU_MISC
 #else
 #define TESTID( s ) s##_CPU_MISC
@@ -26,7 +26,7 @@
 #define XTESTID( s ) TESTID( s )
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -59,7 +59,7 @@ namespace mg5amcCpu
 
 TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testmisc )
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/testxxx.cc b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/testxxx.cc
index 3361fe5aa9..e5167de00c 100644
--- a/epochX/cudacpp/gg_tt01g.mad/SubProcesses/testxxx.cc
+++ b/epochX/cudacpp/gg_tt01g.mad/SubProcesses/testxxx.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Apr 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*xxx to run only testxxx.cc tests
 //----------------------------------------------------------------------------
@@ -24,7 +24,7 @@
 #include <iomanip>
 #include <iostream>
 #include <vector>
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #define TESTID( s ) s##_GPU_XXX
 #else
 #define TESTID( s ) s##_CPU_XXX
@@ -32,7 +32,7 @@
 
 #define XTESTID( s ) TESTID( s )
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -42,7 +42,7 @@ namespace mg5amcCpu
   int FPEhandlerIevt = -1;
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU): '" << FPEhandlerMessage << "' ievt=" << FPEhandlerIevt << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU neppV=" << neppV << "): '" << FPEhandlerMessage << "' ievt=" << FPEhandlerIevt << std::endl;
@@ -53,7 +53,7 @@ namespace mg5amcCpu
 
 TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -77,7 +77,7 @@ TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
   assert( nevt % neppM == 0 ); // nevt must be a multiple of neppM
   assert( nevt % neppV == 0 ); // nevt must be a multiple of neppV
   // Fill in the input momenta
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   mg5amcGpu::PinnedHostBufferMomenta hstMomenta( nevt ); // AOSOA[npagM][npar=4][np4=4][neppM]
 #else
   mg5amcCpu::HostBufferMomenta hstMomenta( nevt ); // AOSOA[npagM][npar=4][np4=4][neppM]
@@ -322,7 +322,7 @@ TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
   {
     for( int ievt = 0; ievt < nevt; ievt++ )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       using namespace mg5amcGpu;
 #else
       using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_tt01g.mad/bin/internal/launch_plugin.py b/epochX/cudacpp/gg_tt01g.mad/bin/internal/launch_plugin.py
index 3b09713e12..90af6b7053 100644
--- a/epochX/cudacpp/gg_tt01g.mad/bin/internal/launch_plugin.py
+++ b/epochX/cudacpp/gg_tt01g.mad/bin/internal/launch_plugin.py
@@ -103,7 +103,7 @@ def default_setup(self):
                        fct_mod=(self.reset_makeopts,(),{}),
                        allowed=['auto', 'none', 'sse4', 'avx2','512y','512z'])
         self.add_param('cudacpp_backend', 'CPP', include=False, hidden=False,
-                       allowed=['Fortan', 'CPP', 'CUDA'])
+                       allowed=['Fortran', 'CPP', 'CUDA'])
         self['vector_size'] = 16 # already setup in default class (just change value)
         self['aloha_flag'] = '--fast-math'
         self['matrix_flag'] = '-O3'
diff --git a/epochX/cudacpp/gg_tt01g.mad/src/HelAmps_sm.h b/epochX/cudacpp/gg_tt01g.mad/src/HelAmps_sm.h
index 361b488401..0dd0f3ebba 100644
--- a/epochX/cudacpp/gg_tt01g.mad/src/HelAmps_sm.h
+++ b/epochX/cudacpp/gg_tt01g.mad/src/HelAmps_sm.h
@@ -5,7 +5,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -28,7 +28,7 @@
 //#include <iomanip>
 //#include <iostream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt01g.mad/src/Parameters_sm.cc b/epochX/cudacpp/gg_tt01g.mad/src/Parameters_sm.cc
index 64fc3fea62..067445b198 100644
--- a/epochX/cudacpp/gg_tt01g.mad/src/Parameters_sm.cc
+++ b/epochX/cudacpp/gg_tt01g.mad/src/Parameters_sm.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -17,7 +17,7 @@
 #include <iomanip>
 #include <iostream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using namespace mg5amcGpu;
 #else
 using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_tt01g.mad/src/Parameters_sm.h b/epochX/cudacpp/gg_tt01g.mad/src/Parameters_sm.h
index b6568d3761..9581d66e0e 100644
--- a/epochX/cudacpp/gg_tt01g.mad/src/Parameters_sm.h
+++ b/epochX/cudacpp/gg_tt01g.mad/src/Parameters_sm.h
@@ -27,7 +27,7 @@
 #include "read_slha.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -93,7 +93,7 @@ namespace mg5amcCpu
 #include <limits>
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -218,7 +218,7 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -239,7 +239,7 @@ namespace mg5amcCpu
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wunused-variable"  // e.g. <<warning: unused variable ‘mdl_G__exp__2’ [-Wunused-variable]>>
 #pragma GCC diagnostic ignored "-Wunused-parameter" // e.g. <<warning: unused parameter ‘G’ [-Wunused-parameter]>>
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #pragma nv_diagnostic push
 #pragma nv_diag_suppress 177 // e.g. <<warning #177-D: variable "mdl_G__exp__2" was declared but never referenced>>
 #endif
@@ -267,7 +267,7 @@ namespace mg5amcCpu
       // End SM implementation - no special handling of vectors of floats as in EFT (#439)
       return out;
     }
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #pragma GCC diagnostic pop
 #pragma nv_diagnostic pop
 #endif
diff --git a/epochX/cudacpp/gg_tt01g.mad/src/cudacpp_src.mk b/epochX/cudacpp/gg_tt01g.mad/src/cudacpp_src.mk
index d4cc628aec..b2b9da5288 100644
--- a/epochX/cudacpp/gg_tt01g.mad/src/cudacpp_src.mk
+++ b/epochX/cudacpp/gg_tt01g.mad/src/cudacpp_src.mk
@@ -19,7 +19,7 @@ SHELL := /bin/bash
 #=== Configure common compiler flags for CUDA and C++
 
 INCFLAGS = -I.
-OPTFLAGS = -O3 # this ends up in CUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
+OPTFLAGS = -O3 # this ends up in GPUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
 
 #-------------------------------------------------------------------------------
 
@@ -45,13 +45,13 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure the CUDA compiler (note: NVCC is already exported including ccache)
+#=== Configure the CUDA compiler (note: GPUCC is already exported including ccache)
 
-###$(info NVCC=$(NVCC))
+###$(info GPUCC=$(GPUCC))
 
 #-------------------------------------------------------------------------------
 
-#=== Configure ccache for C++ builds (note: NVCC is already exported including ccache)
+#=== Configure ccache for C++ builds (note: GPUCC is already exported including ccache)
 
 # Enable ccache if USECCACHE=1
 ifeq ($(USECCACHE)$(shell echo $(CXX) | grep ccache),1)
@@ -92,6 +92,13 @@ endif
 ###$(info OMPFLAGS=$(OMPFLAGS))
 CXXFLAGS += $(OMPFLAGS)
 
+# Add correct flags for nvcc (-x cu) and hipcc (-x hip) for GPU code (see #810)
+ifeq ($(findstring nvcc,$(GPUCC)),nvcc)
+  GPUFLAGS += -Xcompiler -fPIC -c -x cu
+else ifeq ($(findstring hipcc,$(GPUCC)),hipcc)
+  GPUFLAGS += -fPIC -c -x hip
+endif
+
 # Set the build flags appropriate to each AVX choice (example: "make AVX=none")
 # [NB MGONGPU_PVW512 is needed because "-mprefer-vector-width=256" is not exposed in a macro]
 # [See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96476]
@@ -253,20 +260,20 @@ $(BUILDDIR)/%.o : %.cc *.h $(BUILDDIR)/.build.$(TAG)
 # Generic target and build rules: objects from CUDA compilation
 $(BUILDDIR)/%_cu.o : %.cc *.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c -x cu $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $< -o $@
 
 #-------------------------------------------------------------------------------
 
 cxx_objects=$(addprefix $(BUILDDIR)/, Parameters_sm.o read_slha.o)
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 cu_objects=$(addprefix $(BUILDDIR)/, Parameters_sm_cu.o)
 endif
 
 # Target (and build rules): common (src) library
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so : $(cxx_objects) $(cu_objects)
 	@if [ ! -d $(LIBDIR) ]; then echo "mkdir -p $(LIBDIR)"; mkdir -p $(LIBDIR); fi
-	$(NVCC) -shared -o $@ $(cxx_objects) $(cu_objects) $(LDFLAGS)
+	$(GPUCC) -shared -o $@ $(cxx_objects) $(cu_objects) $(LDFLAGS)
 else
 $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so : $(cxx_objects)
 	@if [ ! -d $(LIBDIR) ]; then echo "mkdir -p $(LIBDIR)"; mkdir -p $(LIBDIR); fi
diff --git a/epochX/cudacpp/gg_tt01g.mad/src/mgOnGpuConfig.h b/epochX/cudacpp/gg_tt01g.mad/src/mgOnGpuConfig.h
index 80032e528b..6bde4466d0 100644
--- a/epochX/cudacpp/gg_tt01g.mad/src/mgOnGpuConfig.h
+++ b/epochX/cudacpp/gg_tt01g.mad/src/mgOnGpuConfig.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jul 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUCONFIG_H
 #define MGONGPUCONFIG_H 1
@@ -10,12 +10,26 @@
 // There are two different code bases for standalone_cudacpp (without multichannel) and madevent+cudacpp (with multichannel)
 #define MGONGPU_SUPPORTS_MULTICHANNEL 1
 
+// Is this a GPU (CUDA, HIP) or CPU implementation?
+#ifdef __CUDACC__
+#define MGONGPUCPP_GPUIMPL cuda
+#elif defined __HIPCC__
+#define MGONGPUCPP_GPUIMPL hip
+#include "hip/hip_runtime.h" // needed for blockDim, blockIdx, threadIdx: better in mgOnGpuConfig.h than in GpuAbstraction.h
+#else
+#undef MGONGPUCPP_GPUIMPL
+#endif
+
 // ** NB1 Throughputs (e.g. 6.8E8) are events/sec for "./gcheck.exe -p 65536 128 12"
 // ** NB2 Baseline on b7g47n0004 fluctuates (probably depends on load on other VMs)
 
 // Choose if curand is supported for generating random numbers
+// For HIP, by default, do not use curand (common random numbers will be used instead)
 // For both CUDA and C++, by default, do not inline, but allow this macro to be set from outside with e.g. -DMGONGPU_HAS_NO_CURAND
-// (there exist CUDA installations, e.g. using the HPC package, which do not include curand - see PR #784)
+// (there exist CUDA installations, e.g. using the HPC package, which do not include curand - see PR #784 and #785)
+#if defined __HIPCC__
+#define MGONGPU_HAS_NO_CURAND 1
+#else
 //#ifdef __CUDACC__
 //#undef MGONGPU_HAS_NO_CURAND // default
 ////#define MGONGPU_HAS_NO_CURAND 1
@@ -23,6 +37,7 @@
 //#undef MGONGPU_HAS_NO_CURAND // default
 ////#define MGONGPU_HAS_NO_CURAND 1
 //#endif
+#endif
 
 // Choose floating point precision (for everything but color algebra #537)
 // If one of these macros has been set from outside with e.g. -DMGONGPU_FPTYPE_FLOAT, nothing happens (issue #167)
@@ -54,23 +69,31 @@
 //#undef MGONGPU_HARDCODE_PARAM // default
 ////#define MGONGPU_HARDCODE_PARAM 1
 
-// Complex type in c++: std::complex or cxsmpl (CHOOSE ONLY ONE)
-#ifndef __CUDACC__
-//#define MGONGPU_CPPCXTYPE_STDCOMPLEX 1 // ~8 percent slower on float, same on double (5.1E6/double, 9.4E6/float)
-#define MGONGPU_CPPCXTYPE_CXSMPL 1 // new default (5.1E6/double, 10.2E6/float)
-#endif
-
-// Complex type in cuda: thrust or cucomplex or cxsmpl (CHOOSE ONLY ONE)
+// Complex type in CUDA: thrust or cucomplex or cxsmpl (CHOOSE ONLY ONE)
+// (NB THIS IS MGONGPU_*CU*CXTYPE_xxx)
 #ifdef __CUDACC__
 #define MGONGPU_CUCXTYPE_THRUST 1 // default (~1.15E9/double, ~3.2E9/float)
 //#define MGONGPU_CUCXTYPE_CUCOMPLEX 1 // ~10 percent slower (1.03E9/double, ~2.8E9/float)
 //#define MGONGPU_CUCXTYPE_CXSMPL 1 // ~10 percent slower (1.00E9/double, ~2.9E9/float)
+
+// Complex type in HIP: cxsmpl (ONLY ONE OPTION POSSIBLE? #810)
+// (NB THIS IS MGONGPU_*HIP*CXTYPE_xxx)
+#elif defined __HIPCC__
+#define MGONGPU_HIPCXTYPE_CXSMPL 1 // default for HIP
+
+// Complex type in C++: std::complex or cxsmpl (CHOOSE ONLY ONE)
+// (NB THIS IS MGONGPU_*CPP*CXTYPE_xxx)
+#else
+//#define MGONGPU_CPPCXTYPE_STDCOMPLEX 1 // ~8 percent slower on float, same on double (5.1E6/double, 9.4E6/float)
+#define MGONGPU_CPPCXTYPE_CXSMPL 1 // new default (5.1E6/double, 10.2E6/float)
 #endif
 
-// Cuda nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
+// CUDA nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
 #ifdef __CUDACC__
-#undef MGONGPU_NSIGHT_DEBUG // default
+#undef MGONGPU_NSIGHT_DEBUG // default in CUDA
 //#define MGONGPU_NSIGHT_DEBUG 1
+#else
+#undef MGONGPU_NSIGHT_DEBUG // only option in HIP or C++
 #endif
 
 // SANITY CHECKS (floating point precision for everything but color algebra #537)
@@ -86,17 +109,21 @@
 #error You cannot use double precision for color algebra and single precision elsewhere
 #endif
 
-// SANITY CHECKS (c++ complex number implementation)
-#ifndef __CUDACC__
-#if defined MGONGPU_CPPCXTYPE_STDCOMPLEX and defined MGONGPU_CPPCXTYPE_CXSMPL
-#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CPPCXTYPE_STDCOMPLEX or MGONGPU_CPPCXTYPE_CXSMPL
+// SANITY CHECKS (CUDA complex number implementation)
+#ifdef __CUDACC__
+#if defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CUCOMPLEX
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX for CUDA
+#elif defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CXSMPL for CUDA
+#elif defined MGONGPU_CUCXTYPE_CUCOMPLEX and defined MGONGPU_CUCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE OF MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL for CUDA
 #endif
 #endif
 
-// SANITY CHECKS (cuda complex number implementation)
-#ifdef __CUDACC__
-#if defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CUCOMPLEX and defined MGONGPU_CUCXTYPE_CXSMPL
-#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL
+// SANITY CHECKS (C++ complex number implementation)
+#ifndef MGONGPUCPP_GPUIMPL
+#if defined MGONGPU_CPPCXTYPE_STDCOMPLEX and defined MGONGPU_CPPCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CPPCXTYPE_STDCOMPLEX or MGONGPU_CPPCXTYPE_CXSMPL for C++
 #endif
 #endif
 
@@ -134,7 +161,7 @@ namespace mgOnGpu
   // Alignment requirement for using reinterpret_cast with SIMD vectorized code
   // (using reinterpret_cast with non aligned memory may lead to segmentation faults!)
   // Only needed for C++ code but can be enforced also in NVCC builds of C++ code using CUDA>=11.2 and C++17 (#318, #319, #333)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   constexpr int cppAlign = 64; // alignment requirement for SIMD vectorization (64-byte i.e. 512-bit)
 #endif
 
@@ -145,7 +172,7 @@ using mgOnGpu::fptype;
 using mgOnGpu::fptype2;
 
 // C++ SIMD vectorization width (this will be used to set neppV)
-#ifdef __CUDACC__ // CUDA implementation has no SIMD
+#ifdef MGONGPUCPP_GPUIMPL // CUDA and HIP implementations have no SIMD
 #undef MGONGPU_CPPSIMD
 #elif defined __AVX512VL__ && defined MGONGPU_PVW512 // C++ "512z" AVX512 with 512 width (512-bit ie 64-byte): 8 (DOUBLE) or 16 (FLOAT)
 #ifdef MGONGPU_FPTYPE_DOUBLE
@@ -175,9 +202,9 @@ using mgOnGpu::fptype2;
 #undef MGONGPU_CPPSIMD
 #endif
 
-// Cuda nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
+// CUDA nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
 // Arguments (not used so far): text is __FUNCTION__, code is 0 (start) or 1 (end)
-#if defined __CUDACC__ && defined MGONGPU_NSIGHT_DEBUG /* clang-format off */
+#if defined __CUDA__ && defined MGONGPU_NSIGHT_DEBUG /* clang-format off */
 #define mgDebugDeclare() __shared__ float mgDebugCounter[mgOnGpu::ntpbMAX];
 #define mgDebugInitialise() { mgDebugCounter[threadIdx.x] = 0; }
 #define mgDebug( code, text ) { mgDebugCounter[threadIdx.x] += 1; }
@@ -189,8 +216,8 @@ using mgOnGpu::fptype2;
 #define mgDebugFinalise() { /*noop*/ }
 #endif /* clang-format on */
 
-// Define empty CUDA declaration specifiers for C++
-#ifndef __CUDACC__
+// Define empty CUDA/HIP declaration specifiers for C++
+#ifndef MGONGPUCPP_GPUIMPL
 #define __global__
 #define __host__
 #define __device__
diff --git a/epochX/cudacpp/gg_tt01g.mad/src/mgOnGpuCxtypes.h b/epochX/cudacpp/gg_tt01g.mad/src/mgOnGpuCxtypes.h
index ca9a9f00c0..7ede1dbfae 100644
--- a/epochX/cudacpp/gg_tt01g.mad/src/mgOnGpuCxtypes.h
+++ b/epochX/cudacpp/gg_tt01g.mad/src/mgOnGpuCxtypes.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022, based on earlier work by D. Smith) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUCXTYPES_H
 #define MGONGPUCXTYPES_H 1
@@ -19,7 +19,7 @@
 #include <complex>
 
 // Complex type in cuda: thrust or cucomplex or cxsmpl
-#ifdef __CUDACC__
+#ifdef __CUDACC__ // this must be __CUDAC__ (not MGONGPUCPP_GPUIMPL)
 #if defined MGONGPU_CUCXTYPE_THRUST
 #pragma clang diagnostic push
 #pragma clang diagnostic ignored "-Wtautological-compare" // for icpx2021/clang13 (https://stackoverflow.com/a/15864661)
@@ -30,8 +30,13 @@
 #elif not defined MGONGPU_CUCXTYPE_CXSMPL
 #error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL
 #endif
+// Complex type in HIP: cxsmpl
+#elif defined __HIPCC__
+#if not defined MGONGPU_HIPCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_HIPCXTYPE_CXSMPL
+#endif
 #else
-// Complex type in c++: std::complex or cxsmpl
+// Complex type in c++ or HIP: std::complex or cxsmpl
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
 #include <cmath>
 #elif not defined MGONGPU_CPPCXTYPE_CXSMPL
@@ -82,7 +87,7 @@ namespace mgOnGpu /* clang-format off */
 using mgOnGpu::cxsmpl;
 
 // Printout to stream for user defined types
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -92,7 +97,7 @@ namespace mg5amcCpu
   inline __host__ std::ostream&
   operator<<( std::ostream& out, const cxsmpl<FP>& c )
   {
-    out << std::complex( c.real(), c.imag() );
+    out << std::complex<FP>( c.real(), c.imag() );
     return out;
   }
 
@@ -215,14 +220,14 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
   // --- Type definitions (complex type: cxtype)
-#ifdef __CUDACC__ // cuda
+#ifdef __CUDACC__ // this must be __CUDAC__ (not MGONGPUCPP_GPUIMPL)
 #if defined MGONGPU_CUCXTYPE_THRUST
   typedef thrust::complex<fptype> cxtype;
 #elif defined MGONGPU_CUCXTYPE_CUCOMPLEX
@@ -255,13 +260,13 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
-#if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+#if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
 
   //------------------------------
   // CUDA or C++ - using cxsmpl
@@ -303,11 +308,11 @@ namespace mg5amcCpu
     return cxmake( c.real(), c.imag() );
   }
 
-#endif // #if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+#endif // #if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
 
   //==========================================================================
 
-#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_THRUST // cuda + thrust
+#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_THRUST // cuda + thrust (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // CUDA - using thrust::complex
@@ -347,7 +352,7 @@ namespace mg5amcCpu
 
   //==========================================================================
 
-#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_CUCOMPLEX // cuda + cucomplex
+#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_CUCOMPLEX // cuda + cucomplex (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // CUDA - using cuComplex
@@ -566,7 +571,7 @@ namespace mg5amcCpu
 
   //==========================================================================
 
-#if not defined __CUDACC__ and defined MGONGPU_CPPCXTYPE_STDCOMPLEX // c++ + stdcomplex
+#if not defined __CUDACC__ and defined MGONGPU_CPPCXTYPE_STDCOMPLEX // c++/hip + stdcomplex (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // C++ - using std::complex
@@ -633,7 +638,7 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_tt01g.mad/src/mgOnGpuFptypes.h b/epochX/cudacpp/gg_tt01g.mad/src/mgOnGpuFptypes.h
index 905c97d700..fa3a02664b 100644
--- a/epochX/cudacpp/gg_tt01g.mad/src/mgOnGpuFptypes.h
+++ b/epochX/cudacpp/gg_tt01g.mad/src/mgOnGpuFptypes.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUFPTYPES_H
 #define MGONGPUFPTYPES_H 1
@@ -12,7 +12,7 @@
 #include <cmath>
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL // cuda
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -20,7 +20,7 @@ namespace mg5amcCpu
 {
   //==========================================================================
 
-#ifdef __CUDACC__ // cuda
+#ifdef MGONGPUCPP_GPUIMPL // cuda
 
   //------------------------------
   // Floating point types - Cuda
@@ -64,11 +64,11 @@ namespace mg5amcCpu
 #endif
   }
 
-#endif // #ifdef __CUDACC__
+#endif // #ifdef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
   //------------------------------
   // Floating point types - C++
@@ -92,7 +92,7 @@ namespace mg5amcCpu
     return std::sqrt( f );
   }
 
-#endif // #ifndef __CUDACC__
+#endif // #ifndef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
diff --git a/epochX/cudacpp/gg_tt01g.mad/src/mgOnGpuVectors.h b/epochX/cudacpp/gg_tt01g.mad/src/mgOnGpuVectors.h
index e1299ba81e..cdae04326b 100644
--- a/epochX/cudacpp/gg_tt01g.mad/src/mgOnGpuVectors.h
+++ b/epochX/cudacpp/gg_tt01g.mad/src/mgOnGpuVectors.h
@@ -32,7 +32,7 @@
 #endif
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -131,7 +131,7 @@ namespace mg5amcCpu
 #endif
 #endif
 
-#else // i.e #ifndef MGONGPU_CPPSIMD (this includes #ifdef __CUDACC__)
+#else // i.e #ifndef MGONGPU_CPPSIMD (this includes #ifdef MGONGPUCPP_GPUIMPL)
 
   const int neppV = 1;
 
@@ -153,13 +153,13 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
   // Printout to stream for user defined types
 
@@ -805,11 +805,11 @@ namespace mg5amcCpu
 
 #endif // #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
 
-#endif // #ifndef __CUDACC__
+#endif // #ifndef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
   //------------------------------
   // Vector types - CUDA
@@ -853,12 +853,12 @@ namespace mg5amcCpu
     return mask;
   }
 
-#endif // #ifdef __CUDACC__
+#endif // #ifdef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
   // Scalar-or-vector types: scalar in CUDA, vector or scalar in C++
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   typedef bool bool_sv;
   typedef fptype fptype_sv;
   typedef fptype2 fptype2_sv;
@@ -879,7 +879,7 @@ namespace mg5amcCpu
 #endif
 
   // Scalar-or-vector zeros: scalar in CUDA, vector or scalar in C++
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   inline __host__ __device__ cxtype cxzero_sv(){ return cxtype( 0, 0 ); }
 #elif defined MGONGPU_CPPSIMD
   inline cxtype_v cxzero_sv() { return cxtype_v(); } // RRRR=0000 IIII=0000
diff --git a/epochX/cudacpp/gg_tt01g.mad/src/rambo.h b/epochX/cudacpp/gg_tt01g.mad/src/rambo.h
index e02ea52496..cd7e1008ea 100644
--- a/epochX/cudacpp/gg_tt01g.mad/src/rambo.h
+++ b/epochX/cudacpp/gg_tt01g.mad/src/rambo.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -18,7 +18,7 @@
 #include <iostream>
 
 // Simplified rambo version for 2 to N (with N>=2) processes with massless particles
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +83,7 @@ namespace mg5amcCpu
       static bool first = true;
       if( first )
       {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
         if constexpr( M_ACCESS::isOnDevice() ) // avoid
         {
           const int ievt0 = 0;
@@ -166,7 +166,7 @@ namespace mg5amcCpu
     wt = po2log;
     if( nparf != 2 ) wt = ( 2. * nparf - 4. ) * log( energy ) + z[nparf - 1];
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     // issue warnings if weight is too small or too large
     static int iwarn[5] = { 0, 0, 0, 0, 0 };
     if( wt < -180. )
diff --git a/epochX/cudacpp/gg_tt01g.mad/src/read_slha.cc b/epochX/cudacpp/gg_tt01g.mad/src/read_slha.cc
index 055b19a779..f8e46f2e66 100644
--- a/epochX/cudacpp/gg_tt01g.mad/src/read_slha.cc
+++ b/epochX/cudacpp/gg_tt01g.mad/src/read_slha.cc
@@ -11,7 +11,11 @@
 
 #include <algorithm>
 #include <cstdlib>
-#include <filesystem>
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
 #include <fstream>
 #include <iostream>
 
@@ -60,7 +64,14 @@ SLHAReader::read_slha_file( std::string file_name, bool verbose )
     {
       std::cout << "WARNING! Card file '" << file_name << "' does not exist:"
                 << " look for the file in directory $" << envpath << "='" << getenv( envpath ) << "'" << std::endl;
+      /*
+#ifdef __HIPCC__
+      const std::string file_name2 = std::experimental::filesystem::path( getenv( envpath ) ) / std::experimental::filesystem::path( file_name ).filename();
+#else
       const std::string file_name2 = std::filesystem::path( getenv( envpath ) ) / std::filesystem::path( file_name ).filename();
+#endif
+      */
+      const std::string file_name2 = std::string( getenv( envpath ) ) + "/" + file_name; // bypass std::filesystem #803
       param_card.open( file_name2.c_str(), std::ifstream::in );
       if( param_card.good() )
       {
diff --git a/epochX/cudacpp/gg_ttg.mad/CODEGEN_mad_gg_ttg_log.txt b/epochX/cudacpp/gg_ttg.mad/CODEGEN_mad_gg_ttg_log.txt
index 37ba5c7297..ffc3d1d3ef 100644
--- a/epochX/cudacpp/gg_ttg.mad/CODEGEN_mad_gg_ttg_log.txt
+++ b/epochX/cudacpp/gg_ttg.mad/CODEGEN_mad_gg_ttg_log.txt
@@ -52,7 +52,7 @@ Note that you can still compile and run aMC@NLO with the built-in PDFs
 
 Using default text editor "vi". Set another one in ./input/mg5_configuration.txt
 Using default eps viewer "evince". Set another one in ./input/mg5_configuration.txt
-No valid web browser found. Please set in ./input/mg5_configuration.txt
+Using default web browser "firefox". Set another one in ./input/mg5_configuration.txt
 import /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_mad_gg_ttg.mg
 The import format was not given, so we guess it as command
 set stdout_level DEBUG
@@ -62,7 +62,7 @@ generate g g > t t~ g
 No model currently active, so we import the Standard Model
 INFO: load particles 
 INFO: load vertices 
-DEBUG: model prefixing  takes 0.005791187286376953 
+DEBUG: model prefixing  takes 0.005362510681152344 
 INFO: Restrict model sm with file models/sm/restrict_default.dat . 
 DEBUG: Simplifying conditional expressions 
 DEBUG: remove interactions: u s w+ at order: QED=1 
@@ -155,7 +155,7 @@ INFO: Please specify coupling orders to bypass this step.
 INFO: Trying coupling order WEIGHTED<=3: WEIGTHED IS QCD+2*QED 
 INFO: Trying process: g g > t t~ g WEIGHTED<=3 @1  
 INFO: Process has 16 diagrams 
-1 processes with 16 diagrams generated in 0.022 s
+1 processes with 16 diagrams generated in 0.021 s
 Total: 1 processes with 16 diagrams
 output madevent ../TMPOUT/CODEGEN_mad_gg_ttg --hel_recycling=False --vector_size=32 --me_exporter=standalone_cudacpp
 Load PLUGIN.CUDACPP_OUTPUT
@@ -175,7 +175,7 @@ INFO: Generating Helas calls for process: g g > t t~ g WEIGHTED<=3 @1
 INFO: Processing color information for process: g g > t t~ g @1 
 INFO: Creating files in directory P1_gg_ttxg 
 DEBUG:  kwargs[prefix] = 0 [model_handling.py at line 1058] 
-DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7f9d4565e700> [export_v4.py at line 6262] 
+DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7fad80fd26a0> [export_v4.py at line 6262] 
 INFO: Creating files in directory . 
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.h
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.cc
@@ -190,8 +190,8 @@ INFO: Created files CPPProcess.h and CPPProcess.cc in directory ./.
 DEBUG:  vector, subproc_group,self.opt['vector_size'] =  32 True 32 [export_v4.py at line 1872] 
 INFO: Generating Feynman diagrams for Process: g g > t t~ g WEIGHTED<=3 @1 
 INFO: Finding symmetric diagrams for subprocess group gg_ttxg 
-Generated helas calls for 1 subprocesses (16 diagrams) in 0.039 s
-Wrote files for 36 helas calls in 0.153 s
+Generated helas calls for 1 subprocesses (16 diagrams) in 0.037 s
+Wrote files for 36 helas calls in 0.149 s
 ALOHA: aloha starts to compute helicity amplitudes
 ALOHA: aloha creates VVV1 routines
 ALOHA: aloha creates FFV1 routines
@@ -206,7 +206,7 @@ ALOHA: aloha creates FFV1 routines
 ALOHA: aloha creates VVVV1 set of routines with options: P0
 ALOHA: aloha creates VVVV3 set of routines with options: P0
 ALOHA: aloha creates VVVV4 set of routines with options: P0
-ALOHA: aloha creates 10 routines in  0.315 s
+ALOHA: aloha creates 10 routines in  0.308 s
 <class 'aloha.create_aloha.AbstractRoutine'> VVV1
 <class 'aloha.create_aloha.AbstractRoutine'> VVV1
 <class 'aloha.create_aloha.AbstractRoutine'> FFV1
@@ -252,10 +252,10 @@ Type "launch" to generate events from this process, or see
 Run "open index.html" to see more information about this process.
 quit
 
-real	0m2.208s
-user	0m1.988s
-sys	0m0.221s
-Code generation completed in 2 seconds
+real	0m2.177s
+user	0m1.931s
+sys	0m0.233s
+Code generation completed in 3 seconds
 ************************************************************
 *                                                          *
 *                      W E L C O M E to                    *
@@ -281,7 +281,7 @@ INFO: load configuration from /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/mg5amc
 INFO: load configuration from /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_mad_gg_ttg/Cards/me5_configuration.txt  
 Using default text editor "vi". Set another one in ./input/mg5_configuration.txt
 Using default eps viewer "evince". Set another one in ./input/mg5_configuration.txt
-No valid web browser found. Please set in ./input/mg5_configuration.txt
+Using default web browser "firefox". Set another one in ./input/mg5_configuration.txt
 treatcards run
 quit
 INFO:  
@@ -311,7 +311,7 @@ INFO: load configuration from /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/mg5amc
 INFO: load configuration from /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_mad_gg_ttg/Cards/me5_configuration.txt  
 Using default text editor "vi". Set another one in ./input/mg5_configuration.txt
 Using default eps viewer "evince". Set another one in ./input/mg5_configuration.txt
-No valid web browser found. Please set in ./input/mg5_configuration.txt
+Using default web browser "firefox". Set another one in ./input/mg5_configuration.txt
 treatcards param
 quit
 INFO:  
diff --git a/epochX/cudacpp/gg_ttg.mad/COPYRIGHT b/epochX/cudacpp/gg_ttg.mad/COPYRIGHT
index a134b5fef9..9036d9260a 100644
--- a/epochX/cudacpp/gg_ttg.mad/COPYRIGHT
+++ b/epochX/cudacpp/gg_ttg.mad/COPYRIGHT
@@ -15,6 +15,7 @@ The full development team currently includes the following authors :
   Stephan Hageboeck (CERN)
   Olivier Mattelaer (Universite Catholique de Louvain, original author)
   Stefan Roiser (CERN, original author)
+  Jorgen Teig (CERN)
   Andrea Valassi (CERN, original author)
   Zenny Wettersten (CERN)
 See https://github.com/madgraph5/madgraph4gpu for more details. For the full
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/Bridge.h b/epochX/cudacpp/gg_ttg.mad/SubProcesses/Bridge.h
index bf8b5e024d..f9ed70dfde 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/Bridge.h
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/Bridge.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Roiser (Nov 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef BRIDGE_H
 #define BRIDGE_H 1
@@ -14,16 +14,23 @@
 #include "MemoryAccessMomenta.h"  // for MemoryAccessMomenta::neppM
 #include "MemoryBuffers.h"        // for HostBufferMomenta, DeviceBufferMomenta etc
 
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
+
+#include <sys/stat.h> // bypass std::filesystem #803
+
 #include <algorithm>
 #include <cassert>
 #include <cmath>
 #include <cstring>
-#include <filesystem>
 #include <iostream>
 #include <memory>
 #include <type_traits>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +90,7 @@ namespace mg5amcCpu
     Bridge& operator=( const Bridge& ) = delete;
     Bridge& operator=( Bridge&& ) = delete;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     /**
      * Set the gpublocks and gputhreads for the gpusequence - throws if evnt != gpublocks*gputhreads
      * (this is needed for BridgeKernel tests rather than for actual production use in Fortran)
@@ -150,7 +157,7 @@ namespace mg5amcCpu
     unsigned int m_nevt; // number of events
     int m_nGoodHel;      // the number of good helicities (-1 initially when they have not yet been calculated)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     int m_gputhreads; // number of gpu threads (default set from number of events, can be modified)
     int m_gpublocks;  // number of gpu blocks (default set from number of events, can be modified)
     DeviceBuffer<FORTRANFPTYPE, sizePerEventMomenta> m_devMomentaF;
@@ -187,12 +194,12 @@ namespace mg5amcCpu
   // Forward declare transposition methods
   //
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
   template<typename Tin, typename Tout>
   __global__ void dev_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt );
 
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
 
   template<typename Tin, typename Tout>
   void hst_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt );
@@ -209,7 +216,7 @@ namespace mg5amcCpu
   Bridge<FORTRANFPTYPE>::Bridge( unsigned int nevtF, unsigned int nparF, unsigned int np4F )
     : m_nevt( nevtF )
     , m_nGoodHel( -1 )
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     , m_gputhreads( 256 )                  // default number of gpu threads
     , m_gpublocks( m_nevt / m_gputhreads ) // this ensures m_nevt <= m_gpublocks*m_gputhreads
     , m_devMomentaF( m_nevt )
@@ -233,7 +240,7 @@ namespace mg5amcCpu
   {
     if( nparF != CPPProcess::npar ) throw std::runtime_error( "Bridge constructor: npar mismatch" );
     if( np4F != CPPProcess::np4 ) throw std::runtime_error( "Bridge constructor: np4 mismatch" );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( ( m_nevt < s_gputhreadsmin ) || ( m_nevt % s_gputhreadsmin != 0 ) )
       throw std::runtime_error( "Bridge constructor: nevt should be a multiple of " + std::to_string( s_gputhreadsmin ) );
     while( m_nevt != m_gpublocks * m_gputhreads )
@@ -249,20 +256,28 @@ namespace mg5amcCpu
 #else
     std::cout << "WARNING! Instantiate host Bridge (nevt=" << m_nevt << ")" << std::endl;
     m_pmek.reset( new MatrixElementKernelHost( m_hstMomentaC, m_hstGs, m_hstRndHel, m_hstRndCol, m_hstMEs, m_hstSelHel, m_hstSelCol, m_nevt ) );
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
     // Create a process object, read param card and set parameters
     // FIXME: the process instance can happily go out of scope because it is only needed to read parameters?
     // FIXME: the CPPProcess should really be a singleton? what if fbridgecreate is called from several Fortran threads?
     CPPProcess process( /*verbose=*/false );
     std::string paramCard = "../../Cards/param_card.dat";
-    if( !std::filesystem::exists( paramCard ) )
-    {
-      paramCard = "../" + paramCard;
-    }
+    /*
+#ifdef __HIPCC__
+    if( !std::experimental::filesystem::exists( paramCard ) ) paramCard = "../" + paramCard;
+#else
+    if( !std::filesystem::exists( paramCard ) ) paramCard = "../" + paramCard;
+#endif
+    */
+    //struct stat dummybuffer; // bypass std::filesystem #803
+    //if( !( stat( paramCard.c_str(), &dummyBuffer ) == 0 ) ) paramCard = "../" + paramCard; //
+    auto fileExists = []( std::string& fileName )
+    { struct stat buffer; return stat( fileName.c_str(), &buffer ) == 0; };
+    if( !fileExists( paramCard ) ) paramCard = "../" + paramCard; // bypass std::filesystem #803
     process.initProc( paramCard );
   }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::set_gpugrid( const int gpublocks, const int gputhreads )
   {
@@ -276,7 +291,7 @@ namespace mg5amcCpu
   }
 #endif
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::gpu_sequence( const FORTRANFPTYPE* momenta,
                                             const FORTRANFPTYPE* gs,
@@ -291,14 +306,14 @@ namespace mg5amcCpu
     constexpr int neppM = MemoryAccessMomenta::neppM;
     if constexpr( neppM == 1 && std::is_same_v<FORTRANFPTYPE, fptype> )
     {
-      checkCuda( cudaMemcpy( m_devMomentaC.data(), momenta, m_devMomentaC.bytes(), cudaMemcpyHostToDevice ) );
+      gpuMemcpy( m_devMomentaC.data(), momenta, m_devMomentaC.bytes(), gpuMemcpyHostToDevice );
     }
     else
     {
-      checkCuda( cudaMemcpy( m_devMomentaF.data(), momenta, m_devMomentaF.bytes(), cudaMemcpyHostToDevice ) );
+      gpuMemcpy( m_devMomentaF.data(), momenta, m_devMomentaF.bytes(), gpuMemcpyHostToDevice );
       const int thrPerEvt = CPPProcess::npar * CPPProcess::np4; // AV: transpose alg does 1 element per thread (NOT 1 event per thread)
       //const int thrPerEvt = 1; // AV: try new alg with 1 event per thread... this seems slower
-      dev_transposeMomentaF2C<<<m_gpublocks * thrPerEvt, m_gputhreads>>>( m_devMomentaF.data(), m_devMomentaC.data(), m_nevt );
+      gpuLaunchKernel( dev_transposeMomentaF2C, m_gpublocks * thrPerEvt, m_gputhreads, m_devMomentaF.data(), m_devMomentaC.data(), m_nevt );
     }
     if constexpr( std::is_same_v<FORTRANFPTYPE, fptype> )
     {
@@ -341,7 +356,7 @@ namespace mg5amcCpu
   }
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::cpu_sequence( const FORTRANFPTYPE* momenta,
                                             const FORTRANFPTYPE* gs,
@@ -396,7 +411,7 @@ namespace mg5amcCpu
   // - C++ array: momenta[npagM][npar][np4][neppM] with nevt=npagM*neppM (AOSOA)
   //
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename Tin, typename Tout>
   __global__ void dev_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt )
   {
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/BridgeKernels.cc b/epochX/cudacpp/gg_ttg.mad/SubProcesses/BridgeKernels.cc
index d58066c9c1..eaf4037a24 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/BridgeKernels.cc
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/BridgeKernels.cc
@@ -1,17 +1,18 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "BridgeKernels.h"
 
+#include "GpuAbstraction.h"
 #include "MemoryAccessMomenta.h"
 
 #include <sstream>
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -45,7 +46,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu
 {
 
@@ -96,7 +97,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/BridgeKernels.h b/epochX/cudacpp/gg_ttg.mad/SubProcesses/BridgeKernels.h
index 15eb4bff4d..3efef8ce97 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/BridgeKernels.h
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/BridgeKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef BRIDGEKERNELS_H
 #define BRIDGEKERNELS_H 1
@@ -12,7 +12,7 @@
 #include "MatrixElementKernels.h"
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -49,7 +49,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A Bridge wrapper class encapsulating matrix element calculations on a CPU host
   class BridgeKernelHost final : public BridgeKernelBase
   {
@@ -89,7 +89,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A Bridge wrapper class encapsulating matrix element calculations on a GPU device
   class BridgeKernelDevice : public BridgeKernelBase
   {
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/CommonRandomNumberKernel.cc b/epochX/cudacpp/gg_ttg.mad/SubProcesses/CommonRandomNumberKernel.cc
index 985b39f576..010bc4cbd0 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/CommonRandomNumberKernel.cc
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/CommonRandomNumberKernel.cc
@@ -1,15 +1,16 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "CommonRandomNumbers.h"
+#include "GpuAbstraction.h"
 #include "MemoryBuffers.h"
 #include "RandomNumberKernels.h"
 
 #include <cassert>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/CrossSectionKernels.cc b/epochX/cudacpp/gg_ttg.mad/SubProcesses/CrossSectionKernels.cc
index 0b355a3c8d..c15b39844d 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/CrossSectionKernels.cc
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/CrossSectionKernels.cc
@@ -1,10 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "CrossSectionKernels.h"
 
+#include "GpuAbstraction.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessWeights.h"
 #include "MemoryBuffers.h"
@@ -77,7 +78,7 @@ debug_me_is_abnormal( const fptype& me, size_t ievtALL )
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -185,7 +186,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/CrossSectionKernels.h b/epochX/cudacpp/gg_ttg.mad/SubProcesses/CrossSectionKernels.h
index 7933ca4bbf..4d9659e04e 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/CrossSectionKernels.h
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/CrossSectionKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef CROSSSECTIONKERNELS_H
 #define CROSSSECTIONKERNELS_H 1
@@ -13,7 +13,7 @@
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -96,7 +96,7 @@ namespace mg5amcCpu
   //--------------------------------------------------------------------------
 
   /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating the calculation of event statistics on a GPU device
   class CrossSectionKernelDevice : public CrossSectionKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/CudaRuntime.h b/epochX/cudacpp/gg_ttg.mad/SubProcesses/CudaRuntime.h
deleted file mode 100644
index 64ce52f4b3..0000000000
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/CudaRuntime.h
+++ /dev/null
@@ -1,85 +0,0 @@
-// Copyright (C) 2020-2023 CERN and UCLouvain.
-// Licensed under the GNU Lesser General Public License (version 3 or later).
-// Created by: S. Roiser (Jul 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
-
-#ifndef MG5AMC_CUDARUNTIME_H
-#define MG5AMC_CUDARUNTIME_H 1
-
-// MG5AMC on GPU uses the CUDA runtime API, not the lower level CUDA driver API
-// See https://docs.nvidia.com/cuda/cuda-runtime-api/driver-vs-runtime-api.html#driver-vs-runtime-api
-
-#include <cassert>
-#include <iostream>
-
-//--------------------------------------------------------------------------
-
-// See https://stackoverflow.com/a/14038590
-#ifdef __CUDACC__ /* clang-format off */
-#define checkCuda( code ) { assertCuda( code, __FILE__, __LINE__ ); }
-inline void assertCuda( cudaError_t code, const char* file, int line, bool abort = true )
-{
-  if( code != cudaSuccess )
-  {
-    printf( "ERROR! assertCuda: '%s' (%d) in %s:%d\n", cudaGetErrorString( code ), code, file, line );
-    if( abort ) assert( code == cudaSuccess );
-  }
-}
-#endif /* clang-format on */
-
-//--------------------------------------------------------------------------
-
-#ifdef __CUDACC__
-namespace mg5amcGpu
-{
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  // *** FIXME! This will all need to be designed differently when going to multi-GPU nodes! ***
-  struct CudaRuntime final
-  {
-    CudaRuntime( const bool debug = true )
-      : m_debug( debug ) { setUp( m_debug ); }
-    ~CudaRuntime() { tearDown( m_debug ); }
-    CudaRuntime( const CudaRuntime& ) = delete;
-    CudaRuntime( CudaRuntime&& ) = delete;
-    CudaRuntime& operator=( const CudaRuntime& ) = delete;
-    CudaRuntime& operator=( CudaRuntime&& ) = delete;
-    bool m_debug;
-
-    // Set up CUDA application
-    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
-    // Calling cudaSetDevice on startup is useful to properly book-keep the time spent in CUDA initialization
-    static void setUp( const bool debug = true )
-    {
-      // ** NB: it is useful to call cudaSetDevice, or cudaFree, to properly book-keep the time spent in CUDA initialization
-      // ** NB: otherwise, the first CUDA operation (eg a cudaMemcpyToSymbol in CPPProcess ctor) appears to take much longer!
-      /*
-      // [We initially added cudaFree(0) to "ease profile analysis" only because it shows up as a big recognizable block!]
-      // No explicit initialization is needed: https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#initialization
-      // It is not clear what cudaFree(0) does at all: https://stackoverflow.com/questions/69967813/
-      if ( debug ) std::cout << "__CudaRuntime: calling cudaFree(0)" << std::endl;
-      checkCuda( cudaFree( 0 ) ); // SLOW!
-      */
-      // Replace cudaFree(0) by cudaSetDevice(0), even if it is not really needed either
-      // (but see https://developer.nvidia.com/blog/cuda-pro-tip-always-set-current-device-avoid-multithreading-bugs)
-      if( debug ) std::cout << "__CudaRuntime: calling cudaSetDevice(0)" << std::endl;
-      checkCuda( cudaSetDevice( 0 ) ); // SLOW!
-    }
-
-    // Tear down CUDA application (call cudaDeviceReset)
-    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
-    // Calling cudaDeviceReset on shutdown is only needed for checking memory leaks in cuda-memcheck
-    // See https://docs.nvidia.com/cuda/cuda-memcheck/index.html#leak-checking
-    static void tearDown( const bool debug = true )
-    {
-      if( debug ) std::cout << "__CudaRuntime: calling cudaDeviceReset()" << std::endl;
-      checkCuda( cudaDeviceReset() );
-    }
-  };
-
-}
-#endif
-
-//--------------------------------------------------------------------------
-
-#endif // MG5AMC_CUDARUNTIME_H
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/CurandRandomNumberKernel.cc b/epochX/cudacpp/gg_ttg.mad/SubProcesses/CurandRandomNumberKernel.cc
index eb56333b03..08a16f6f2c 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/CurandRandomNumberKernel.cc
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/CurandRandomNumberKernel.cc
@@ -1,9 +1,9 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "MemoryBuffers.h"
 #include "RandomNumberKernels.h"
 
@@ -22,7 +22,7 @@ inline void assertCurand( curandStatus_t code, const char *file, int line, bool
 }
 #endif /* clang-format on */
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -36,7 +36,7 @@ namespace mg5amcCpu
   {
     if( m_isOnDevice )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if( !m_rnarray.isOnDevice() )
         throw std::runtime_error( "CurandRandomNumberKernel on device with a host random number array" );
 #else
@@ -114,7 +114,7 @@ namespace mg5amcCpu
     /*
     printf( "\nCurandRandomNumberKernel::generateRnarray size = %d\n", (int)m_rnarray.size() );
     fptype* data = m_rnarray.data();
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( m_rnarray.isOnDevice() )
     {
       data = new fptype[m_rnarray.size()]();
@@ -123,7 +123,7 @@ namespace mg5amcCpu
 #endif
     for( int i = 0; i < ( (int)m_rnarray.size() / 4 ); i++ )
       printf( "[%4d] %f %f %f %f\n", i * 4, data[i * 4], data[i * 4 + 2], data[i * 4 + 2], data[i * 4 + 3] );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( m_rnarray.isOnDevice() ) delete[] data;
 #endif
     */
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/EventStatistics.h b/epochX/cudacpp/gg_ttg.mad/SubProcesses/EventStatistics.h
index 48b51e0a49..b425a5bade 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/EventStatistics.h
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/EventStatistics.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef EventStatistics_H
 #define EventStatistics_H 1
@@ -16,7 +16,7 @@
 #include <limits>
 #include <string>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/GpuAbstraction.h b/epochX/cudacpp/gg_ttg.mad/SubProcesses/GpuAbstraction.h
new file mode 100644
index 0000000000..9c467b1e04
--- /dev/null
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/GpuAbstraction.h
@@ -0,0 +1,69 @@
+// Copyright (C) 2020-2023 CERN and UCLouvain.
+// Licensed under the GNU Lesser General Public License (version 3 or later).
+// Created by: J. Teig (Jul 2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+
+#ifndef MG5AMC_GPUABSTRACTION_H
+#define MG5AMC_GPUABSTRACTION_H 1
+
+#include <cassert>
+
+//--------------------------------------------------------------------------
+
+#ifdef __CUDACC__
+
+#define gpuError_t cudaError_t
+#define gpuPeekAtLastError cudaPeekAtLastError
+#define gpuGetErrorString cudaGetErrorString
+#define gpuSuccess cudaSuccess
+
+#define gpuMallocHost( ptr, size ) checkGpu( cudaMallocHost( ptr, size ) )
+#define gpuMalloc( ptr, size ) checkGpu( cudaMalloc( ptr, size ) )
+
+#define gpuMemcpy( dstData, srcData, srcBytes, func ) checkGpu( cudaMemcpy( dstData, srcData, srcBytes, func ) )
+#define gpuMemcpyHostToDevice cudaMemcpyHostToDevice
+#define gpuMemcpyDeviceToHost cudaMemcpyDeviceToHost
+#define gpuMemcpyToSymbol( type1, type2, size ) checkGpu( cudaMemcpyToSymbol( type1, type2, size ) )
+
+#define gpuFree( ptr ) checkGpu( cudaFree( ptr ) )
+#define gpuFreeHost( ptr ) checkGpu( cudaFreeHost( ptr ) )
+
+#define gpuSetDevice cudaSetDevice
+#define gpuDeviceSynchronize cudaDeviceSynchronize
+#define gpuDeviceReset cudaDeviceReset
+
+#define gpuLaunchKernel( kernel, blocks, threads, ... ) kernel<<<blocks, threads>>>( __VA_ARGS__ )
+#define gpuLaunchKernelSharedMem( kernel, blocks, threads, sharedMem, ... ) kernel<<<blocks, threads, sharedMem>>>( __VA_ARGS__ )
+
+//--------------------------------------------------------------------------
+
+#elif defined __HIPCC__
+
+#define gpuError_t hipError_t
+#define gpuPeekAtLastError hipPeekAtLastError
+#define gpuGetErrorString hipGetErrorString
+#define gpuSuccess hipSuccess
+
+#define gpuMallocHost( ptr, size ) checkGpu( hipHostMalloc( ptr, size ) ) // HostMalloc better
+#define gpuMalloc( ptr, size ) checkGpu( hipMalloc( ptr, size ) )
+
+#define gpuMemcpy( dstData, srcData, srcBytes, func ) checkGpu( hipMemcpy( dstData, srcData, srcBytes, func ) )
+#define gpuMemcpyHostToDevice hipMemcpyHostToDevice
+#define gpuMemcpyDeviceToHost hipMemcpyDeviceToHost
+#define gpuMemcpyToSymbol( type1, type2, size ) checkGpu( hipMemcpyToSymbol( type1, type2, size ) )
+
+#define gpuFree( ptr ) checkGpu( hipFree( ptr ) )
+#define gpuFreeHost( ptr ) checkGpu( hipHostFree( ptr ) )
+
+#define gpuSetDevice hipSetDevice
+#define gpuDeviceSynchronize hipDeviceSynchronize
+#define gpuDeviceReset hipDeviceReset
+
+#define gpuLaunchKernel( kernel, blocks, threads, ... ) kernel<<<blocks, threads>>>( __VA_ARGS__ )
+#define gpuLaunchKernelSharedMem( kernel, blocks, threads, sharedMem, ... ) kernel<<<blocks, threads, sharedMem>>>( __VA_ARGS__ )
+
+//--------------------------------------------------------------------------
+
+#endif
+
+#endif // MG5AMC_GPUABSTRACTION_H
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/GpuRuntime.h b/epochX/cudacpp/gg_ttg.mad/SubProcesses/GpuRuntime.h
new file mode 100644
index 0000000000..93579ef08b
--- /dev/null
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/GpuRuntime.h
@@ -0,0 +1,85 @@
+// Copyright (C) 2020-2023 CERN and UCLouvain.
+// Licensed under the GNU Lesser General Public License (version 3 or later).
+// Created by: J. Teig (Jun 2023, based on earlier work by S. Roiser) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+
+#ifndef MG5AMC_GPURUNTIME_H
+#define MG5AMC_GPURUNTIME_H 1
+
+// MG5AMC on GPU uses the CUDA runtime API, not the lower level CUDA driver API
+// See https://docs.nvidia.com/cuda/cuda-runtime-api/driver-vs-runtime-api.html#driver-vs-runtime-api
+
+#include "GpuAbstraction.h"
+
+#include <iostream>
+
+//--------------------------------------------------------------------------
+
+// See https://stackoverflow.com/a/14038590
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
+#define checkGpu( code ) { assertGpu( code, __FILE__, __LINE__ ); }
+inline void assertGpu( gpuError_t code, const char* file, int line, bool abort = true )
+{
+  if( code != gpuSuccess )
+  {
+    printf( "ERROR! assertGpu: '%s' (%d) in %s:%d\n", gpuGetErrorString( code ), code, file, line );
+    if( abort ) assert( code == gpuSuccess );
+  }
+}
+#endif /* clang-format on */
+
+//--------------------------------------------------------------------------
+
+#ifdef MGONGPUCPP_GPUIMPL
+namespace mg5amcGpu
+{
+  // Instantiate a GpuRuntime at the beginnining of the application's main to
+  // invoke gpuSetDevice(0) in the constructor and book a gpuDeviceReset() call in the destructor
+  // *** FIXME! This will all need to be designed differently when going to multi-GPU nodes! ***
+  struct GpuRuntime final
+  {
+    GpuRuntime( const bool debug = true )
+      : m_debug( debug ) { setUp( m_debug ); }
+    ~GpuRuntime() { tearDown( m_debug ); }
+    GpuRuntime( const GpuRuntime& ) = delete;
+    GpuRuntime( GpuRuntime&& ) = delete;
+    GpuRuntime& operator=( const GpuRuntime& ) = delete;
+    GpuRuntime& operator=( GpuRuntime&& ) = delete;
+    bool m_debug;
+
+    // Set up CUDA application
+    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
+    // Calling cudaSetDevice on startup is useful to properly book-keep the time spent in CUDA initialization
+    static void setUp( const bool debug = true )
+    {
+      // ** NB: it is useful to call cudaSetDevice, or cudaFree, to properly book-keep the time spent in CUDA initialization
+      // ** NB: otherwise, the first CUDA operation (eg a cudaMemcpyToSymbol in CPPProcess ctor) appears to take much longer!
+      /*
+      // [We initially added cudaFree(0) to "ease profile analysis" only because it shows up as a big recognizable block!]
+      // No explicit initialization is needed: https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#initialization
+      // It is not clear what cudaFree(0) does at all: https://stackoverflow.com/questions/69967813/
+      if ( debug ) std::cout << "__CudaRuntime: calling cudaFree(0)" << std::endl;
+      checkCuda( cudaFree( 0 ) ); // SLOW!
+      */
+      // Replace cudaFree(0) by cudaSetDevice(0), even if it is not really needed either
+      // (but see https://developer.nvidia.com/blog/cuda-pro-tip-always-set-current-device-avoid-multithreading-bugs)
+      if( debug ) std::cout << "__GpuRuntime: calling GpuSetDevice(0)" << std::endl;
+      checkGpu( gpuSetDevice( 0 ) ); // SLOW!
+    }
+
+    // Tear down CUDA application (call cudaDeviceReset)
+    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
+    // Calling cudaDeviceReset on shutdown is only needed for checking memory leaks in cuda-memcheck
+    // See https://docs.nvidia.com/cuda/cuda-memcheck/index.html#leak-checking
+    static void tearDown( const bool debug = true )
+    {
+      if( debug ) std::cout << "__GpuRuntime: calling GpuDeviceReset()" << std::endl;
+      checkGpu( gpuDeviceReset() );
+    }
+  };
+}
+#endif
+
+//--------------------------------------------------------------------------
+
+#endif // MG5AMC_GPURUNTIME_H
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/MadgraphTest.h b/epochX/cudacpp/gg_ttg.mad/SubProcesses/MadgraphTest.h
index ef40624c88..6054185300 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/MadgraphTest.h
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/MadgraphTest.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Hageboeck (Dec 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MADGRAPHTEST_H_
 #define MADGRAPHTEST_H_ 1
@@ -14,7 +14,11 @@
 
 #include <array>
 #include <cmath>
-#include <filesystem>
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
 #include <fstream>
 #include <iomanip>
 #include <memory>
@@ -22,7 +26,7 @@
 #include <string>
 #include <vector>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using mg5amcGpu::CPPProcess;
 #else
 using mg5amcCpu::CPPProcess;
@@ -201,7 +205,7 @@ class MadgraphTest : public testing::TestWithParam<TestDriverBase*>
 
 // Since we link both the CPU-only and GPU tests into the same executable, we prevent
 // a multiply defined symbol by only compiling this in the non-CUDA phase:
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
 /// Compare momenta and matrix elements.
 /// This uses an implementation of TestDriverBase to run a madgraph workflow,
@@ -219,7 +223,14 @@ TEST_P( MadgraphTest, CompareMomentaAndME )
   const char* dumpEventsC = getenv( "CUDACPP_RUNTEST_DUMPEVENTS" );
   const bool dumpEvents = ( dumpEventsC != 0 ) && ( std::string( dumpEventsC ) != "" );
   const std::string refFileName = testDriver->getRefFileName();
+  /*
+#ifdef __HIPCC__
+  const std::string dumpFileName = std::experimental::filesystem::path( refFileName ).filename();
+#else
   const std::string dumpFileName = std::filesystem::path( refFileName ).filename();
+#endif
+  */
+  const std::string dumpFileName = refFileName; // bypass std::filesystem #803
   std::ofstream dumpFile;
   if( dumpEvents )
   {
@@ -307,6 +318,6 @@ TEST_P( MadgraphTest, CompareMomentaAndME )
   }
 }
 
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
 
 #endif /* MADGRAPHTEST_H_ */
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/MatrixElementKernels.cc b/epochX/cudacpp/gg_ttg.mad/SubProcesses/MatrixElementKernels.cc
index 74b5239ebf..81699dfea9 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/MatrixElementKernels.cc
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/MatrixElementKernels.cc
@@ -1,12 +1,12 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "MatrixElementKernels.h"
 
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h" // Includes the abstraction for Nvidia/AMD compilation
 #include "MemoryAccessMomenta.h"
 #include "MemoryBuffers.h"
 
@@ -14,7 +14,7 @@
 
 //============================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu
 {
 
@@ -150,7 +150,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
@@ -209,13 +209,13 @@ namespace mg5amcGpu
     PinnedHostBufferHelicityMask hstIsGoodHel( ncomb );
     DeviceBufferHelicityMask devIsGoodHel( ncomb );
     // ... 0d1. Compute good helicity mask on the device
-    computeDependentCouplings<<<m_gpublocks, m_gputhreads>>>( m_gs.data(), m_couplings.data() );
+    gpuLaunchKernel( computeDependentCouplings, m_gpublocks, m_gputhreads, m_gs.data(), m_couplings.data() );
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-    sigmaKin_getGoodHel<<<m_gpublocks, m_gputhreads>>>( m_momenta.data(), m_couplings.data(), m_matrixElements.data(), m_numerators.data(), m_denominators.data(), devIsGoodHel.data() );
+    gpuLaunchKernel( sigmaKin_getGoodHel, m_gpublocks, m_gputhreads, m_momenta.data(), m_couplings.data(), m_matrixElements.data(), m_numerators.data(), m_denominators.data(), devIsGoodHel.data() );
 #else
-    sigmaKin_getGoodHel<<<m_gpublocks, m_gputhreads>>>( m_momenta.data(), m_couplings.data(), m_matrixElements.data(), devIsGoodHel.data() );
+    gpuLaunchKernel( sigmaKin_getGoodHel, m_gpublocks, m_gputhreads, m_momenta.data(), m_couplings.data(), m_matrixElements.data(), devIsGoodHel.data() );
 #endif
-    checkCuda( cudaPeekAtLastError() );
+    checkGpu( gpuPeekAtLastError() );
     // ... 0d2. Copy back good helicity mask to the host
     copyHostFromDevice( hstIsGoodHel, devIsGoodHel );
     // ... 0d3. Copy back good helicity list to constant memory on the device
@@ -226,19 +226,19 @@ namespace mg5amcGpu
 
   void MatrixElementKernelDevice::computeMatrixElements( const unsigned int channelId )
   {
-    computeDependentCouplings<<<m_gpublocks, m_gputhreads>>>( m_gs.data(), m_couplings.data() );
+    gpuLaunchKernel( computeDependentCouplings, m_gpublocks, m_gputhreads, m_gs.data(), m_couplings.data() );
 #ifndef MGONGPU_NSIGHT_DEBUG
     constexpr unsigned int sharedMemSize = 0;
 #else
     constexpr unsigned int sharedMemSize = ntpbMAX * sizeof( float );
 #endif
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-    sigmaKin<<<m_gpublocks, m_gputhreads, sharedMemSize>>>( m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), channelId, m_numerators.data(), m_denominators.data(), m_selhel.data(), m_selcol.data() );
+    gpuLaunchKernelSharedMem( sigmaKin, m_gpublocks, m_gputhreads, sharedMemSize, m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), channelId, m_numerators.data(), m_denominators.data(), m_selhel.data(), m_selcol.data() );
 #else
-    sigmaKin<<<m_gpublocks, m_gputhreads, sharedMemSize>>>( m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), m_selhel.data(), m_selcol.data() );
+    gpuLaunchKernelSharedMem( sigmaKin, m_gpublocks, m_gputhreads, sharedMemSize, m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), m_selhel.data(), m_selcol.data() );
 #endif
-    checkCuda( cudaPeekAtLastError() );
-    checkCuda( cudaDeviceSynchronize() );
+    checkGpu( gpuPeekAtLastError() );
+    checkGpu( gpuDeviceSynchronize() );
   }
 
   //--------------------------------------------------------------------------
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/MatrixElementKernels.h b/epochX/cudacpp/gg_ttg.mad/SubProcesses/MatrixElementKernels.h
index 23e84757a2..72bd8f195b 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/MatrixElementKernels.h
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/MatrixElementKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MATRIXELEMENTKERNELS_H
 #define MATRIXELEMENTKERNELS_H 1
@@ -10,7 +10,7 @@
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -81,7 +81,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating matrix element calculations on a CPU host
   class MatrixElementKernelHost final : public MatrixElementKernelBase, public NumberOfEvents
   {
@@ -130,7 +130,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating matrix element calculations on a GPU device
   class MatrixElementKernelDevice : public MatrixElementKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessAmplitudes.h b/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessAmplitudes.h
index 573b3bbbc9..ffb76e93de 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessAmplitudes.h
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessAmplitudes.h
@@ -15,7 +15,7 @@
 #define MGONGPU_TRIVIAL_AMPLITUDES 1
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessCouplings.h b/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessCouplings.h
index 35a3af42e0..3afdf3e554 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessCouplings.h
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessCouplings.h
@@ -15,7 +15,7 @@
 #include "MemoryBuffers.h"       // for HostBufferCouplings::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessCouplingsFixed.h b/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessCouplingsFixed.h
index dc0d93afff..ffcdf4dbef 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessCouplingsFixed.h
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessCouplingsFixed.h
@@ -14,7 +14,7 @@
 //#include "MemoryAccessHelpers.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessDenominators.h b/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessDenominators.h
index 3bce635718..66f2d32a6b 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessDenominators.h
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessDenominators.h
@@ -10,7 +10,7 @@
 #include "MemoryAccessGs.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessGs.h b/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessGs.h
index 31311aa375..4c726b30f3 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessGs.h
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessGs.h
@@ -13,7 +13,7 @@
 #include "MemoryBuffers.h" // for HostBufferMatrixElements::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessHelpers.h b/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessHelpers.h
index c82a6c7635..db73e4e064 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessHelpers.h
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessHelpers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessHelpers_H
 #define MemoryAccessHelpers_H 1
@@ -105,7 +105,7 @@ class KernelAccessHelper : public MemoryAccessHelper<T>
     }
     else
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
       //printf( "kernelAccessRecord: ievt=%d threadId=%d\n", ievt, threadIdx.x );
       return T::ieventAccessRecord( buffer, ievt ); // NB fptype and fptype_sv coincide for CUDA
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessMatrixElements.h b/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessMatrixElements.h
index f32e6fea5b..3741011971 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessMatrixElements.h
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessMatrixElements.h
@@ -13,7 +13,7 @@
 #include "MemoryBuffers.h" // for HostBufferMatrixElements::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessMomenta.h b/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessMomenta.h
index 29266de32c..3be229d392 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessMomenta.h
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessMomenta.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessMomenta_H
 #define MemoryAccessMomenta_H 1
@@ -13,7 +13,7 @@
 #include "MemoryAccessVectors.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -30,7 +30,7 @@ namespace mg5amcCpu
 
     // Number of Events Per Page in the momenta AOSOA memory buffer layout
     // (these are all best kept as a compile-time constants: see issue #23)
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
     // -----------------------------------------------------------------------------------------------
     // --- GPUs: neppM is best set to a power of 2 times the number of fptype's in a 32-byte cacheline
     // --- This is relevant to ensure coalesced access to momenta in global memory
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessNumerators.h b/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessNumerators.h
index b152183b28..18991f4fa6 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessNumerators.h
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessNumerators.h
@@ -10,7 +10,7 @@
 #include "MemoryAccessGs.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessRandomNumbers.h b/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessRandomNumbers.h
index e2988d39f3..40cb089135 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessRandomNumbers.h
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessRandomNumbers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessRandomNumbers_H
 #define MemoryAccessRandomNumbers_H 1
@@ -11,7 +11,7 @@
 #include "CPPProcess.h"
 #include "MemoryAccessHelpers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using mg5amcGpu::CPPProcess;
 #else
 using mg5amcCpu::CPPProcess;
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessVectors.h b/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessVectors.h
index e9b197368e..08faccff0f 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessVectors.h
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessVectors.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessVectors_H
 #define MemoryAccessVectors_H 1
@@ -10,7 +10,7 @@
 
 #include "mgOnGpuVectors.h"
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu // this is only needed for CPU SIMD vectorization
 {
 
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessWavefunctions.h b/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessWavefunctions.h
index 5428aaf933..33bef4559e 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessWavefunctions.h
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryAccessWavefunctions.h
@@ -15,7 +15,7 @@
 #define MGONGPU_TRIVIAL_WAVEFUNCTIONS 1
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryBuffers.h b/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryBuffers.h
index 3093e6ed18..7756a71621 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryBuffers.h
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/MemoryBuffers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021, based on earlier work by S. Hageboeck) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryBuffers_H
 #define MemoryBuffers_H 1
@@ -11,12 +11,12 @@
 #include "mgOnGpuCxtypes.h"
 
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "Parameters_sm.h"
 
 #include <sstream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -87,7 +87,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   constexpr bool HostBufferALIGNED = false;   // ismisaligned=false
   constexpr bool HostBufferMISALIGNED = true; // ismisaligned=true
 
@@ -119,7 +119,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA pinned host buffer
   template<typename T>
   class PinnedHostBufferBase : public BufferBase<T>
@@ -128,18 +128,18 @@ namespace mg5amcCpu
     PinnedHostBufferBase( const size_t size )
       : BufferBase<T>( size, false )
     {
-      checkCuda( cudaMallocHost( &( this->m_data ), this->bytes() ) );
+      gpuMallocHost( &( this->m_data ), this->bytes() );
     }
     virtual ~PinnedHostBufferBase()
     {
-      checkCuda( cudaFreeHost( this->m_data ) );
+      gpuFreeHost( this->m_data );
     }
   };
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA device buffer
   template<typename T>
   class DeviceBufferBase : public BufferBase<T>
@@ -148,18 +148,18 @@ namespace mg5amcCpu
     DeviceBufferBase( const size_t size )
       : BufferBase<T>( size, true )
     {
-      checkCuda( cudaMalloc( &( this->m_data ), this->bytes() ) );
+      gpuMalloc( &( this->m_data ), this->bytes() );
     }
     virtual ~DeviceBufferBase()
     {
-      checkCuda( cudaFree( this->m_data ) );
+      gpuFree( this->m_data );
     }
   };
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for a given number of events
   template<typename T, size_t sizePerEvent, bool ismisaligned>
   class HostBuffer : public HostBufferBase<T, ismisaligned>, virtual private NumberOfEvents
@@ -175,7 +175,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA pinned host buffer for a given number of events
   template<typename T, size_t sizePerEvent>
   class PinnedHostBuffer : public PinnedHostBufferBase<T>, virtual private NumberOfEvents
@@ -191,7 +191,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA device buffer for a given number of events
   template<typename T, size_t sizePerEvent>
   class DeviceBuffer : public DeviceBufferBase<T>, virtual private NumberOfEvents
@@ -213,7 +213,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for momenta random numbers
   constexpr size_t sizePerEventRndNumMomenta = MemoryBuffers::np4 * MemoryBuffers::nparf;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for momenta random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumMomenta, HostBufferALIGNED> HostBufferRndNumMomenta;
 #else
@@ -232,7 +232,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer with ONE fptype per event
   constexpr size_t sizePerEventOneFp = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer with ONE fptype per event
   typedef HostBuffer<fptype, sizePerEventOneFp, HostBufferALIGNED> HostBufferOneFp;
 #else
@@ -257,7 +257,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for Gs
   constexpr size_t sizePerEventGs = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventGs, HostBufferALIGNED> HostBufferGs;
 #else
@@ -276,7 +276,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for numerators
   constexpr size_t sizePerEventNumerators = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventNumerators, HostBufferALIGNED> HostBufferNumerators;
 #else
@@ -296,7 +296,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for denominators
   constexpr size_t sizePerEventDenominators = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventDenominators, HostBufferALIGNED> HostBufferDenominators;
 #else
@@ -315,7 +315,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for random numbers
   constexpr size_t sizePerEventCouplings = MemoryBuffers::ndcoup * MemoryBuffers::nx2;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventCouplings, HostBufferALIGNED> HostBufferCouplings;
 #else
@@ -333,7 +333,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for momenta
   constexpr size_t sizePerEventMomenta = MemoryBuffers::np4 * MemoryBuffers::npar;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for momenta
   typedef HostBuffer<fptype, sizePerEventMomenta, HostBufferALIGNED> HostBufferMomenta;
   //typedef HostBuffer<fptype, sizePerEventMomenta, HostBufferMISALIGNED> HostBufferMomenta; // TEST MISALIGNMENT!
@@ -352,7 +352,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for sampling weights
   constexpr size_t sizePerEventWeights = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for sampling weights
   typedef HostBuffer<fptype, sizePerEventWeights, HostBufferALIGNED> HostBufferWeights;
 #else
@@ -370,7 +370,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for matrix elements
   constexpr size_t sizePerEventMatrixElements = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for matrix elements
   typedef HostBuffer<fptype, sizePerEventMatrixElements, HostBufferALIGNED> HostBufferMatrixElements;
 #else
@@ -385,7 +385,7 @@ namespace mg5amcCpu
   // A base class encapsulating a memory buffer for the helicity mask
   typedef BufferBase<bool> BufferHelicityMask;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for the helicity mask
   typedef HostBufferBase<bool, HostBufferALIGNED> HostBufferHelicityMask;
 #else
@@ -403,7 +403,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for wavefunctions
   constexpr size_t sizePerEventWavefunctions = MemoryBuffers::nw6 * MemoryBuffers::nx2;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for wavefunctions
   typedef HostBuffer<fptype, sizePerEventWavefunctions, HostBufferALIGNED> HostBufferWavefunctions;
 #else
@@ -421,7 +421,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for helicity random numbers
   constexpr size_t sizePerEventRndNumHelicity = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for helicity random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumHelicity, HostBufferALIGNED> HostBufferRndNumHelicity;
 #else
@@ -439,7 +439,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for color random numbers
   constexpr size_t sizePerEventRndNumColor = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for color random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumColor, HostBufferALIGNED> HostBufferRndNumColor;
 #else
@@ -457,7 +457,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for helicity selection
   constexpr size_t sizePerEventSelectedHelicity = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for helicity selection
   typedef HostBuffer<int, sizePerEventSelectedHelicity, HostBufferALIGNED> HostBufferSelectedHelicity;
 #else
@@ -475,7 +475,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for color selection
   constexpr size_t sizePerEventSelectedColor = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for color selection
   typedef HostBuffer<int, sizePerEventSelectedColor, HostBufferALIGNED> HostBufferSelectedColor;
 #else
@@ -487,7 +487,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<class Tdst, class Tsrc>
   void copyDeviceFromHost( Tdst& dst, const Tsrc& src ) // keep the same order of arguments as in memcpy
   {
@@ -504,13 +504,13 @@ namespace mg5amcCpu
       throw std::runtime_error( sstr.str() );
     }
     // NB (PR #45): cudaMemcpy involves an intermediate memcpy to pinned memory if host array is a not a pinned host array
-    checkCuda( cudaMemcpy( dst.data(), src.data(), src.bytes(), cudaMemcpyHostToDevice ) );
+    gpuMemcpy( dst.data(), src.data(), src.bytes(), gpuMemcpyHostToDevice );
   }
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<class Tdst, class Tsrc>
   void copyHostFromDevice( Tdst& dst, const Tsrc& src ) // keep the same order of arguments as in memcpy
   {
@@ -527,7 +527,7 @@ namespace mg5amcCpu
       throw std::runtime_error( sstr.str() );
     }
     // NB (PR #45): cudaMemcpy involves an intermediate memcpy to pinned memory if host array is a not a pinned host array
-    checkCuda( cudaMemcpy( dst.data(), src.data(), src.bytes(), cudaMemcpyDeviceToHost ) );
+    gpuMemcpy( dst.data(), src.data(), src.bytes(), gpuMemcpyDeviceToHost );
   }
 #endif
 
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/CPPProcess.cc b/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/CPPProcess.cc
index afeebde3c6..2fa9b4f651 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/CPPProcess.cc
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_sm.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -46,7 +45,7 @@
 // Class member functions for calculating the matrix elements for
 // Process: g g > t t~ g WEIGHTED<=3 @1
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -80,7 +79,7 @@ namespace mg5amcCpu
   __device__ const fptype cIPD[2] = { (fptype)Parameters_sm::mdl_MT, (fptype)Parameters_sm::mdl_WT };
   __device__ const fptype* cIPC = nullptr; // unused as nicoup=0
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ fptype cIPD[2];
   __device__ __constant__ fptype* cIPC = nullptr; // unused as nicoup=0
 #else
@@ -90,7 +89,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -118,13 +117,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -151,7 +150,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -187,7 +186,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -200,8 +199,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -505,7 +506,7 @@ namespace mg5amcCpu
         { 1, -8, 10, 1, 64, -8 },
         { 10, 1, 1, -8, -8, 64 } }; // 2-D array[6][6]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -562,7 +563,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -621,7 +622,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -684,8 +685,8 @@ namespace mg5amcCpu
       { 1, 1, 1, 1, 1 },
       { 1, 1, 1, -1, -1 },
       { 1, 1, 1, -1, 1 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -726,9 +727,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     const fptype tIPD[2] = { (fptype)m_pars->mdl_MT, (fptype)m_pars->mdl_WT };
     //const cxtype tIPC[0] = { ... }; // nicoup=0
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) ) );
-    //checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ) ); // nicoup=0
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) );
+    //gpuMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
 #else
     memcpy( cIPD, tIPD, 2 * sizeof( fptype ) );
     //memcpy( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
@@ -765,7 +766,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -794,6 +795,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -807,6 +812,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -830,12 +837,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -856,7 +863,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -982,9 +989,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -1008,7 +1015,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -1028,7 +1035,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 256 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -1042,9 +1049,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -1072,7 +1082,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -1282,7 +1292,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/CPPProcess.h b/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/CPPProcess.h
index 37d6ebe981..11f562273e 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/CPPProcess.h
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -107,7 +107,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -120,7 +120,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -150,7 +150,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/CudaRuntime.h b/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/GpuAbstraction.h b/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/GpuRuntime.h b/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/check_sa.cc b/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/check_sa.cc
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/RamboSamplingKernels.cc b/epochX/cudacpp/gg_ttg.mad/SubProcesses/RamboSamplingKernels.cc
index da68aa9255..79abbcc4f8 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/RamboSamplingKernels.cc
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/RamboSamplingKernels.cc
@@ -1,11 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "RamboSamplingKernels.h"
 
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "MemoryAccessMomenta.h"
 #include "MemoryAccessRandomNumbers.h"
 #include "MemoryAccessWeights.h"
@@ -14,7 +14,7 @@
 
 #include <sstream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -92,7 +92,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingKernelDevice::RamboSamplingKernelDevice( const fptype energy,               // input: energy
                                                         const BufferRndNumMomenta& rndmom, // input: random numbers in [0,1]
                                                         BufferMomenta& momenta,            // output: momenta
@@ -135,7 +135,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   getMomentaInitialDevice( const fptype energy,
                            fptype* momenta )
@@ -147,17 +147,17 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   void
   RamboSamplingKernelDevice::getMomentaInitial()
   {
-    getMomentaInitialDevice<<<m_gpublocks, m_gputhreads>>>( m_energy, m_momenta.data() );
+    gpuLaunchKernel( getMomentaInitialDevice, m_gpublocks, m_gputhreads, m_energy, m_momenta.data() );
   }
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   getMomentaFinalDevice( const fptype energy,
                          const fptype* rndmom,
@@ -171,11 +171,11 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   void
   RamboSamplingKernelDevice::getMomentaFinal()
   {
-    getMomentaFinalDevice<<<m_gpublocks, m_gputhreads>>>( m_energy, m_rndmom.data(), m_momenta.data(), m_weights.data() );
+    gpuLaunchKernel( getMomentaFinalDevice, m_gpublocks, m_gputhreads, m_energy, m_rndmom.data(), m_momenta.data(), m_weights.data() );
   }
 #endif
 
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/RamboSamplingKernels.h b/epochX/cudacpp/gg_ttg.mad/SubProcesses/RamboSamplingKernels.h
index 184089efd7..7c214cd74b 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/RamboSamplingKernels.h
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/RamboSamplingKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef RAMBOSAMPLINGKERNELS_H
 #define RAMBOSAMPLINGKERNELS_H 1
@@ -10,7 +10,7 @@
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -93,7 +93,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating RAMBO phase space sampling on a GPU device
   class RamboSamplingKernelDevice final : public SamplingKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/RandomNumberKernels.h b/epochX/cudacpp/gg_ttg.mad/SubProcesses/RandomNumberKernels.h
index 188a72c2c9..21d63beeac 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/RandomNumberKernels.h
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/RandomNumberKernels.h
@@ -1,14 +1,14 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef RANDOMNUMBERKERNELS_H
 #define RANDOMNUMBERKERNELS_H 1
 
 #include "mgOnGpuConfig.h"
 
-// NB This must come AFTER mgOnGpuConfig.h which contains our definition of __global__ when __CUDACC__ is not defined
+// NB This must come AFTER mgOnGpuConfig.h which contains our definition of __global__ when MGONGPUCPP_GPUIMPL is not defined
 #ifndef MGONGPU_HAS_NO_CURAND
 //#include "curand.h"
 struct curandGenerator_st; // forward definition from curand.h
@@ -16,7 +16,7 @@ struct curandGenerator_st; // forward definition from curand.h
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/cudacpp.mk b/epochX/cudacpp/gg_ttg.mad/SubProcesses/cudacpp.mk
index 509307506b..df74dfc284 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/cudacpp.mk
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/cudacpp.mk
@@ -1,7 +1,7 @@
 # Copyright (C) 2020-2023 CERN and UCLouvain.
 # Licensed under the GNU Lesser General Public License (version 3 or later).
 # Created by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-# Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+# Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #=== Determine the name of this makefile (https://ftp.gnu.org/old-gnu/Manuals/make-3.80/html_node/make_17.html)
 #=== NB: use ':=' to ensure that the value of CUDACPP_MAKEFILE is not modified further down after including make_opts
@@ -42,10 +42,10 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure common compiler flags for C++ and CUDA
+#=== Configure common compiler flags for C++ and CUDA/HIP
 
 INCFLAGS = -I.
-OPTFLAGS = -O3 # this ends up in CUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
+OPTFLAGS = -O3 # this ends up in GPUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
 
 # Dependency on src directory
 MG5AMC_COMMONLIB = mg5amc_common
@@ -121,24 +121,45 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure the CUDA compiler
+#=== Configure the GPU compiler (CUDA or HIP)
+
+# FIXME! (AV 24.01.2024)
+# In the current implementation (without separate builds for C++ and CUDA/HIP), we first check for cudacc and hipcc in CUDA_HOME and HIP_HOME.
+# If CUDA_HOME or HIP_HOME are not set, try to determine them from the path to cudacc and hipcc.
+# While convoluted, this is currently necessary to allow disabling CUDA/HIP builds by setting CUDA_HOME or HIP_HOME to invalid paths.
+# This will (probably?) be fixed when separate C++ and CUDA/HIP builds are implemented (PR #775).
 
 # If CXX is not a single word (example "clang++ --gcc-toolchain...") then disable CUDA builds (issue #505)
-# This is because it is impossible to pass this to "CUFLAGS += -ccbin <host-compiler>" below
+# This is because it is impossible to pass this to "GPUFLAGS += -ccbin <host-compiler>" below
 ifneq ($(words $(subst ccache ,,$(CXX))),1) # allow at most "CXX=ccache <host-compiler>" from outside
   $(warning CUDA builds are not supported for multi-word CXX "$(CXX)")
   override CUDA_HOME=disabled
 endif
 
-# If CUDA_HOME is not set, try to set it from the location of nvcc
+# If CUDA_HOME is not set, try to set it from the path to nvcc
 ifndef CUDA_HOME
-  CUDA_HOME = $(patsubst %bin/nvcc,%,$(shell which nvcc 2>/dev/null))
+  CUDA_HOME = $(patsubst %/bin/nvcc,%,$(shell which nvcc 2>/dev/null))
   $(warning CUDA_HOME was not set: using "$(CUDA_HOME)")
 endif
 
-# Set NVCC as $(CUDA_HOME)/bin/nvcc if it exists
+# If HIP_HOME is not set, try to set it from the path to hipcc
+ifndef HIP_HOME
+  HIP_HOME = $(patsubst %/bin/hipcc,%,$(shell which hipcc 2>/dev/null))
+  $(warning HIP_HOME was not set: using "$(HIP_HOME)")
+endif
+
+# FIXME! (AV 24.01.2024)
+# In the current implementation (without separate builds for C++ and CUDA/HIP),
+# builds are performed for HIP only if CUDA is not found in the path.
+# If both CUDA and HIP are installed, HIP builds can be triggered by unsetting CUDA_HOME.
+# This will be fixed when separate C++ and CUDA/HIP builds are implemented (PR #775).
+
+#--- Option 1: CUDA exists -> use CUDA
+
+# Set GPUCC as $(CUDA_HOME)/bin/nvcc if it exists
 ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
-  NVCC = $(CUDA_HOME)/bin/nvcc
+
+  GPUCC = $(CUDA_HOME)/bin/nvcc
   USE_NVTX ?=-DUSE_NVTX
   # See https://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/index.html
   # See https://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/
@@ -148,7 +169,7 @@ ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
   # Examples: use 60 for P100 (Piz Daint), 80 for A100 (Juwels Booster, NVidia raplab/Curiosity).
   MADGRAPH_CUDA_ARCHITECTURE ?= 70
   ###CUARCHFLAGS = -gencode arch=compute_$(MADGRAPH_CUDA_ARCHITECTURE),code=compute_$(MADGRAPH_CUDA_ARCHITECTURE) -gencode arch=compute_$(MADGRAPH_CUDA_ARCHITECTURE),code=sm_$(MADGRAPH_CUDA_ARCHITECTURE) # Older implementation (AV): go back to this one for multi-GPU support #533
-  ###CUARCHFLAGS = --gpu-architecture=compute_$(MADGRAPH_CUDA_ARCHITECTURE) --gpu-code=sm_$(MADGRAPH_CUDA_ARCHITECTURE),compute_$(MADGRAPH_CUDA_ARCHITECTURE) # Newer implementation (SH): cannot use this as-is for multi-GPU support #533
+  ###CUARCHFLAGS = --gpu-architecture=compute_$(MADGRAPH_CUDA_ARCHITECTURE) --gpu-code=sm_$(MADGRAPH_CUDA_ARCHITECTURE),compute_$(MADGRAPH_CUDA_ARCHITECTURE)  # Newer implementation (SH): cannot use this as-is for multi-GPU support #533
   comma:=,
   CUARCHFLAGS = $(foreach arch,$(subst $(comma), ,$(MADGRAPH_CUDA_ARCHITECTURE)),-gencode arch=compute_$(arch),code=compute_$(arch) -gencode arch=compute_$(arch),code=sm_$(arch))
   CUINC = -I$(CUDA_HOME)/include/
@@ -158,41 +179,79 @@ ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
     CURANDLIBFLAGS = -L$(CUDA_HOME)/lib64/ -lcurand # NB: -lcuda is not needed here!
   endif
   CUOPTFLAGS = -lineinfo
-  CUFLAGS = $(foreach opt, $(OPTFLAGS), -Xcompiler $(opt)) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
-  ###CUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
-  ###NVCC_VERSION = $(shell $(NVCC) --version | grep 'Cuda compilation tools' | cut -d' ' -f5 | cut -d, -f1)
-  CUFLAGS += -std=c++17 # need CUDA >= 11.2 (see #333): this is enforced in mgOnGpuConfig.h
+  ###GPUFLAGS = $(OPTFLAGS) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
+  GPUFLAGS = $(foreach opt, $(OPTFLAGS), -Xcompiler $(opt)) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
+  ###GPUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
+  ###GPUCC_VERSION = $(shell $(GPUCC) --version | grep 'Cuda compilation tools' | cut -d' ' -f5 | cut -d, -f1)
+  GPUFLAGS += -std=c++17 # need CUDA >= 11.2 (see #333): this is enforced in mgOnGpuConfig.h
   # Without -maxrregcount: baseline throughput: 6.5E8 (16384 32 12) up to 7.3E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 160 # improves throughput: 6.9E8 (16384 32 12) up to 7.7E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 128 # improves throughput: 7.3E8 (16384 32 12) up to 7.6E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 96 # degrades throughput: 4.1E8 (16384 32 12) up to 4.5E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 64 # degrades throughput: 1.7E8 (16384 32 12) flat at 1.7E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 160 # improves throughput: 6.9E8 (16384 32 12) up to 7.7E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 128 # improves throughput: 7.3E8 (16384 32 12) up to 7.6E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 96 # degrades throughput: 4.1E8 (16384 32 12) up to 4.5E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 64 # degrades throughput: 1.7E8 (16384 32 12) flat at 1.7E8 (65536 128 12)
+  CUBUILDRULEFLAGS = -Xcompiler -fPIC -c
+  CCBUILDRULEFLAGS = -Xcompiler -fPIC -c -x cu
+  CUDATESTFLAGS = -lcuda
+
+  # Set the host C++ compiler for GPUCC via "-ccbin <host-compiler>"
+  # (NB issue #505: this must be a single word, "clang++ --gcc-toolchain..." is not supported)
+  GPUFLAGS += -ccbin $(shell which $(subst ccache ,,$(CXX)))
+
+  # Allow newer (unsupported) C++ compilers with older versions of CUDA if ALLOW_UNSUPPORTED_COMPILER_IN_CUDA is set (#504)
+  ifneq ($(origin ALLOW_UNSUPPORTED_COMPILER_IN_CUDA),undefined)
+  GPUFLAGS += -allow-unsupported-compiler
+  endif
+
 else ifneq ($(origin REQUIRE_CUDA),undefined)
+
   # If REQUIRE_CUDA is set but no cuda is found, stop here (e.g. for CI tests on GPU #443)
-  $(error No cuda installation found (set CUDA_HOME or make nvcc visible in PATH))
+  $(error No cuda installation found (set CUDA_HOME or make GPUCC visible in PATH))
+
+#--- Option 2: CUDA does not exist, HIP exists -> use HIP
+
+# Set GPUCC as $(HIP_HOME)/bin/hipcc if it exists
+else ifneq ($(wildcard $(HIP_HOME)/bin/hipcc),)
+
+  GPUCC = $(HIP_HOME)/bin/hipcc
+  #USE_NVTX ?=-DUSE_NVTX # should maybe find something equivalent to this in HIP?
+  HIPARCHFLAGS = -target x86_64-linux-gnu --offload-arch=gfx90a
+  HIPINC = -I$(HIP_HOME)/include/
+  # Note: -DHIP_FAST_MATH is equivalent to -use_fast_math in HIP 
+  # (but only for single precision line 208: https://rocm-developer-tools.github.io/HIP/hcc__detail_2math__functions_8h_source.html)
+  # Note: CUOPTFLAGS should not be used for HIP, it had been added here but was then removed (#808)
+  GPUFLAGS = $(OPTFLAGS) $(INCFLAGS) $(HIPINC) $(HIPARCHFLAGS) -DHIP_FAST_MATH -DHIP_PLATFORM=amd -fPIC
+  ###GPUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
+  GPUFLAGS += -std=c++17
+  ###GPUFLAGS+= --maxrregcount 255 # (AV: is this option valid on HIP and meaningful on AMD GPUs?)
+  CUBUILDRULEFLAGS = -fPIC -c
+  CCBUILDRULEFLAGS = -fPIC -c -x hip
+
+else ifneq ($(origin REQUIRE_HIP),undefined)
+
+  # If REQUIRE_HIP is set but no HIP is found, stop here (e.g. for CI tests on GPU #443)
+  $(error No hip installation found (set HIP_HOME or make GPUCC visible in PATH))
+
+#--- Option 3: CUDA does not exist, HIP does not exist -> switch off both CUDA and HIP
+
 else
-  # No cuda. Switch cuda compilation off and go to common random numbers in C++
+
+  # No cudacc and no hipcc: switch CUDA and HIP compilation off and go to common random numbers in C++
   $(warning CUDA_HOME is not set or is invalid: export CUDA_HOME to compile with cuda)
-  override NVCC=
+  $(warning HIP_HOME is not set or is invalid: export HIP_HOME to compile with hip)
+  override GPUCC=
   override USE_NVTX=
   override CUINC=
   override CURANDLIBFLAGS=
-endif
-export NVCC
-export CUFLAGS
 
-# Set the host C++ compiler for nvcc via "-ccbin <host-compiler>"
-# (NB issue #505: this must be a single word, "clang++ --gcc-toolchain..." is not supported)
-CUFLAGS += -ccbin $(shell which $(subst ccache ,,$(CXX)))
-
-# Allow newer (unsupported) C++ compilers with older versions of CUDA if ALLOW_UNSUPPORTED_COMPILER_IN_CUDA is set (#504)
-ifneq ($(origin ALLOW_UNSUPPORTED_COMPILER_IN_CUDA),undefined)
-CUFLAGS += -allow-unsupported-compiler
 endif
 
+# Export GPUCC (so that it can also be used in cudacpp_src.mk?)
+export GPUCC
+export GPUFLAGS
+
 #-------------------------------------------------------------------------------
 
-#=== Configure ccache for C++ and CUDA builds
+#=== Configure ccache for C++ and CUDA/HIP builds
 
 # Enable ccache if USECCACHE=1
 ifeq ($(USECCACHE)$(shell echo $(CXX) | grep ccache),1)
@@ -201,15 +260,15 @@ endif
 #ifeq ($(USECCACHE)$(shell echo $(AR) | grep ccache),1)
 #  override AR:=ccache $(AR)
 #endif
-ifneq ($(NVCC),)
-  ifeq ($(USECCACHE)$(shell echo $(NVCC) | grep ccache),1)
-    override NVCC:=ccache $(NVCC)
+ifneq ($(GPUCC),)
+  ifeq ($(USECCACHE)$(shell echo $(GPUCC) | grep ccache),1)
+    override GPUCC:=ccache $(GPUCC)
   endif
 endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure PowerPC-specific compiler flags for C++ and CUDA
+#=== Configure PowerPC-specific compiler flags for C++ and CUDA/HIP
 
 # PowerPC-specific CXX compiler flags (being reviewed)
 ifeq ($(UNAME_P),ppc64le)
@@ -225,9 +284,9 @@ else
   ######CXXFLAGS+= -fno-semantic-interposition # no benefit (neither alone, nor combined with -flto)
 endif
 
-# PowerPC-specific CUDA compiler flags (to be reviewed!)
+# PowerPC-specific CUDA/HIP compiler flags (to be reviewed!)
 ifeq ($(UNAME_P),ppc64le)
-  CUFLAGS+= -Xcompiler -mno-float128
+  GPUFLAGS+= -Xcompiler -mno-float128
 endif
 
 #-------------------------------------------------------------------------------
@@ -235,9 +294,11 @@ endif
 #=== Configure defaults and check if user-defined choices exist for OMPFLAGS, AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
 
 # Set the default OMPFLAGS choice
-ifneq ($(shell $(CXX) --version | egrep '^Intel'),)
+ifneq ($(findstring hipcc,$(GPUCC)),)
+override OMPFLAGS = # disable OpenMP MT when using hipcc #802
+else ifneq ($(shell $(CXX) --version | egrep '^Intel'),)
 override OMPFLAGS = -fopenmp
-###override OMPFLAGS = # disable OpenMP MT on Intel (was ok without nvcc but not ok with nvcc before #578)
+###override OMPFLAGS = # disable OpenMP MT on Intel (was ok without GPUCC but not ok with GPUCC before #578)
 else ifneq ($(shell $(CXX) --version | egrep '^(clang)'),)
 override OMPFLAGS = -fopenmp
 ###override OMPFLAGS = # disable OpenMP MT on clang (was not ok without or with nvcc before #578)
@@ -293,7 +354,10 @@ endif
 
 # Set the default RNDGEN (random number generator) choice
 ifeq ($(RNDGEN),)
-  ifeq ($(NVCC),)
+  ifeq ($(GPUCC),)
+    override RNDGEN = hasNoCurand
+  # Edgecase for HIP compilation
+  else ifeq ($(findstring hipcc,$(GPUCC)),hipcc)
     override RNDGEN = hasNoCurand
   else ifeq ($(RNDGEN),)
     override RNDGEN = hasCurand
@@ -310,7 +374,7 @@ export OMPFLAGS
 
 #-------------------------------------------------------------------------------
 
-#=== Set the CUDA/C++ compiler flags appropriate to user-defined choices of AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
+#=== Set the CUDA/HIP/C++ compiler flags appropriate to user-defined choices of AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
 
 # Set the build flags appropriate to OMPFLAGS
 $(info OMPFLAGS=$(OMPFLAGS))
@@ -368,13 +432,13 @@ CXXFLAGS+= $(AVXFLAGS)
 $(info FPTYPE=$(FPTYPE))
 ifeq ($(FPTYPE),d)
   CXXFLAGS += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
-  CUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
+  GPUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
 else ifeq ($(FPTYPE),f)
   CXXFLAGS += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
-  CUFLAGS  += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
+  GPUFLAGS  += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
 else ifeq ($(FPTYPE),m)
   CXXFLAGS += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
-  CUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
+  GPUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
 else
   $(error Unknown FPTYPE='$(FPTYPE)': only 'd', 'f' and 'm' are supported)
 endif
@@ -383,7 +447,7 @@ endif
 $(info HELINL=$(HELINL))
 ifeq ($(HELINL),1)
   CXXFLAGS += -DMGONGPU_INLINE_HELAMPS
-  CUFLAGS  += -DMGONGPU_INLINE_HELAMPS
+  GPUFLAGS  += -DMGONGPU_INLINE_HELAMPS
 else ifneq ($(HELINL),0)
   $(error Unknown HELINL='$(HELINL)': only '0' and '1' are supported)
 endif
@@ -392,7 +456,7 @@ endif
 $(info HRDCOD=$(HRDCOD))
 ifeq ($(HRDCOD),1)
   CXXFLAGS += -DMGONGPU_HARDCODE_PARAM
-  CUFLAGS  += -DMGONGPU_HARDCODE_PARAM
+  GPUFLAGS  += -DMGONGPU_HARDCODE_PARAM
 else ifneq ($(HRDCOD),0)
   $(error Unknown HRDCOD='$(HRDCOD)': only '0' and '1' are supported)
 endif
@@ -444,11 +508,11 @@ ifeq ($(UNAME_S),Darwin)
   override CULIBFLAGSRPATH2 =
 else
   # RPATH to cuda/cpp libs when linking executables
-  override CXXLIBFLAGSRPATH = -Wl,-rpath,$(LIBDIRRPATH)
-  override CULIBFLAGSRPATH = -Xlinker -rpath,$(LIBDIRRPATH)
+  override CXXLIBFLAGSRPATH = -Wl,-rpath=$(LIBDIRRPATH)
+  override CULIBFLAGSRPATH = -Xlinker -rpath=$(LIBDIRRPATH)
   # RPATH to common lib when linking cuda/cpp libs
-  override CXXLIBFLAGSRPATH2 = -Wl,-rpath,'$$ORIGIN'
-  override CULIBFLAGSRPATH2 = -Xlinker -rpath,'$$ORIGIN'
+  override CXXLIBFLAGSRPATH2 = -Wl,-rpath='$$ORIGIN'
+  override CULIBFLAGSRPATH2 = -Xlinker -rpath='$$ORIGIN'
 endif
 
 # Setting LD_LIBRARY_PATH or DYLD_LIBRARY_PATH in the RUNTIME is no longer necessary (neither on Linux nor on Mac)
@@ -461,7 +525,7 @@ override RUNTIME =
 cxx_main=$(BUILDDIR)/check.exe
 fcxx_main=$(BUILDDIR)/fcheck.exe
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 cu_main=$(BUILDDIR)/gcheck.exe
 fcu_main=$(BUILDDIR)/fgcheck.exe
 else
@@ -492,14 +556,15 @@ $(BUILDDIR)/.build.$(TAG):
 	@touch $(BUILDDIR)/.build.$(TAG)
 
 # Generic target and build rules: objects from CUDA compilation
-ifneq ($(NVCC),)
+# NB: CCBUILDRULEFLAGS includes "-x cu" for nvcc and "-x hip" for hipcc (#810)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/%.o : %.cu *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $(CUBUILDRULEFLAGS) $< -o $@
 
 $(BUILDDIR)/%_cu.o : %.cc *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c -x cu $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $(CCBUILDRULEFLAGS) $< -o $@
 endif
 
 # Generic target and build rules: objects from C++ compilation
@@ -509,11 +574,14 @@ $(BUILDDIR)/%.o : %.cc *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	$(CXX) $(CPPFLAGS) $(CXXFLAGS) -fPIC -c $< -o $@
 
 # Apply special build flags only to CrossSectionKernel.cc and gCrossSectionKernel.cu (no fast math, see #117 and #516)
+# Added edgecase for HIP compilation
 ifeq ($(shell $(CXX) --version | grep ^nvc++),)
 $(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS := $(filter-out -ffast-math,$(CXXFLAGS))
 $(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS += -fno-fast-math
-ifneq ($(NVCC),)
-$(BUILDDIR)/gCrossSectionKernels.o: CUFLAGS += -Xcompiler -fno-fast-math
+ifeq ($(findstring nvcc,$(GPUCC)),nvcc)
+  $(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -Xcompiler -fno-fast-math
+else
+  $(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -fno-fast-math
 endif
 endif
 
@@ -530,10 +598,10 @@ ifeq ($(RNDGEN),hasCurand)
 $(BUILDDIR)/CurandRandomNumberKernel.o: CXXFLAGS += $(CUINC)
 endif
 
-# Avoid "warning: builtin __has_trivial_... is deprecated; use __is_trivially_... instead" in nvcc with icx2023 (#592)
+# Avoid "warning: builtin __has_trivial_... is deprecated; use __is_trivially_... instead" in GPUCC with icx2023 (#592)
 ifneq ($(shell $(CXX) --version | egrep '^(Intel)'),)
-ifneq ($(NVCC),)
-CUFLAGS += -Xcompiler -Wno-deprecated-builtins
+ifneq ($(GPUCC),)
+GPUFLAGS += -Wno-deprecated-builtins
 endif
 endif
 
@@ -541,8 +609,8 @@ endif
 # This patch does remove the warning, but I prefer to keep it disabled for the moment...
 ###ifneq ($(shell $(CXX) --version | egrep '^(clang|Apple clang|Intel)'),)
 ###$(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS += -Wno-overriding-t-option
-###ifneq ($(NVCC),)
-###$(BUILDDIR)/gCrossSectionKernels.o: CUFLAGS += -Xcompiler -Wno-overriding-t-option
+###ifneq ($(GPUCC),)
+###$(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -Xcompiler -Wno-overriding-t-option
 ###endif
 ###endif
 
@@ -569,7 +637,7 @@ MG5AMC_CXXLIB = mg5amc_$(processid_short)_cpp
 cxx_objects_lib=$(BUILDDIR)/CPPProcess.o $(BUILDDIR)/MatrixElementKernels.o $(BUILDDIR)/BridgeKernels.o $(BUILDDIR)/CrossSectionKernels.o
 cxx_objects_exe=$(BUILDDIR)/CommonRandomNumberKernel.o $(BUILDDIR)/RamboSamplingKernels.o
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 MG5AMC_CULIB = mg5amc_$(processid_short)_cuda
 cu_objects_lib=$(BUILDDIR)/gCPPProcess.o $(BUILDDIR)/gMatrixElementKernels.o $(BUILDDIR)/gBridgeKernels.o $(BUILDDIR)/gCrossSectionKernels.o
 cu_objects_exe=$(BUILDDIR)/gCommonRandomNumberKernel.o $(BUILDDIR)/gRamboSamplingKernels.o
@@ -581,11 +649,17 @@ $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: cxx_objects_lib += $(BUILDDIR)/fbridge.o
 $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib)
 	$(CXX) -shared -o $@ $(cxx_objects_lib) $(CXXLIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: $(BUILDDIR)/fbridge_cu.o
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: cu_objects_lib += $(BUILDDIR)/fbridge_cu.o
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cu_objects_lib)
-	$(NVCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+# Bypass std::filesystem completely to ease portability on LUMI #803
+#ifneq ($(findstring hipcc,$(GPUCC)),)
+#	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB) -lstdc++fs
+#else
+#	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+#endif
 endif
 
 #-------------------------------------------------------------------------------
@@ -602,16 +676,16 @@ $(cxx_main): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PAT
 $(cxx_main): $(BUILDDIR)/check_sa.o $(LIBDIR)/lib$(MG5AMC_CXXLIB).so $(cxx_objects_exe) $(BUILDDIR)/CurandRandomNumberKernel.o
 	$(CXX) -o $@ $(BUILDDIR)/check_sa.o $(OMPFLAGS) -ldl -pthread $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe) $(BUILDDIR)/CurandRandomNumberKernel.o $(CURANDLIBFLAGS)
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(cu_main): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(cu_main): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(cu_main): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(cu_main): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 else ifneq ($(shell $(CXX) --version | grep ^nvc++),) # support nvc++ #531
 $(cu_main): LIBFLAGS += -L$(patsubst %bin/nvc++,%lib,$(subst ccache ,,$(CXX))) -lnvhpcatm -lnvcpumath -lnvc
 endif
 $(cu_main): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(cu_main): $(BUILDDIR)/gcheck_sa.o $(LIBDIR)/lib$(MG5AMC_CULIB).so $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o
-	$(NVCC) -o $@ $(BUILDDIR)/gcheck_sa.o $(CUARCHFLAGS) $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o $(CURANDLIBFLAGS)
+	$(GPUCC) -o $@ $(BUILDDIR)/gcheck_sa.o $(CUARCHFLAGS) $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o $(CURANDLIBFLAGS)
 endif
 
 #-------------------------------------------------------------------------------
@@ -635,19 +709,27 @@ $(fcxx_main): LIBFLAGS += -L$(shell dirname $(shell $(FC) --print-file-name libg
 endif
 $(fcxx_main): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(fcxx_main): $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler.o $(LIBDIR)/lib$(MG5AMC_CXXLIB).so $(cxx_objects_exe)
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(BUILDDIR)/fcheck_sa.o $(OMPFLAGS) $(BUILDDIR)/fsampler.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe) -lstdc++
+else
 	$(CXX) -o $@ $(BUILDDIR)/fcheck_sa.o $(OMPFLAGS) $(BUILDDIR)/fsampler.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe)
+endif
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(fcu_main): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(fcu_main): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(fcu_main): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(fcu_main): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 endif
 ifeq ($(UNAME_S),Darwin)
 $(fcu_main): LIBFLAGS += -L$(shell dirname $(shell $(FC) --print-file-name libgfortran.dylib)) # add path to libgfortran on Mac #375
 endif
 $(fcu_main): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(fcu_main): $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBDIR)/lib$(MG5AMC_CULIB).so $(cu_objects_exe)
-	$(NVCC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe)
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) -lstdc++ -L$(shell dirname $(shell $(GPUCC) -print-prog-name=clang))/../../lib -lamdhip64
+else
+	$(GPUCC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe)
+endif
 endif
 
 #-------------------------------------------------------------------------------
@@ -659,7 +741,7 @@ $(BUILDDIR)/testxxx.o: testxxx_cc_ref.txt
 $(testmain): $(BUILDDIR)/testxxx.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/testxxx.o # Comment out this line to skip the C++ test of xxx functions
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/testxxx_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/testxxx_cu.o: INCFLAGS += $(GTESTINC)
 $(BUILDDIR)/testxxx_cu.o: testxxx_cc_ref.txt
@@ -672,7 +754,7 @@ $(BUILDDIR)/testmisc.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/testmisc.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/testmisc.o # Comment out this line to skip the C++ miscellaneous tests
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/testmisc_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/testmisc_cu.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/testmisc_cu.o
@@ -684,12 +766,12 @@ $(BUILDDIR)/runTest.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/runTest.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/runTest.o
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/runTest_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/runTest_cu.o: INCFLAGS += $(GTESTINC)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(testmain): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(testmain): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(testmain): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(testmain): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 else ifneq ($(shell $(CXX) --version | grep ^nvc++),) # support nvc++ #531
 $(testmain): LIBFLAGS += -L$(patsubst %bin/nvc++,%lib,$(subst ccache ,,$(CXX))) -lnvhpcatm -lnvcpumath -lnvc
 endif
@@ -713,14 +795,23 @@ $(testmain): LIBFLAGS += -lgomp
 endif
 endif
 
-ifeq ($(NVCC),) # link only runTest.o
+# Bypass std::filesystem completely to ease portability on LUMI #803
+#ifneq ($(findstring hipcc,$(GPUCC)),)
+#$(testmain): LIBFLAGS += -lstdc++fs
+#endif
+
+ifeq ($(GPUCC),) # link only runTest.o
 $(testmain): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(testmain): $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib) $(cxx_objects_exe) $(GTESTLIBS)
 	$(CXX) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) -ldl -pthread $(LIBFLAGS)
 else # link both runTest.o and runTest_cu.o
 $(testmain): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(testmain): $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) $(GTESTLIBS)
-	$(NVCC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) -lcuda
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) $(CUDATESTFLAGS) -lstdc++ -lpthread  -L$(shell dirname $(shell $(GPUCC) -print-prog-name=clang))/../../lib -lamdhip64
+else
+	$(GPUCC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) $(CUDATESTFLAGS)
+endif
 endif
 
 # Use target gtestlibs to build only googletest
@@ -829,9 +920,9 @@ ifeq ($(USECCACHE),1)
 	ccache --version | head -1
 endif
 	@echo ""
-	@echo NVCC=$(NVCC)
-ifneq ($(NVCC),)
-	$(NVCC) --version
+	@echo GPUCC=$(GPUCC)
+ifneq ($(GPUCC),)
+	$(GPUCC) --version
 endif
 	@echo ""
 	@echo CXX=$(CXX)
@@ -850,7 +941,7 @@ endif
 
 # Target: check (run the C++ test executable)
 # [NB THIS IS WHAT IS USED IN THE GITHUB CI!]
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 check: runTest cmpFcheck cmpFGcheck
 else
 check: runTest cmpFcheck
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/fbridge.cc b/epochX/cudacpp/gg_ttg.mad/SubProcesses/fbridge.cc
index 2d2b36d560..22ce3f5115 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/fbridge.cc
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/fbridge.cc
@@ -1,11 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Roiser (Oct 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "Bridge.h"
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 
 extern "C"
 {
@@ -22,7 +22,7 @@ extern "C"
    * Using the same Fortran MadEvent code, linking to the hetrerogeneous library would allow access to both CPU and GPU implementations.
    * The specific heterogeneous configuration (how many GPUs, how many threads on each CPU, etc) could be loaded in CUDA/C++ from a data file.
    */
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -46,8 +46,8 @@ extern "C"
    */
   void fbridgecreate_( CppObjectInFortran** ppbridge, const int* pnevtF, const int* pnparF, const int* pnp4F )
   {
-#ifdef __CUDACC__
-    CudaRuntime::setUp();
+#ifdef MGONGPUCPP_GPUIMPL
+    GpuRuntime::setUp();
 #endif
     // (NB: CPPProcess::initProc no longer needs to be executed here because it is called in the Bridge constructor)
     // FIXME: disable OMP in Bridge when called from Fortran
@@ -65,8 +65,8 @@ extern "C"
     Bridge<FORTRANFPTYPE>* pbridge = dynamic_cast<Bridge<FORTRANFPTYPE>*>( *ppbridge );
     if( pbridge == 0 ) throw std::runtime_error( "fbridgedelete_: invalid Bridge address" );
     delete pbridge;
-#ifdef __CUDACC__
-    CudaRuntime::tearDown();
+#ifdef MGONGPUCPP_GPUIMPL
+    GpuRuntime::tearDown();
 #endif
   }
 
@@ -96,7 +96,7 @@ extern "C"
   {
     Bridge<FORTRANFPTYPE>* pbridge = dynamic_cast<Bridge<FORTRANFPTYPE>*>( *ppbridge );
     if( pbridge == 0 ) throw std::runtime_error( "fbridgesequence_: invalid Bridge address" );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Use the device/GPU implementation in the CUDA library
     // (there is also a host implementation in this library)
     pbridge->gpu_sequence( momenta, gs, rndhel, rndcol, *pchannelId, mes, selhel, selcol );
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/fsampler.cc b/epochX/cudacpp/gg_ttg.mad/SubProcesses/fsampler.cc
index 2fb445372d..3743934f41 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/fsampler.cc
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/fsampler.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Feb 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "mgOnGpuConfig.h"
 
@@ -13,7 +13,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -40,7 +40,7 @@ namespace mg5amcCpu
   private:
     const int m_nevt; // The number of events in each iteration
     int m_iiter;      // The iteration counter (for random number seeding)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     HostBufferRndNumMomenta m_hstRndmom; // Memory buffers for random numbers
     HostBufferMomenta m_hstMomenta;      // Memory buffers for momenta
     HostBufferWeights m_hstWeights;      // Memory buffers for sampling weights
@@ -105,7 +105,7 @@ namespace mg5amcCpu
 
 extern "C"
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/runTest.cc b/epochX/cudacpp/gg_ttg.mad/SubProcesses/runTest.cc
index d4a760a71b..de327f2321 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/runTest.cc
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/runTest.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Hageboeck (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*xxx to run only testxxx.cc tests
 //----------------------------------------------------------------------------
@@ -18,7 +18,7 @@
 #include "RandomNumberKernels.h"
 #include "epoch_process_id.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using namespace mg5amcGpu;
 #else
 using namespace mg5amcCpu;
@@ -35,7 +35,7 @@ struct CUDA_CPU_TestBase : public TestDriverBase
     : TestDriverBase( npar, refFileName ) {}
 };
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 struct CPUTest : public CUDA_CPU_TestBase
 {
   // Struct data members (process, and memory structures for random numbers, momenta, matrix elements and weights on host and device)
@@ -119,7 +119,7 @@ struct CPUTest : public CUDA_CPU_TestBase
 };
 #endif
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 struct CUDATest : public CUDA_CPU_TestBase
 {
   // Reset the device when our test goes out of scope. Note that this should happen after
@@ -128,7 +128,7 @@ struct CUDATest : public CUDA_CPU_TestBase
   {
     ~DeviceReset()
     {
-      checkCuda( cudaDeviceReset() ); // this is needed by cuda-memcheck --leak-check full
+      checkGpu( gpuDeviceReset() ); // this is needed by cuda-memcheck --leak-check full
     }
   } deviceResetter;
 
@@ -256,7 +256,7 @@ INSTANTIATE_TEST_SUITE_P( prefix, \
                           test_suite_name, \
                           testing::Values( new CUDATest( MG_EPOCH_REFERENCE_FILE_NAME ) ) );
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 MG_INSTANTIATE_TEST_SUITE_GPU( XTESTID_GPU( MG_EPOCH_PROCESS_ID ), MadgraphTest );
 #else
 MG_INSTANTIATE_TEST_SUITE_CPU( XTESTID_CPU( MG_EPOCH_PROCESS_ID ), MadgraphTest );
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/testmisc.cc b/epochX/cudacpp/gg_ttg.mad/SubProcesses/testmisc.cc
index 895d6eeb56..ba9e59a8a3 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/testmisc.cc
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/testmisc.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*misc to run only testmisc.cc tests
 //----------------------------------------------------------------------------
@@ -17,7 +17,7 @@
 #include <sstream>
 #include <typeinfo>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #define TESTID( s ) s##_GPU_MISC
 #else
 #define TESTID( s ) s##_CPU_MISC
@@ -26,7 +26,7 @@
 #define XTESTID( s ) TESTID( s )
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -59,7 +59,7 @@ namespace mg5amcCpu
 
 TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testmisc )
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_ttg.mad/SubProcesses/testxxx.cc b/epochX/cudacpp/gg_ttg.mad/SubProcesses/testxxx.cc
index 3361fe5aa9..e5167de00c 100644
--- a/epochX/cudacpp/gg_ttg.mad/SubProcesses/testxxx.cc
+++ b/epochX/cudacpp/gg_ttg.mad/SubProcesses/testxxx.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Apr 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*xxx to run only testxxx.cc tests
 //----------------------------------------------------------------------------
@@ -24,7 +24,7 @@
 #include <iomanip>
 #include <iostream>
 #include <vector>
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #define TESTID( s ) s##_GPU_XXX
 #else
 #define TESTID( s ) s##_CPU_XXX
@@ -32,7 +32,7 @@
 
 #define XTESTID( s ) TESTID( s )
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -42,7 +42,7 @@ namespace mg5amcCpu
   int FPEhandlerIevt = -1;
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU): '" << FPEhandlerMessage << "' ievt=" << FPEhandlerIevt << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU neppV=" << neppV << "): '" << FPEhandlerMessage << "' ievt=" << FPEhandlerIevt << std::endl;
@@ -53,7 +53,7 @@ namespace mg5amcCpu
 
 TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -77,7 +77,7 @@ TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
   assert( nevt % neppM == 0 ); // nevt must be a multiple of neppM
   assert( nevt % neppV == 0 ); // nevt must be a multiple of neppV
   // Fill in the input momenta
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   mg5amcGpu::PinnedHostBufferMomenta hstMomenta( nevt ); // AOSOA[npagM][npar=4][np4=4][neppM]
 #else
   mg5amcCpu::HostBufferMomenta hstMomenta( nevt ); // AOSOA[npagM][npar=4][np4=4][neppM]
@@ -322,7 +322,7 @@ TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
   {
     for( int ievt = 0; ievt < nevt; ievt++ )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       using namespace mg5amcGpu;
 #else
       using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_ttg.mad/bin/internal/launch_plugin.py b/epochX/cudacpp/gg_ttg.mad/bin/internal/launch_plugin.py
index 3b09713e12..90af6b7053 100644
--- a/epochX/cudacpp/gg_ttg.mad/bin/internal/launch_plugin.py
+++ b/epochX/cudacpp/gg_ttg.mad/bin/internal/launch_plugin.py
@@ -103,7 +103,7 @@ def default_setup(self):
                        fct_mod=(self.reset_makeopts,(),{}),
                        allowed=['auto', 'none', 'sse4', 'avx2','512y','512z'])
         self.add_param('cudacpp_backend', 'CPP', include=False, hidden=False,
-                       allowed=['Fortan', 'CPP', 'CUDA'])
+                       allowed=['Fortran', 'CPP', 'CUDA'])
         self['vector_size'] = 16 # already setup in default class (just change value)
         self['aloha_flag'] = '--fast-math'
         self['matrix_flag'] = '-O3'
diff --git a/epochX/cudacpp/gg_ttg.mad/src/HelAmps_sm.h b/epochX/cudacpp/gg_ttg.mad/src/HelAmps_sm.h
index 361b488401..0dd0f3ebba 100644
--- a/epochX/cudacpp/gg_ttg.mad/src/HelAmps_sm.h
+++ b/epochX/cudacpp/gg_ttg.mad/src/HelAmps_sm.h
@@ -5,7 +5,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -28,7 +28,7 @@
 //#include <iomanip>
 //#include <iostream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttg.mad/src/Parameters_sm.cc b/epochX/cudacpp/gg_ttg.mad/src/Parameters_sm.cc
index 64fc3fea62..067445b198 100644
--- a/epochX/cudacpp/gg_ttg.mad/src/Parameters_sm.cc
+++ b/epochX/cudacpp/gg_ttg.mad/src/Parameters_sm.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -17,7 +17,7 @@
 #include <iomanip>
 #include <iostream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using namespace mg5amcGpu;
 #else
 using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_ttg.mad/src/Parameters_sm.h b/epochX/cudacpp/gg_ttg.mad/src/Parameters_sm.h
index b6568d3761..9581d66e0e 100644
--- a/epochX/cudacpp/gg_ttg.mad/src/Parameters_sm.h
+++ b/epochX/cudacpp/gg_ttg.mad/src/Parameters_sm.h
@@ -27,7 +27,7 @@
 #include "read_slha.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -93,7 +93,7 @@ namespace mg5amcCpu
 #include <limits>
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -218,7 +218,7 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -239,7 +239,7 @@ namespace mg5amcCpu
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wunused-variable"  // e.g. <<warning: unused variable ‘mdl_G__exp__2’ [-Wunused-variable]>>
 #pragma GCC diagnostic ignored "-Wunused-parameter" // e.g. <<warning: unused parameter ‘G’ [-Wunused-parameter]>>
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #pragma nv_diagnostic push
 #pragma nv_diag_suppress 177 // e.g. <<warning #177-D: variable "mdl_G__exp__2" was declared but never referenced>>
 #endif
@@ -267,7 +267,7 @@ namespace mg5amcCpu
       // End SM implementation - no special handling of vectors of floats as in EFT (#439)
       return out;
     }
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #pragma GCC diagnostic pop
 #pragma nv_diagnostic pop
 #endif
diff --git a/epochX/cudacpp/gg_ttg.mad/src/cudacpp_src.mk b/epochX/cudacpp/gg_ttg.mad/src/cudacpp_src.mk
index d4cc628aec..b2b9da5288 100644
--- a/epochX/cudacpp/gg_ttg.mad/src/cudacpp_src.mk
+++ b/epochX/cudacpp/gg_ttg.mad/src/cudacpp_src.mk
@@ -19,7 +19,7 @@ SHELL := /bin/bash
 #=== Configure common compiler flags for CUDA and C++
 
 INCFLAGS = -I.
-OPTFLAGS = -O3 # this ends up in CUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
+OPTFLAGS = -O3 # this ends up in GPUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
 
 #-------------------------------------------------------------------------------
 
@@ -45,13 +45,13 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure the CUDA compiler (note: NVCC is already exported including ccache)
+#=== Configure the CUDA compiler (note: GPUCC is already exported including ccache)
 
-###$(info NVCC=$(NVCC))
+###$(info GPUCC=$(GPUCC))
 
 #-------------------------------------------------------------------------------
 
-#=== Configure ccache for C++ builds (note: NVCC is already exported including ccache)
+#=== Configure ccache for C++ builds (note: GPUCC is already exported including ccache)
 
 # Enable ccache if USECCACHE=1
 ifeq ($(USECCACHE)$(shell echo $(CXX) | grep ccache),1)
@@ -92,6 +92,13 @@ endif
 ###$(info OMPFLAGS=$(OMPFLAGS))
 CXXFLAGS += $(OMPFLAGS)
 
+# Add correct flags for nvcc (-x cu) and hipcc (-x hip) for GPU code (see #810)
+ifeq ($(findstring nvcc,$(GPUCC)),nvcc)
+  GPUFLAGS += -Xcompiler -fPIC -c -x cu
+else ifeq ($(findstring hipcc,$(GPUCC)),hipcc)
+  GPUFLAGS += -fPIC -c -x hip
+endif
+
 # Set the build flags appropriate to each AVX choice (example: "make AVX=none")
 # [NB MGONGPU_PVW512 is needed because "-mprefer-vector-width=256" is not exposed in a macro]
 # [See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96476]
@@ -253,20 +260,20 @@ $(BUILDDIR)/%.o : %.cc *.h $(BUILDDIR)/.build.$(TAG)
 # Generic target and build rules: objects from CUDA compilation
 $(BUILDDIR)/%_cu.o : %.cc *.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c -x cu $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $< -o $@
 
 #-------------------------------------------------------------------------------
 
 cxx_objects=$(addprefix $(BUILDDIR)/, Parameters_sm.o read_slha.o)
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 cu_objects=$(addprefix $(BUILDDIR)/, Parameters_sm_cu.o)
 endif
 
 # Target (and build rules): common (src) library
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so : $(cxx_objects) $(cu_objects)
 	@if [ ! -d $(LIBDIR) ]; then echo "mkdir -p $(LIBDIR)"; mkdir -p $(LIBDIR); fi
-	$(NVCC) -shared -o $@ $(cxx_objects) $(cu_objects) $(LDFLAGS)
+	$(GPUCC) -shared -o $@ $(cxx_objects) $(cu_objects) $(LDFLAGS)
 else
 $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so : $(cxx_objects)
 	@if [ ! -d $(LIBDIR) ]; then echo "mkdir -p $(LIBDIR)"; mkdir -p $(LIBDIR); fi
diff --git a/epochX/cudacpp/gg_ttg.mad/src/mgOnGpuConfig.h b/epochX/cudacpp/gg_ttg.mad/src/mgOnGpuConfig.h
index 80032e528b..6bde4466d0 100644
--- a/epochX/cudacpp/gg_ttg.mad/src/mgOnGpuConfig.h
+++ b/epochX/cudacpp/gg_ttg.mad/src/mgOnGpuConfig.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jul 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUCONFIG_H
 #define MGONGPUCONFIG_H 1
@@ -10,12 +10,26 @@
 // There are two different code bases for standalone_cudacpp (without multichannel) and madevent+cudacpp (with multichannel)
 #define MGONGPU_SUPPORTS_MULTICHANNEL 1
 
+// Is this a GPU (CUDA, HIP) or CPU implementation?
+#ifdef __CUDACC__
+#define MGONGPUCPP_GPUIMPL cuda
+#elif defined __HIPCC__
+#define MGONGPUCPP_GPUIMPL hip
+#include "hip/hip_runtime.h" // needed for blockDim, blockIdx, threadIdx: better in mgOnGpuConfig.h than in GpuAbstraction.h
+#else
+#undef MGONGPUCPP_GPUIMPL
+#endif
+
 // ** NB1 Throughputs (e.g. 6.8E8) are events/sec for "./gcheck.exe -p 65536 128 12"
 // ** NB2 Baseline on b7g47n0004 fluctuates (probably depends on load on other VMs)
 
 // Choose if curand is supported for generating random numbers
+// For HIP, by default, do not use curand (common random numbers will be used instead)
 // For both CUDA and C++, by default, do not inline, but allow this macro to be set from outside with e.g. -DMGONGPU_HAS_NO_CURAND
-// (there exist CUDA installations, e.g. using the HPC package, which do not include curand - see PR #784)
+// (there exist CUDA installations, e.g. using the HPC package, which do not include curand - see PR #784 and #785)
+#if defined __HIPCC__
+#define MGONGPU_HAS_NO_CURAND 1
+#else
 //#ifdef __CUDACC__
 //#undef MGONGPU_HAS_NO_CURAND // default
 ////#define MGONGPU_HAS_NO_CURAND 1
@@ -23,6 +37,7 @@
 //#undef MGONGPU_HAS_NO_CURAND // default
 ////#define MGONGPU_HAS_NO_CURAND 1
 //#endif
+#endif
 
 // Choose floating point precision (for everything but color algebra #537)
 // If one of these macros has been set from outside with e.g. -DMGONGPU_FPTYPE_FLOAT, nothing happens (issue #167)
@@ -54,23 +69,31 @@
 //#undef MGONGPU_HARDCODE_PARAM // default
 ////#define MGONGPU_HARDCODE_PARAM 1
 
-// Complex type in c++: std::complex or cxsmpl (CHOOSE ONLY ONE)
-#ifndef __CUDACC__
-//#define MGONGPU_CPPCXTYPE_STDCOMPLEX 1 // ~8 percent slower on float, same on double (5.1E6/double, 9.4E6/float)
-#define MGONGPU_CPPCXTYPE_CXSMPL 1 // new default (5.1E6/double, 10.2E6/float)
-#endif
-
-// Complex type in cuda: thrust or cucomplex or cxsmpl (CHOOSE ONLY ONE)
+// Complex type in CUDA: thrust or cucomplex or cxsmpl (CHOOSE ONLY ONE)
+// (NB THIS IS MGONGPU_*CU*CXTYPE_xxx)
 #ifdef __CUDACC__
 #define MGONGPU_CUCXTYPE_THRUST 1 // default (~1.15E9/double, ~3.2E9/float)
 //#define MGONGPU_CUCXTYPE_CUCOMPLEX 1 // ~10 percent slower (1.03E9/double, ~2.8E9/float)
 //#define MGONGPU_CUCXTYPE_CXSMPL 1 // ~10 percent slower (1.00E9/double, ~2.9E9/float)
+
+// Complex type in HIP: cxsmpl (ONLY ONE OPTION POSSIBLE? #810)
+// (NB THIS IS MGONGPU_*HIP*CXTYPE_xxx)
+#elif defined __HIPCC__
+#define MGONGPU_HIPCXTYPE_CXSMPL 1 // default for HIP
+
+// Complex type in C++: std::complex or cxsmpl (CHOOSE ONLY ONE)
+// (NB THIS IS MGONGPU_*CPP*CXTYPE_xxx)
+#else
+//#define MGONGPU_CPPCXTYPE_STDCOMPLEX 1 // ~8 percent slower on float, same on double (5.1E6/double, 9.4E6/float)
+#define MGONGPU_CPPCXTYPE_CXSMPL 1 // new default (5.1E6/double, 10.2E6/float)
 #endif
 
-// Cuda nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
+// CUDA nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
 #ifdef __CUDACC__
-#undef MGONGPU_NSIGHT_DEBUG // default
+#undef MGONGPU_NSIGHT_DEBUG // default in CUDA
 //#define MGONGPU_NSIGHT_DEBUG 1
+#else
+#undef MGONGPU_NSIGHT_DEBUG // only option in HIP or C++
 #endif
 
 // SANITY CHECKS (floating point precision for everything but color algebra #537)
@@ -86,17 +109,21 @@
 #error You cannot use double precision for color algebra and single precision elsewhere
 #endif
 
-// SANITY CHECKS (c++ complex number implementation)
-#ifndef __CUDACC__
-#if defined MGONGPU_CPPCXTYPE_STDCOMPLEX and defined MGONGPU_CPPCXTYPE_CXSMPL
-#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CPPCXTYPE_STDCOMPLEX or MGONGPU_CPPCXTYPE_CXSMPL
+// SANITY CHECKS (CUDA complex number implementation)
+#ifdef __CUDACC__
+#if defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CUCOMPLEX
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX for CUDA
+#elif defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CXSMPL for CUDA
+#elif defined MGONGPU_CUCXTYPE_CUCOMPLEX and defined MGONGPU_CUCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE OF MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL for CUDA
 #endif
 #endif
 
-// SANITY CHECKS (cuda complex number implementation)
-#ifdef __CUDACC__
-#if defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CUCOMPLEX and defined MGONGPU_CUCXTYPE_CXSMPL
-#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL
+// SANITY CHECKS (C++ complex number implementation)
+#ifndef MGONGPUCPP_GPUIMPL
+#if defined MGONGPU_CPPCXTYPE_STDCOMPLEX and defined MGONGPU_CPPCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CPPCXTYPE_STDCOMPLEX or MGONGPU_CPPCXTYPE_CXSMPL for C++
 #endif
 #endif
 
@@ -134,7 +161,7 @@ namespace mgOnGpu
   // Alignment requirement for using reinterpret_cast with SIMD vectorized code
   // (using reinterpret_cast with non aligned memory may lead to segmentation faults!)
   // Only needed for C++ code but can be enforced also in NVCC builds of C++ code using CUDA>=11.2 and C++17 (#318, #319, #333)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   constexpr int cppAlign = 64; // alignment requirement for SIMD vectorization (64-byte i.e. 512-bit)
 #endif
 
@@ -145,7 +172,7 @@ using mgOnGpu::fptype;
 using mgOnGpu::fptype2;
 
 // C++ SIMD vectorization width (this will be used to set neppV)
-#ifdef __CUDACC__ // CUDA implementation has no SIMD
+#ifdef MGONGPUCPP_GPUIMPL // CUDA and HIP implementations have no SIMD
 #undef MGONGPU_CPPSIMD
 #elif defined __AVX512VL__ && defined MGONGPU_PVW512 // C++ "512z" AVX512 with 512 width (512-bit ie 64-byte): 8 (DOUBLE) or 16 (FLOAT)
 #ifdef MGONGPU_FPTYPE_DOUBLE
@@ -175,9 +202,9 @@ using mgOnGpu::fptype2;
 #undef MGONGPU_CPPSIMD
 #endif
 
-// Cuda nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
+// CUDA nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
 // Arguments (not used so far): text is __FUNCTION__, code is 0 (start) or 1 (end)
-#if defined __CUDACC__ && defined MGONGPU_NSIGHT_DEBUG /* clang-format off */
+#if defined __CUDA__ && defined MGONGPU_NSIGHT_DEBUG /* clang-format off */
 #define mgDebugDeclare() __shared__ float mgDebugCounter[mgOnGpu::ntpbMAX];
 #define mgDebugInitialise() { mgDebugCounter[threadIdx.x] = 0; }
 #define mgDebug( code, text ) { mgDebugCounter[threadIdx.x] += 1; }
@@ -189,8 +216,8 @@ using mgOnGpu::fptype2;
 #define mgDebugFinalise() { /*noop*/ }
 #endif /* clang-format on */
 
-// Define empty CUDA declaration specifiers for C++
-#ifndef __CUDACC__
+// Define empty CUDA/HIP declaration specifiers for C++
+#ifndef MGONGPUCPP_GPUIMPL
 #define __global__
 #define __host__
 #define __device__
diff --git a/epochX/cudacpp/gg_ttg.mad/src/mgOnGpuCxtypes.h b/epochX/cudacpp/gg_ttg.mad/src/mgOnGpuCxtypes.h
index ca9a9f00c0..7ede1dbfae 100644
--- a/epochX/cudacpp/gg_ttg.mad/src/mgOnGpuCxtypes.h
+++ b/epochX/cudacpp/gg_ttg.mad/src/mgOnGpuCxtypes.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022, based on earlier work by D. Smith) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUCXTYPES_H
 #define MGONGPUCXTYPES_H 1
@@ -19,7 +19,7 @@
 #include <complex>
 
 // Complex type in cuda: thrust or cucomplex or cxsmpl
-#ifdef __CUDACC__
+#ifdef __CUDACC__ // this must be __CUDAC__ (not MGONGPUCPP_GPUIMPL)
 #if defined MGONGPU_CUCXTYPE_THRUST
 #pragma clang diagnostic push
 #pragma clang diagnostic ignored "-Wtautological-compare" // for icpx2021/clang13 (https://stackoverflow.com/a/15864661)
@@ -30,8 +30,13 @@
 #elif not defined MGONGPU_CUCXTYPE_CXSMPL
 #error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL
 #endif
+// Complex type in HIP: cxsmpl
+#elif defined __HIPCC__
+#if not defined MGONGPU_HIPCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_HIPCXTYPE_CXSMPL
+#endif
 #else
-// Complex type in c++: std::complex or cxsmpl
+// Complex type in c++ or HIP: std::complex or cxsmpl
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
 #include <cmath>
 #elif not defined MGONGPU_CPPCXTYPE_CXSMPL
@@ -82,7 +87,7 @@ namespace mgOnGpu /* clang-format off */
 using mgOnGpu::cxsmpl;
 
 // Printout to stream for user defined types
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -92,7 +97,7 @@ namespace mg5amcCpu
   inline __host__ std::ostream&
   operator<<( std::ostream& out, const cxsmpl<FP>& c )
   {
-    out << std::complex( c.real(), c.imag() );
+    out << std::complex<FP>( c.real(), c.imag() );
     return out;
   }
 
@@ -215,14 +220,14 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
   // --- Type definitions (complex type: cxtype)
-#ifdef __CUDACC__ // cuda
+#ifdef __CUDACC__ // this must be __CUDAC__ (not MGONGPUCPP_GPUIMPL)
 #if defined MGONGPU_CUCXTYPE_THRUST
   typedef thrust::complex<fptype> cxtype;
 #elif defined MGONGPU_CUCXTYPE_CUCOMPLEX
@@ -255,13 +260,13 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
-#if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+#if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
 
   //------------------------------
   // CUDA or C++ - using cxsmpl
@@ -303,11 +308,11 @@ namespace mg5amcCpu
     return cxmake( c.real(), c.imag() );
   }
 
-#endif // #if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+#endif // #if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
 
   //==========================================================================
 
-#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_THRUST // cuda + thrust
+#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_THRUST // cuda + thrust (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // CUDA - using thrust::complex
@@ -347,7 +352,7 @@ namespace mg5amcCpu
 
   //==========================================================================
 
-#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_CUCOMPLEX // cuda + cucomplex
+#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_CUCOMPLEX // cuda + cucomplex (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // CUDA - using cuComplex
@@ -566,7 +571,7 @@ namespace mg5amcCpu
 
   //==========================================================================
 
-#if not defined __CUDACC__ and defined MGONGPU_CPPCXTYPE_STDCOMPLEX // c++ + stdcomplex
+#if not defined __CUDACC__ and defined MGONGPU_CPPCXTYPE_STDCOMPLEX // c++/hip + stdcomplex (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // C++ - using std::complex
@@ -633,7 +638,7 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttg.mad/src/mgOnGpuFptypes.h b/epochX/cudacpp/gg_ttg.mad/src/mgOnGpuFptypes.h
index 905c97d700..fa3a02664b 100644
--- a/epochX/cudacpp/gg_ttg.mad/src/mgOnGpuFptypes.h
+++ b/epochX/cudacpp/gg_ttg.mad/src/mgOnGpuFptypes.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUFPTYPES_H
 #define MGONGPUFPTYPES_H 1
@@ -12,7 +12,7 @@
 #include <cmath>
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL // cuda
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -20,7 +20,7 @@ namespace mg5amcCpu
 {
   //==========================================================================
 
-#ifdef __CUDACC__ // cuda
+#ifdef MGONGPUCPP_GPUIMPL // cuda
 
   //------------------------------
   // Floating point types - Cuda
@@ -64,11 +64,11 @@ namespace mg5amcCpu
 #endif
   }
 
-#endif // #ifdef __CUDACC__
+#endif // #ifdef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
   //------------------------------
   // Floating point types - C++
@@ -92,7 +92,7 @@ namespace mg5amcCpu
     return std::sqrt( f );
   }
 
-#endif // #ifndef __CUDACC__
+#endif // #ifndef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
diff --git a/epochX/cudacpp/gg_ttg.mad/src/mgOnGpuVectors.h b/epochX/cudacpp/gg_ttg.mad/src/mgOnGpuVectors.h
index e1299ba81e..cdae04326b 100644
--- a/epochX/cudacpp/gg_ttg.mad/src/mgOnGpuVectors.h
+++ b/epochX/cudacpp/gg_ttg.mad/src/mgOnGpuVectors.h
@@ -32,7 +32,7 @@
 #endif
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -131,7 +131,7 @@ namespace mg5amcCpu
 #endif
 #endif
 
-#else // i.e #ifndef MGONGPU_CPPSIMD (this includes #ifdef __CUDACC__)
+#else // i.e #ifndef MGONGPU_CPPSIMD (this includes #ifdef MGONGPUCPP_GPUIMPL)
 
   const int neppV = 1;
 
@@ -153,13 +153,13 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
   // Printout to stream for user defined types
 
@@ -805,11 +805,11 @@ namespace mg5amcCpu
 
 #endif // #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
 
-#endif // #ifndef __CUDACC__
+#endif // #ifndef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
   //------------------------------
   // Vector types - CUDA
@@ -853,12 +853,12 @@ namespace mg5amcCpu
     return mask;
   }
 
-#endif // #ifdef __CUDACC__
+#endif // #ifdef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
   // Scalar-or-vector types: scalar in CUDA, vector or scalar in C++
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   typedef bool bool_sv;
   typedef fptype fptype_sv;
   typedef fptype2 fptype2_sv;
@@ -879,7 +879,7 @@ namespace mg5amcCpu
 #endif
 
   // Scalar-or-vector zeros: scalar in CUDA, vector or scalar in C++
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   inline __host__ __device__ cxtype cxzero_sv(){ return cxtype( 0, 0 ); }
 #elif defined MGONGPU_CPPSIMD
   inline cxtype_v cxzero_sv() { return cxtype_v(); } // RRRR=0000 IIII=0000
diff --git a/epochX/cudacpp/gg_ttg.mad/src/rambo.h b/epochX/cudacpp/gg_ttg.mad/src/rambo.h
index e02ea52496..cd7e1008ea 100644
--- a/epochX/cudacpp/gg_ttg.mad/src/rambo.h
+++ b/epochX/cudacpp/gg_ttg.mad/src/rambo.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -18,7 +18,7 @@
 #include <iostream>
 
 // Simplified rambo version for 2 to N (with N>=2) processes with massless particles
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +83,7 @@ namespace mg5amcCpu
       static bool first = true;
       if( first )
       {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
         if constexpr( M_ACCESS::isOnDevice() ) // avoid
         {
           const int ievt0 = 0;
@@ -166,7 +166,7 @@ namespace mg5amcCpu
     wt = po2log;
     if( nparf != 2 ) wt = ( 2. * nparf - 4. ) * log( energy ) + z[nparf - 1];
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     // issue warnings if weight is too small or too large
     static int iwarn[5] = { 0, 0, 0, 0, 0 };
     if( wt < -180. )
diff --git a/epochX/cudacpp/gg_ttg.mad/src/read_slha.cc b/epochX/cudacpp/gg_ttg.mad/src/read_slha.cc
index 055b19a779..f8e46f2e66 100644
--- a/epochX/cudacpp/gg_ttg.mad/src/read_slha.cc
+++ b/epochX/cudacpp/gg_ttg.mad/src/read_slha.cc
@@ -11,7 +11,11 @@
 
 #include <algorithm>
 #include <cstdlib>
-#include <filesystem>
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
 #include <fstream>
 #include <iostream>
 
@@ -60,7 +64,14 @@ SLHAReader::read_slha_file( std::string file_name, bool verbose )
     {
       std::cout << "WARNING! Card file '" << file_name << "' does not exist:"
                 << " look for the file in directory $" << envpath << "='" << getenv( envpath ) << "'" << std::endl;
+      /*
+#ifdef __HIPCC__
+      const std::string file_name2 = std::experimental::filesystem::path( getenv( envpath ) ) / std::experimental::filesystem::path( file_name ).filename();
+#else
       const std::string file_name2 = std::filesystem::path( getenv( envpath ) ) / std::filesystem::path( file_name ).filename();
+#endif
+      */
+      const std::string file_name2 = std::string( getenv( envpath ) ) + "/" + file_name; // bypass std::filesystem #803
       param_card.open( file_name2.c_str(), std::ifstream::in );
       if( param_card.good() )
       {
diff --git a/epochX/cudacpp/gg_ttg.sa/CODEGEN_cudacpp_gg_ttg_log.txt b/epochX/cudacpp/gg_ttg.sa/CODEGEN_cudacpp_gg_ttg_log.txt
index adda711aad..f034db4427 100644
--- a/epochX/cudacpp/gg_ttg.sa/CODEGEN_cudacpp_gg_ttg_log.txt
+++ b/epochX/cudacpp/gg_ttg.sa/CODEGEN_cudacpp_gg_ttg_log.txt
@@ -52,7 +52,7 @@ Note that you can still compile and run aMC@NLO with the built-in PDFs
 
 Using default text editor "vi". Set another one in ./input/mg5_configuration.txt
 Using default eps viewer "evince". Set another one in ./input/mg5_configuration.txt
-No valid web browser found. Please set in ./input/mg5_configuration.txt
+Using default web browser "firefox". Set another one in ./input/mg5_configuration.txt
 import /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_cudacpp_gg_ttg.mg
 The import format was not given, so we guess it as command
 set stdout_level DEBUG
@@ -62,7 +62,7 @@ generate g g > t t~ g
 No model currently active, so we import the Standard Model
 INFO: load particles 
 INFO: load vertices 
-DEBUG: model prefixing  takes 0.005533933639526367 
+DEBUG: model prefixing  takes 0.00534820556640625 
 INFO: Restrict model sm with file models/sm/restrict_default.dat . 
 DEBUG: Simplifying conditional expressions 
 DEBUG: remove interactions: u s w+ at order: QED=1 
@@ -175,7 +175,7 @@ INFO: Creating files in directory /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TM
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_cudacpp_gg_ttg/SubProcesses/P1_Sigma_sm_gg_ttxg/./CPPProcess.h
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_cudacpp_gg_ttg/SubProcesses/P1_Sigma_sm_gg_ttxg/./CPPProcess.cc
 INFO: Created files CPPProcess.h and CPPProcess.cc in directory /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_cudacpp_gg_ttg/SubProcesses/P1_Sigma_sm_gg_ttxg/. 
-Generated helas calls for 1 subprocesses (16 diagrams) in 0.037 s
+Generated helas calls for 1 subprocesses (16 diagrams) in 0.036 s
 DEBUG:  Entering PLUGIN_ProcessExporter.convert_model (create the model) [output.py at line 202] 
 ALOHA: aloha starts to compute helicity amplitudes
 ALOHA: aloha creates VVV1 routines
@@ -183,7 +183,7 @@ ALOHA: aloha creates FFV1 routines
 ALOHA: aloha creates VVVV1 set of routines with options: P0
 ALOHA: aloha creates VVVV3 set of routines with options: P0
 ALOHA: aloha creates VVVV4 set of routines with options: P0
-ALOHA: aloha creates 5 routines in  0.328 s
+ALOHA: aloha creates 5 routines in  0.325 s
 <class 'aloha.create_aloha.AbstractRoutine'> VVV1
 <class 'aloha.create_aloha.AbstractRoutine'> VVV1
 <class 'aloha.create_aloha.AbstractRoutine'> FFV1
@@ -203,7 +203,7 @@ INFO: Created files Parameters_sm.h and Parameters_sm.cc in directory
 INFO: /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_cudacpp_gg_ttg/src/. and /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_cudacpp_gg_ttg/src/. 
 quit
 
-real	0m0.787s
-user	0m0.730s
-sys	0m0.049s
-Code generation completed in 0 seconds
+real	0m0.783s
+user	0m0.718s
+sys	0m0.051s
+Code generation completed in 1 seconds
diff --git a/epochX/cudacpp/gg_ttg.sa/COPYRIGHT b/epochX/cudacpp/gg_ttg.sa/COPYRIGHT
index a134b5fef9..9036d9260a 100644
--- a/epochX/cudacpp/gg_ttg.sa/COPYRIGHT
+++ b/epochX/cudacpp/gg_ttg.sa/COPYRIGHT
@@ -15,6 +15,7 @@ The full development team currently includes the following authors :
   Stephan Hageboeck (CERN)
   Olivier Mattelaer (Universite Catholique de Louvain, original author)
   Stefan Roiser (CERN, original author)
+  Jorgen Teig (CERN)
   Andrea Valassi (CERN, original author)
   Zenny Wettersten (CERN)
 See https://github.com/madgraph5/madgraph4gpu for more details. For the full
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/Bridge.h b/epochX/cudacpp/gg_ttg.sa/SubProcesses/Bridge.h
index bf8b5e024d..f9ed70dfde 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/Bridge.h
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/Bridge.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Roiser (Nov 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef BRIDGE_H
 #define BRIDGE_H 1
@@ -14,16 +14,23 @@
 #include "MemoryAccessMomenta.h"  // for MemoryAccessMomenta::neppM
 #include "MemoryBuffers.h"        // for HostBufferMomenta, DeviceBufferMomenta etc
 
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
+
+#include <sys/stat.h> // bypass std::filesystem #803
+
 #include <algorithm>
 #include <cassert>
 #include <cmath>
 #include <cstring>
-#include <filesystem>
 #include <iostream>
 #include <memory>
 #include <type_traits>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +90,7 @@ namespace mg5amcCpu
     Bridge& operator=( const Bridge& ) = delete;
     Bridge& operator=( Bridge&& ) = delete;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     /**
      * Set the gpublocks and gputhreads for the gpusequence - throws if evnt != gpublocks*gputhreads
      * (this is needed for BridgeKernel tests rather than for actual production use in Fortran)
@@ -150,7 +157,7 @@ namespace mg5amcCpu
     unsigned int m_nevt; // number of events
     int m_nGoodHel;      // the number of good helicities (-1 initially when they have not yet been calculated)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     int m_gputhreads; // number of gpu threads (default set from number of events, can be modified)
     int m_gpublocks;  // number of gpu blocks (default set from number of events, can be modified)
     DeviceBuffer<FORTRANFPTYPE, sizePerEventMomenta> m_devMomentaF;
@@ -187,12 +194,12 @@ namespace mg5amcCpu
   // Forward declare transposition methods
   //
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
   template<typename Tin, typename Tout>
   __global__ void dev_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt );
 
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
 
   template<typename Tin, typename Tout>
   void hst_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt );
@@ -209,7 +216,7 @@ namespace mg5amcCpu
   Bridge<FORTRANFPTYPE>::Bridge( unsigned int nevtF, unsigned int nparF, unsigned int np4F )
     : m_nevt( nevtF )
     , m_nGoodHel( -1 )
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     , m_gputhreads( 256 )                  // default number of gpu threads
     , m_gpublocks( m_nevt / m_gputhreads ) // this ensures m_nevt <= m_gpublocks*m_gputhreads
     , m_devMomentaF( m_nevt )
@@ -233,7 +240,7 @@ namespace mg5amcCpu
   {
     if( nparF != CPPProcess::npar ) throw std::runtime_error( "Bridge constructor: npar mismatch" );
     if( np4F != CPPProcess::np4 ) throw std::runtime_error( "Bridge constructor: np4 mismatch" );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( ( m_nevt < s_gputhreadsmin ) || ( m_nevt % s_gputhreadsmin != 0 ) )
       throw std::runtime_error( "Bridge constructor: nevt should be a multiple of " + std::to_string( s_gputhreadsmin ) );
     while( m_nevt != m_gpublocks * m_gputhreads )
@@ -249,20 +256,28 @@ namespace mg5amcCpu
 #else
     std::cout << "WARNING! Instantiate host Bridge (nevt=" << m_nevt << ")" << std::endl;
     m_pmek.reset( new MatrixElementKernelHost( m_hstMomentaC, m_hstGs, m_hstRndHel, m_hstRndCol, m_hstMEs, m_hstSelHel, m_hstSelCol, m_nevt ) );
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
     // Create a process object, read param card and set parameters
     // FIXME: the process instance can happily go out of scope because it is only needed to read parameters?
     // FIXME: the CPPProcess should really be a singleton? what if fbridgecreate is called from several Fortran threads?
     CPPProcess process( /*verbose=*/false );
     std::string paramCard = "../../Cards/param_card.dat";
-    if( !std::filesystem::exists( paramCard ) )
-    {
-      paramCard = "../" + paramCard;
-    }
+    /*
+#ifdef __HIPCC__
+    if( !std::experimental::filesystem::exists( paramCard ) ) paramCard = "../" + paramCard;
+#else
+    if( !std::filesystem::exists( paramCard ) ) paramCard = "../" + paramCard;
+#endif
+    */
+    //struct stat dummybuffer; // bypass std::filesystem #803
+    //if( !( stat( paramCard.c_str(), &dummyBuffer ) == 0 ) ) paramCard = "../" + paramCard; //
+    auto fileExists = []( std::string& fileName )
+    { struct stat buffer; return stat( fileName.c_str(), &buffer ) == 0; };
+    if( !fileExists( paramCard ) ) paramCard = "../" + paramCard; // bypass std::filesystem #803
     process.initProc( paramCard );
   }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::set_gpugrid( const int gpublocks, const int gputhreads )
   {
@@ -276,7 +291,7 @@ namespace mg5amcCpu
   }
 #endif
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::gpu_sequence( const FORTRANFPTYPE* momenta,
                                             const FORTRANFPTYPE* gs,
@@ -291,14 +306,14 @@ namespace mg5amcCpu
     constexpr int neppM = MemoryAccessMomenta::neppM;
     if constexpr( neppM == 1 && std::is_same_v<FORTRANFPTYPE, fptype> )
     {
-      checkCuda( cudaMemcpy( m_devMomentaC.data(), momenta, m_devMomentaC.bytes(), cudaMemcpyHostToDevice ) );
+      gpuMemcpy( m_devMomentaC.data(), momenta, m_devMomentaC.bytes(), gpuMemcpyHostToDevice );
     }
     else
     {
-      checkCuda( cudaMemcpy( m_devMomentaF.data(), momenta, m_devMomentaF.bytes(), cudaMemcpyHostToDevice ) );
+      gpuMemcpy( m_devMomentaF.data(), momenta, m_devMomentaF.bytes(), gpuMemcpyHostToDevice );
       const int thrPerEvt = CPPProcess::npar * CPPProcess::np4; // AV: transpose alg does 1 element per thread (NOT 1 event per thread)
       //const int thrPerEvt = 1; // AV: try new alg with 1 event per thread... this seems slower
-      dev_transposeMomentaF2C<<<m_gpublocks * thrPerEvt, m_gputhreads>>>( m_devMomentaF.data(), m_devMomentaC.data(), m_nevt );
+      gpuLaunchKernel( dev_transposeMomentaF2C, m_gpublocks * thrPerEvt, m_gputhreads, m_devMomentaF.data(), m_devMomentaC.data(), m_nevt );
     }
     if constexpr( std::is_same_v<FORTRANFPTYPE, fptype> )
     {
@@ -341,7 +356,7 @@ namespace mg5amcCpu
   }
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::cpu_sequence( const FORTRANFPTYPE* momenta,
                                             const FORTRANFPTYPE* gs,
@@ -396,7 +411,7 @@ namespace mg5amcCpu
   // - C++ array: momenta[npagM][npar][np4][neppM] with nevt=npagM*neppM (AOSOA)
   //
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename Tin, typename Tout>
   __global__ void dev_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt )
   {
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/BridgeKernels.cc b/epochX/cudacpp/gg_ttg.sa/SubProcesses/BridgeKernels.cc
index d58066c9c1..eaf4037a24 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/BridgeKernels.cc
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/BridgeKernels.cc
@@ -1,17 +1,18 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "BridgeKernels.h"
 
+#include "GpuAbstraction.h"
 #include "MemoryAccessMomenta.h"
 
 #include <sstream>
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -45,7 +46,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu
 {
 
@@ -96,7 +97,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/BridgeKernels.h b/epochX/cudacpp/gg_ttg.sa/SubProcesses/BridgeKernels.h
index 15eb4bff4d..3efef8ce97 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/BridgeKernels.h
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/BridgeKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef BRIDGEKERNELS_H
 #define BRIDGEKERNELS_H 1
@@ -12,7 +12,7 @@
 #include "MatrixElementKernels.h"
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -49,7 +49,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A Bridge wrapper class encapsulating matrix element calculations on a CPU host
   class BridgeKernelHost final : public BridgeKernelBase
   {
@@ -89,7 +89,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A Bridge wrapper class encapsulating matrix element calculations on a GPU device
   class BridgeKernelDevice : public BridgeKernelBase
   {
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/CommonRandomNumberKernel.cc b/epochX/cudacpp/gg_ttg.sa/SubProcesses/CommonRandomNumberKernel.cc
index 985b39f576..010bc4cbd0 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/CommonRandomNumberKernel.cc
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/CommonRandomNumberKernel.cc
@@ -1,15 +1,16 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "CommonRandomNumbers.h"
+#include "GpuAbstraction.h"
 #include "MemoryBuffers.h"
 #include "RandomNumberKernels.h"
 
 #include <cassert>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/CrossSectionKernels.cc b/epochX/cudacpp/gg_ttg.sa/SubProcesses/CrossSectionKernels.cc
index 0b355a3c8d..c15b39844d 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/CrossSectionKernels.cc
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/CrossSectionKernels.cc
@@ -1,10 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "CrossSectionKernels.h"
 
+#include "GpuAbstraction.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessWeights.h"
 #include "MemoryBuffers.h"
@@ -77,7 +78,7 @@ debug_me_is_abnormal( const fptype& me, size_t ievtALL )
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -185,7 +186,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/CrossSectionKernels.h b/epochX/cudacpp/gg_ttg.sa/SubProcesses/CrossSectionKernels.h
index 7933ca4bbf..4d9659e04e 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/CrossSectionKernels.h
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/CrossSectionKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef CROSSSECTIONKERNELS_H
 #define CROSSSECTIONKERNELS_H 1
@@ -13,7 +13,7 @@
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -96,7 +96,7 @@ namespace mg5amcCpu
   //--------------------------------------------------------------------------
 
   /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating the calculation of event statistics on a GPU device
   class CrossSectionKernelDevice : public CrossSectionKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/CudaRuntime.h b/epochX/cudacpp/gg_ttg.sa/SubProcesses/CudaRuntime.h
deleted file mode 100644
index 64ce52f4b3..0000000000
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/CudaRuntime.h
+++ /dev/null
@@ -1,85 +0,0 @@
-// Copyright (C) 2020-2023 CERN and UCLouvain.
-// Licensed under the GNU Lesser General Public License (version 3 or later).
-// Created by: S. Roiser (Jul 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
-
-#ifndef MG5AMC_CUDARUNTIME_H
-#define MG5AMC_CUDARUNTIME_H 1
-
-// MG5AMC on GPU uses the CUDA runtime API, not the lower level CUDA driver API
-// See https://docs.nvidia.com/cuda/cuda-runtime-api/driver-vs-runtime-api.html#driver-vs-runtime-api
-
-#include <cassert>
-#include <iostream>
-
-//--------------------------------------------------------------------------
-
-// See https://stackoverflow.com/a/14038590
-#ifdef __CUDACC__ /* clang-format off */
-#define checkCuda( code ) { assertCuda( code, __FILE__, __LINE__ ); }
-inline void assertCuda( cudaError_t code, const char* file, int line, bool abort = true )
-{
-  if( code != cudaSuccess )
-  {
-    printf( "ERROR! assertCuda: '%s' (%d) in %s:%d\n", cudaGetErrorString( code ), code, file, line );
-    if( abort ) assert( code == cudaSuccess );
-  }
-}
-#endif /* clang-format on */
-
-//--------------------------------------------------------------------------
-
-#ifdef __CUDACC__
-namespace mg5amcGpu
-{
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  // *** FIXME! This will all need to be designed differently when going to multi-GPU nodes! ***
-  struct CudaRuntime final
-  {
-    CudaRuntime( const bool debug = true )
-      : m_debug( debug ) { setUp( m_debug ); }
-    ~CudaRuntime() { tearDown( m_debug ); }
-    CudaRuntime( const CudaRuntime& ) = delete;
-    CudaRuntime( CudaRuntime&& ) = delete;
-    CudaRuntime& operator=( const CudaRuntime& ) = delete;
-    CudaRuntime& operator=( CudaRuntime&& ) = delete;
-    bool m_debug;
-
-    // Set up CUDA application
-    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
-    // Calling cudaSetDevice on startup is useful to properly book-keep the time spent in CUDA initialization
-    static void setUp( const bool debug = true )
-    {
-      // ** NB: it is useful to call cudaSetDevice, or cudaFree, to properly book-keep the time spent in CUDA initialization
-      // ** NB: otherwise, the first CUDA operation (eg a cudaMemcpyToSymbol in CPPProcess ctor) appears to take much longer!
-      /*
-      // [We initially added cudaFree(0) to "ease profile analysis" only because it shows up as a big recognizable block!]
-      // No explicit initialization is needed: https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#initialization
-      // It is not clear what cudaFree(0) does at all: https://stackoverflow.com/questions/69967813/
-      if ( debug ) std::cout << "__CudaRuntime: calling cudaFree(0)" << std::endl;
-      checkCuda( cudaFree( 0 ) ); // SLOW!
-      */
-      // Replace cudaFree(0) by cudaSetDevice(0), even if it is not really needed either
-      // (but see https://developer.nvidia.com/blog/cuda-pro-tip-always-set-current-device-avoid-multithreading-bugs)
-      if( debug ) std::cout << "__CudaRuntime: calling cudaSetDevice(0)" << std::endl;
-      checkCuda( cudaSetDevice( 0 ) ); // SLOW!
-    }
-
-    // Tear down CUDA application (call cudaDeviceReset)
-    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
-    // Calling cudaDeviceReset on shutdown is only needed for checking memory leaks in cuda-memcheck
-    // See https://docs.nvidia.com/cuda/cuda-memcheck/index.html#leak-checking
-    static void tearDown( const bool debug = true )
-    {
-      if( debug ) std::cout << "__CudaRuntime: calling cudaDeviceReset()" << std::endl;
-      checkCuda( cudaDeviceReset() );
-    }
-  };
-
-}
-#endif
-
-//--------------------------------------------------------------------------
-
-#endif // MG5AMC_CUDARUNTIME_H
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/CurandRandomNumberKernel.cc b/epochX/cudacpp/gg_ttg.sa/SubProcesses/CurandRandomNumberKernel.cc
index eb56333b03..08a16f6f2c 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/CurandRandomNumberKernel.cc
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/CurandRandomNumberKernel.cc
@@ -1,9 +1,9 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "MemoryBuffers.h"
 #include "RandomNumberKernels.h"
 
@@ -22,7 +22,7 @@ inline void assertCurand( curandStatus_t code, const char *file, int line, bool
 }
 #endif /* clang-format on */
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -36,7 +36,7 @@ namespace mg5amcCpu
   {
     if( m_isOnDevice )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if( !m_rnarray.isOnDevice() )
         throw std::runtime_error( "CurandRandomNumberKernel on device with a host random number array" );
 #else
@@ -114,7 +114,7 @@ namespace mg5amcCpu
     /*
     printf( "\nCurandRandomNumberKernel::generateRnarray size = %d\n", (int)m_rnarray.size() );
     fptype* data = m_rnarray.data();
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( m_rnarray.isOnDevice() )
     {
       data = new fptype[m_rnarray.size()]();
@@ -123,7 +123,7 @@ namespace mg5amcCpu
 #endif
     for( int i = 0; i < ( (int)m_rnarray.size() / 4 ); i++ )
       printf( "[%4d] %f %f %f %f\n", i * 4, data[i * 4], data[i * 4 + 2], data[i * 4 + 2], data[i * 4 + 3] );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( m_rnarray.isOnDevice() ) delete[] data;
 #endif
     */
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/EventStatistics.h b/epochX/cudacpp/gg_ttg.sa/SubProcesses/EventStatistics.h
index 48b51e0a49..b425a5bade 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/EventStatistics.h
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/EventStatistics.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef EventStatistics_H
 #define EventStatistics_H 1
@@ -16,7 +16,7 @@
 #include <limits>
 #include <string>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/GpuAbstraction.h b/epochX/cudacpp/gg_ttg.sa/SubProcesses/GpuAbstraction.h
new file mode 100644
index 0000000000..9c467b1e04
--- /dev/null
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/GpuAbstraction.h
@@ -0,0 +1,69 @@
+// Copyright (C) 2020-2023 CERN and UCLouvain.
+// Licensed under the GNU Lesser General Public License (version 3 or later).
+// Created by: J. Teig (Jul 2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+
+#ifndef MG5AMC_GPUABSTRACTION_H
+#define MG5AMC_GPUABSTRACTION_H 1
+
+#include <cassert>
+
+//--------------------------------------------------------------------------
+
+#ifdef __CUDACC__
+
+#define gpuError_t cudaError_t
+#define gpuPeekAtLastError cudaPeekAtLastError
+#define gpuGetErrorString cudaGetErrorString
+#define gpuSuccess cudaSuccess
+
+#define gpuMallocHost( ptr, size ) checkGpu( cudaMallocHost( ptr, size ) )
+#define gpuMalloc( ptr, size ) checkGpu( cudaMalloc( ptr, size ) )
+
+#define gpuMemcpy( dstData, srcData, srcBytes, func ) checkGpu( cudaMemcpy( dstData, srcData, srcBytes, func ) )
+#define gpuMemcpyHostToDevice cudaMemcpyHostToDevice
+#define gpuMemcpyDeviceToHost cudaMemcpyDeviceToHost
+#define gpuMemcpyToSymbol( type1, type2, size ) checkGpu( cudaMemcpyToSymbol( type1, type2, size ) )
+
+#define gpuFree( ptr ) checkGpu( cudaFree( ptr ) )
+#define gpuFreeHost( ptr ) checkGpu( cudaFreeHost( ptr ) )
+
+#define gpuSetDevice cudaSetDevice
+#define gpuDeviceSynchronize cudaDeviceSynchronize
+#define gpuDeviceReset cudaDeviceReset
+
+#define gpuLaunchKernel( kernel, blocks, threads, ... ) kernel<<<blocks, threads>>>( __VA_ARGS__ )
+#define gpuLaunchKernelSharedMem( kernel, blocks, threads, sharedMem, ... ) kernel<<<blocks, threads, sharedMem>>>( __VA_ARGS__ )
+
+//--------------------------------------------------------------------------
+
+#elif defined __HIPCC__
+
+#define gpuError_t hipError_t
+#define gpuPeekAtLastError hipPeekAtLastError
+#define gpuGetErrorString hipGetErrorString
+#define gpuSuccess hipSuccess
+
+#define gpuMallocHost( ptr, size ) checkGpu( hipHostMalloc( ptr, size ) ) // HostMalloc better
+#define gpuMalloc( ptr, size ) checkGpu( hipMalloc( ptr, size ) )
+
+#define gpuMemcpy( dstData, srcData, srcBytes, func ) checkGpu( hipMemcpy( dstData, srcData, srcBytes, func ) )
+#define gpuMemcpyHostToDevice hipMemcpyHostToDevice
+#define gpuMemcpyDeviceToHost hipMemcpyDeviceToHost
+#define gpuMemcpyToSymbol( type1, type2, size ) checkGpu( hipMemcpyToSymbol( type1, type2, size ) )
+
+#define gpuFree( ptr ) checkGpu( hipFree( ptr ) )
+#define gpuFreeHost( ptr ) checkGpu( hipHostFree( ptr ) )
+
+#define gpuSetDevice hipSetDevice
+#define gpuDeviceSynchronize hipDeviceSynchronize
+#define gpuDeviceReset hipDeviceReset
+
+#define gpuLaunchKernel( kernel, blocks, threads, ... ) kernel<<<blocks, threads>>>( __VA_ARGS__ )
+#define gpuLaunchKernelSharedMem( kernel, blocks, threads, sharedMem, ... ) kernel<<<blocks, threads, sharedMem>>>( __VA_ARGS__ )
+
+//--------------------------------------------------------------------------
+
+#endif
+
+#endif // MG5AMC_GPUABSTRACTION_H
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/GpuRuntime.h b/epochX/cudacpp/gg_ttg.sa/SubProcesses/GpuRuntime.h
new file mode 100644
index 0000000000..93579ef08b
--- /dev/null
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/GpuRuntime.h
@@ -0,0 +1,85 @@
+// Copyright (C) 2020-2023 CERN and UCLouvain.
+// Licensed under the GNU Lesser General Public License (version 3 or later).
+// Created by: J. Teig (Jun 2023, based on earlier work by S. Roiser) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+
+#ifndef MG5AMC_GPURUNTIME_H
+#define MG5AMC_GPURUNTIME_H 1
+
+// MG5AMC on GPU uses the CUDA runtime API, not the lower level CUDA driver API
+// See https://docs.nvidia.com/cuda/cuda-runtime-api/driver-vs-runtime-api.html#driver-vs-runtime-api
+
+#include "GpuAbstraction.h"
+
+#include <iostream>
+
+//--------------------------------------------------------------------------
+
+// See https://stackoverflow.com/a/14038590
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
+#define checkGpu( code ) { assertGpu( code, __FILE__, __LINE__ ); }
+inline void assertGpu( gpuError_t code, const char* file, int line, bool abort = true )
+{
+  if( code != gpuSuccess )
+  {
+    printf( "ERROR! assertGpu: '%s' (%d) in %s:%d\n", gpuGetErrorString( code ), code, file, line );
+    if( abort ) assert( code == gpuSuccess );
+  }
+}
+#endif /* clang-format on */
+
+//--------------------------------------------------------------------------
+
+#ifdef MGONGPUCPP_GPUIMPL
+namespace mg5amcGpu
+{
+  // Instantiate a GpuRuntime at the beginnining of the application's main to
+  // invoke gpuSetDevice(0) in the constructor and book a gpuDeviceReset() call in the destructor
+  // *** FIXME! This will all need to be designed differently when going to multi-GPU nodes! ***
+  struct GpuRuntime final
+  {
+    GpuRuntime( const bool debug = true )
+      : m_debug( debug ) { setUp( m_debug ); }
+    ~GpuRuntime() { tearDown( m_debug ); }
+    GpuRuntime( const GpuRuntime& ) = delete;
+    GpuRuntime( GpuRuntime&& ) = delete;
+    GpuRuntime& operator=( const GpuRuntime& ) = delete;
+    GpuRuntime& operator=( GpuRuntime&& ) = delete;
+    bool m_debug;
+
+    // Set up CUDA application
+    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
+    // Calling cudaSetDevice on startup is useful to properly book-keep the time spent in CUDA initialization
+    static void setUp( const bool debug = true )
+    {
+      // ** NB: it is useful to call cudaSetDevice, or cudaFree, to properly book-keep the time spent in CUDA initialization
+      // ** NB: otherwise, the first CUDA operation (eg a cudaMemcpyToSymbol in CPPProcess ctor) appears to take much longer!
+      /*
+      // [We initially added cudaFree(0) to "ease profile analysis" only because it shows up as a big recognizable block!]
+      // No explicit initialization is needed: https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#initialization
+      // It is not clear what cudaFree(0) does at all: https://stackoverflow.com/questions/69967813/
+      if ( debug ) std::cout << "__CudaRuntime: calling cudaFree(0)" << std::endl;
+      checkCuda( cudaFree( 0 ) ); // SLOW!
+      */
+      // Replace cudaFree(0) by cudaSetDevice(0), even if it is not really needed either
+      // (but see https://developer.nvidia.com/blog/cuda-pro-tip-always-set-current-device-avoid-multithreading-bugs)
+      if( debug ) std::cout << "__GpuRuntime: calling GpuSetDevice(0)" << std::endl;
+      checkGpu( gpuSetDevice( 0 ) ); // SLOW!
+    }
+
+    // Tear down CUDA application (call cudaDeviceReset)
+    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
+    // Calling cudaDeviceReset on shutdown is only needed for checking memory leaks in cuda-memcheck
+    // See https://docs.nvidia.com/cuda/cuda-memcheck/index.html#leak-checking
+    static void tearDown( const bool debug = true )
+    {
+      if( debug ) std::cout << "__GpuRuntime: calling GpuDeviceReset()" << std::endl;
+      checkGpu( gpuDeviceReset() );
+    }
+  };
+}
+#endif
+
+//--------------------------------------------------------------------------
+
+#endif // MG5AMC_GPURUNTIME_H
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/MadgraphTest.h b/epochX/cudacpp/gg_ttg.sa/SubProcesses/MadgraphTest.h
index ef40624c88..6054185300 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/MadgraphTest.h
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/MadgraphTest.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Hageboeck (Dec 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MADGRAPHTEST_H_
 #define MADGRAPHTEST_H_ 1
@@ -14,7 +14,11 @@
 
 #include <array>
 #include <cmath>
-#include <filesystem>
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
 #include <fstream>
 #include <iomanip>
 #include <memory>
@@ -22,7 +26,7 @@
 #include <string>
 #include <vector>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using mg5amcGpu::CPPProcess;
 #else
 using mg5amcCpu::CPPProcess;
@@ -201,7 +205,7 @@ class MadgraphTest : public testing::TestWithParam<TestDriverBase*>
 
 // Since we link both the CPU-only and GPU tests into the same executable, we prevent
 // a multiply defined symbol by only compiling this in the non-CUDA phase:
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
 /// Compare momenta and matrix elements.
 /// This uses an implementation of TestDriverBase to run a madgraph workflow,
@@ -219,7 +223,14 @@ TEST_P( MadgraphTest, CompareMomentaAndME )
   const char* dumpEventsC = getenv( "CUDACPP_RUNTEST_DUMPEVENTS" );
   const bool dumpEvents = ( dumpEventsC != 0 ) && ( std::string( dumpEventsC ) != "" );
   const std::string refFileName = testDriver->getRefFileName();
+  /*
+#ifdef __HIPCC__
+  const std::string dumpFileName = std::experimental::filesystem::path( refFileName ).filename();
+#else
   const std::string dumpFileName = std::filesystem::path( refFileName ).filename();
+#endif
+  */
+  const std::string dumpFileName = refFileName; // bypass std::filesystem #803
   std::ofstream dumpFile;
   if( dumpEvents )
   {
@@ -307,6 +318,6 @@ TEST_P( MadgraphTest, CompareMomentaAndME )
   }
 }
 
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
 
 #endif /* MADGRAPHTEST_H_ */
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/MatrixElementKernels.cc b/epochX/cudacpp/gg_ttg.sa/SubProcesses/MatrixElementKernels.cc
index 74b5239ebf..81699dfea9 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/MatrixElementKernels.cc
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/MatrixElementKernels.cc
@@ -1,12 +1,12 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "MatrixElementKernels.h"
 
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h" // Includes the abstraction for Nvidia/AMD compilation
 #include "MemoryAccessMomenta.h"
 #include "MemoryBuffers.h"
 
@@ -14,7 +14,7 @@
 
 //============================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu
 {
 
@@ -150,7 +150,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
@@ -209,13 +209,13 @@ namespace mg5amcGpu
     PinnedHostBufferHelicityMask hstIsGoodHel( ncomb );
     DeviceBufferHelicityMask devIsGoodHel( ncomb );
     // ... 0d1. Compute good helicity mask on the device
-    computeDependentCouplings<<<m_gpublocks, m_gputhreads>>>( m_gs.data(), m_couplings.data() );
+    gpuLaunchKernel( computeDependentCouplings, m_gpublocks, m_gputhreads, m_gs.data(), m_couplings.data() );
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-    sigmaKin_getGoodHel<<<m_gpublocks, m_gputhreads>>>( m_momenta.data(), m_couplings.data(), m_matrixElements.data(), m_numerators.data(), m_denominators.data(), devIsGoodHel.data() );
+    gpuLaunchKernel( sigmaKin_getGoodHel, m_gpublocks, m_gputhreads, m_momenta.data(), m_couplings.data(), m_matrixElements.data(), m_numerators.data(), m_denominators.data(), devIsGoodHel.data() );
 #else
-    sigmaKin_getGoodHel<<<m_gpublocks, m_gputhreads>>>( m_momenta.data(), m_couplings.data(), m_matrixElements.data(), devIsGoodHel.data() );
+    gpuLaunchKernel( sigmaKin_getGoodHel, m_gpublocks, m_gputhreads, m_momenta.data(), m_couplings.data(), m_matrixElements.data(), devIsGoodHel.data() );
 #endif
-    checkCuda( cudaPeekAtLastError() );
+    checkGpu( gpuPeekAtLastError() );
     // ... 0d2. Copy back good helicity mask to the host
     copyHostFromDevice( hstIsGoodHel, devIsGoodHel );
     // ... 0d3. Copy back good helicity list to constant memory on the device
@@ -226,19 +226,19 @@ namespace mg5amcGpu
 
   void MatrixElementKernelDevice::computeMatrixElements( const unsigned int channelId )
   {
-    computeDependentCouplings<<<m_gpublocks, m_gputhreads>>>( m_gs.data(), m_couplings.data() );
+    gpuLaunchKernel( computeDependentCouplings, m_gpublocks, m_gputhreads, m_gs.data(), m_couplings.data() );
 #ifndef MGONGPU_NSIGHT_DEBUG
     constexpr unsigned int sharedMemSize = 0;
 #else
     constexpr unsigned int sharedMemSize = ntpbMAX * sizeof( float );
 #endif
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-    sigmaKin<<<m_gpublocks, m_gputhreads, sharedMemSize>>>( m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), channelId, m_numerators.data(), m_denominators.data(), m_selhel.data(), m_selcol.data() );
+    gpuLaunchKernelSharedMem( sigmaKin, m_gpublocks, m_gputhreads, sharedMemSize, m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), channelId, m_numerators.data(), m_denominators.data(), m_selhel.data(), m_selcol.data() );
 #else
-    sigmaKin<<<m_gpublocks, m_gputhreads, sharedMemSize>>>( m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), m_selhel.data(), m_selcol.data() );
+    gpuLaunchKernelSharedMem( sigmaKin, m_gpublocks, m_gputhreads, sharedMemSize, m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), m_selhel.data(), m_selcol.data() );
 #endif
-    checkCuda( cudaPeekAtLastError() );
-    checkCuda( cudaDeviceSynchronize() );
+    checkGpu( gpuPeekAtLastError() );
+    checkGpu( gpuDeviceSynchronize() );
   }
 
   //--------------------------------------------------------------------------
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/MatrixElementKernels.h b/epochX/cudacpp/gg_ttg.sa/SubProcesses/MatrixElementKernels.h
index 23e84757a2..72bd8f195b 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/MatrixElementKernels.h
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/MatrixElementKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MATRIXELEMENTKERNELS_H
 #define MATRIXELEMENTKERNELS_H 1
@@ -10,7 +10,7 @@
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -81,7 +81,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating matrix element calculations on a CPU host
   class MatrixElementKernelHost final : public MatrixElementKernelBase, public NumberOfEvents
   {
@@ -130,7 +130,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating matrix element calculations on a GPU device
   class MatrixElementKernelDevice : public MatrixElementKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessAmplitudes.h b/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessAmplitudes.h
index 573b3bbbc9..ffb76e93de 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessAmplitudes.h
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessAmplitudes.h
@@ -15,7 +15,7 @@
 #define MGONGPU_TRIVIAL_AMPLITUDES 1
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessCouplings.h b/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessCouplings.h
index 35a3af42e0..3afdf3e554 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessCouplings.h
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessCouplings.h
@@ -15,7 +15,7 @@
 #include "MemoryBuffers.h"       // for HostBufferCouplings::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessCouplingsFixed.h b/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessCouplingsFixed.h
index dc0d93afff..ffcdf4dbef 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessCouplingsFixed.h
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessCouplingsFixed.h
@@ -14,7 +14,7 @@
 //#include "MemoryAccessHelpers.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessDenominators.h b/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessDenominators.h
index 3bce635718..66f2d32a6b 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessDenominators.h
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessDenominators.h
@@ -10,7 +10,7 @@
 #include "MemoryAccessGs.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessGs.h b/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessGs.h
index 31311aa375..4c726b30f3 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessGs.h
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessGs.h
@@ -13,7 +13,7 @@
 #include "MemoryBuffers.h" // for HostBufferMatrixElements::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessHelpers.h b/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessHelpers.h
index c82a6c7635..db73e4e064 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessHelpers.h
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessHelpers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessHelpers_H
 #define MemoryAccessHelpers_H 1
@@ -105,7 +105,7 @@ class KernelAccessHelper : public MemoryAccessHelper<T>
     }
     else
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
       //printf( "kernelAccessRecord: ievt=%d threadId=%d\n", ievt, threadIdx.x );
       return T::ieventAccessRecord( buffer, ievt ); // NB fptype and fptype_sv coincide for CUDA
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessMatrixElements.h b/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessMatrixElements.h
index f32e6fea5b..3741011971 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessMatrixElements.h
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessMatrixElements.h
@@ -13,7 +13,7 @@
 #include "MemoryBuffers.h" // for HostBufferMatrixElements::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessMomenta.h b/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessMomenta.h
index 29266de32c..3be229d392 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessMomenta.h
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessMomenta.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessMomenta_H
 #define MemoryAccessMomenta_H 1
@@ -13,7 +13,7 @@
 #include "MemoryAccessVectors.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -30,7 +30,7 @@ namespace mg5amcCpu
 
     // Number of Events Per Page in the momenta AOSOA memory buffer layout
     // (these are all best kept as a compile-time constants: see issue #23)
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
     // -----------------------------------------------------------------------------------------------
     // --- GPUs: neppM is best set to a power of 2 times the number of fptype's in a 32-byte cacheline
     // --- This is relevant to ensure coalesced access to momenta in global memory
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessNumerators.h b/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessNumerators.h
index b152183b28..18991f4fa6 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessNumerators.h
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessNumerators.h
@@ -10,7 +10,7 @@
 #include "MemoryAccessGs.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessRandomNumbers.h b/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessRandomNumbers.h
index e2988d39f3..40cb089135 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessRandomNumbers.h
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessRandomNumbers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessRandomNumbers_H
 #define MemoryAccessRandomNumbers_H 1
@@ -11,7 +11,7 @@
 #include "CPPProcess.h"
 #include "MemoryAccessHelpers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using mg5amcGpu::CPPProcess;
 #else
 using mg5amcCpu::CPPProcess;
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessVectors.h b/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessVectors.h
index e9b197368e..08faccff0f 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessVectors.h
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessVectors.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessVectors_H
 #define MemoryAccessVectors_H 1
@@ -10,7 +10,7 @@
 
 #include "mgOnGpuVectors.h"
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu // this is only needed for CPU SIMD vectorization
 {
 
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessWavefunctions.h b/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessWavefunctions.h
index 5428aaf933..33bef4559e 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessWavefunctions.h
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryAccessWavefunctions.h
@@ -15,7 +15,7 @@
 #define MGONGPU_TRIVIAL_WAVEFUNCTIONS 1
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryBuffers.h b/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryBuffers.h
index 3093e6ed18..7756a71621 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryBuffers.h
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/MemoryBuffers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021, based on earlier work by S. Hageboeck) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryBuffers_H
 #define MemoryBuffers_H 1
@@ -11,12 +11,12 @@
 #include "mgOnGpuCxtypes.h"
 
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "Parameters_sm.h"
 
 #include <sstream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -87,7 +87,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   constexpr bool HostBufferALIGNED = false;   // ismisaligned=false
   constexpr bool HostBufferMISALIGNED = true; // ismisaligned=true
 
@@ -119,7 +119,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA pinned host buffer
   template<typename T>
   class PinnedHostBufferBase : public BufferBase<T>
@@ -128,18 +128,18 @@ namespace mg5amcCpu
     PinnedHostBufferBase( const size_t size )
       : BufferBase<T>( size, false )
     {
-      checkCuda( cudaMallocHost( &( this->m_data ), this->bytes() ) );
+      gpuMallocHost( &( this->m_data ), this->bytes() );
     }
     virtual ~PinnedHostBufferBase()
     {
-      checkCuda( cudaFreeHost( this->m_data ) );
+      gpuFreeHost( this->m_data );
     }
   };
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA device buffer
   template<typename T>
   class DeviceBufferBase : public BufferBase<T>
@@ -148,18 +148,18 @@ namespace mg5amcCpu
     DeviceBufferBase( const size_t size )
       : BufferBase<T>( size, true )
     {
-      checkCuda( cudaMalloc( &( this->m_data ), this->bytes() ) );
+      gpuMalloc( &( this->m_data ), this->bytes() );
     }
     virtual ~DeviceBufferBase()
     {
-      checkCuda( cudaFree( this->m_data ) );
+      gpuFree( this->m_data );
     }
   };
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for a given number of events
   template<typename T, size_t sizePerEvent, bool ismisaligned>
   class HostBuffer : public HostBufferBase<T, ismisaligned>, virtual private NumberOfEvents
@@ -175,7 +175,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA pinned host buffer for a given number of events
   template<typename T, size_t sizePerEvent>
   class PinnedHostBuffer : public PinnedHostBufferBase<T>, virtual private NumberOfEvents
@@ -191,7 +191,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA device buffer for a given number of events
   template<typename T, size_t sizePerEvent>
   class DeviceBuffer : public DeviceBufferBase<T>, virtual private NumberOfEvents
@@ -213,7 +213,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for momenta random numbers
   constexpr size_t sizePerEventRndNumMomenta = MemoryBuffers::np4 * MemoryBuffers::nparf;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for momenta random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumMomenta, HostBufferALIGNED> HostBufferRndNumMomenta;
 #else
@@ -232,7 +232,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer with ONE fptype per event
   constexpr size_t sizePerEventOneFp = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer with ONE fptype per event
   typedef HostBuffer<fptype, sizePerEventOneFp, HostBufferALIGNED> HostBufferOneFp;
 #else
@@ -257,7 +257,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for Gs
   constexpr size_t sizePerEventGs = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventGs, HostBufferALIGNED> HostBufferGs;
 #else
@@ -276,7 +276,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for numerators
   constexpr size_t sizePerEventNumerators = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventNumerators, HostBufferALIGNED> HostBufferNumerators;
 #else
@@ -296,7 +296,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for denominators
   constexpr size_t sizePerEventDenominators = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventDenominators, HostBufferALIGNED> HostBufferDenominators;
 #else
@@ -315,7 +315,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for random numbers
   constexpr size_t sizePerEventCouplings = MemoryBuffers::ndcoup * MemoryBuffers::nx2;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventCouplings, HostBufferALIGNED> HostBufferCouplings;
 #else
@@ -333,7 +333,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for momenta
   constexpr size_t sizePerEventMomenta = MemoryBuffers::np4 * MemoryBuffers::npar;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for momenta
   typedef HostBuffer<fptype, sizePerEventMomenta, HostBufferALIGNED> HostBufferMomenta;
   //typedef HostBuffer<fptype, sizePerEventMomenta, HostBufferMISALIGNED> HostBufferMomenta; // TEST MISALIGNMENT!
@@ -352,7 +352,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for sampling weights
   constexpr size_t sizePerEventWeights = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for sampling weights
   typedef HostBuffer<fptype, sizePerEventWeights, HostBufferALIGNED> HostBufferWeights;
 #else
@@ -370,7 +370,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for matrix elements
   constexpr size_t sizePerEventMatrixElements = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for matrix elements
   typedef HostBuffer<fptype, sizePerEventMatrixElements, HostBufferALIGNED> HostBufferMatrixElements;
 #else
@@ -385,7 +385,7 @@ namespace mg5amcCpu
   // A base class encapsulating a memory buffer for the helicity mask
   typedef BufferBase<bool> BufferHelicityMask;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for the helicity mask
   typedef HostBufferBase<bool, HostBufferALIGNED> HostBufferHelicityMask;
 #else
@@ -403,7 +403,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for wavefunctions
   constexpr size_t sizePerEventWavefunctions = MemoryBuffers::nw6 * MemoryBuffers::nx2;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for wavefunctions
   typedef HostBuffer<fptype, sizePerEventWavefunctions, HostBufferALIGNED> HostBufferWavefunctions;
 #else
@@ -421,7 +421,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for helicity random numbers
   constexpr size_t sizePerEventRndNumHelicity = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for helicity random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumHelicity, HostBufferALIGNED> HostBufferRndNumHelicity;
 #else
@@ -439,7 +439,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for color random numbers
   constexpr size_t sizePerEventRndNumColor = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for color random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumColor, HostBufferALIGNED> HostBufferRndNumColor;
 #else
@@ -457,7 +457,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for helicity selection
   constexpr size_t sizePerEventSelectedHelicity = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for helicity selection
   typedef HostBuffer<int, sizePerEventSelectedHelicity, HostBufferALIGNED> HostBufferSelectedHelicity;
 #else
@@ -475,7 +475,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for color selection
   constexpr size_t sizePerEventSelectedColor = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for color selection
   typedef HostBuffer<int, sizePerEventSelectedColor, HostBufferALIGNED> HostBufferSelectedColor;
 #else
@@ -487,7 +487,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<class Tdst, class Tsrc>
   void copyDeviceFromHost( Tdst& dst, const Tsrc& src ) // keep the same order of arguments as in memcpy
   {
@@ -504,13 +504,13 @@ namespace mg5amcCpu
       throw std::runtime_error( sstr.str() );
     }
     // NB (PR #45): cudaMemcpy involves an intermediate memcpy to pinned memory if host array is a not a pinned host array
-    checkCuda( cudaMemcpy( dst.data(), src.data(), src.bytes(), cudaMemcpyHostToDevice ) );
+    gpuMemcpy( dst.data(), src.data(), src.bytes(), gpuMemcpyHostToDevice );
   }
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<class Tdst, class Tsrc>
   void copyHostFromDevice( Tdst& dst, const Tsrc& src ) // keep the same order of arguments as in memcpy
   {
@@ -527,7 +527,7 @@ namespace mg5amcCpu
       throw std::runtime_error( sstr.str() );
     }
     // NB (PR #45): cudaMemcpy involves an intermediate memcpy to pinned memory if host array is a not a pinned host array
-    checkCuda( cudaMemcpy( dst.data(), src.data(), src.bytes(), cudaMemcpyDeviceToHost ) );
+    gpuMemcpy( dst.data(), src.data(), src.bytes(), gpuMemcpyDeviceToHost );
   }
 #endif
 
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/P1_Sigma_sm_gg_ttxg/CPPProcess.cc b/epochX/cudacpp/gg_ttg.sa/SubProcesses/P1_Sigma_sm_gg_ttxg/CPPProcess.cc
index 2988a13b82..661197ace8 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/P1_Sigma_sm_gg_ttxg/CPPProcess.cc
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/P1_Sigma_sm_gg_ttxg/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_sm.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -46,7 +45,7 @@
 // Class member functions for calculating the matrix elements for
 // Process: g g > t t~ g WEIGHTED<=3 @1
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -80,7 +79,7 @@ namespace mg5amcCpu
   __device__ const fptype cIPD[2] = { (fptype)Parameters_sm::mdl_MT, (fptype)Parameters_sm::mdl_WT };
   __device__ const fptype* cIPC = nullptr; // unused as nicoup=0
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ fptype cIPD[2];
   __device__ __constant__ fptype* cIPC = nullptr; // unused as nicoup=0
 #else
@@ -90,7 +89,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -118,13 +117,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -151,7 +150,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -187,7 +186,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -200,8 +199,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -499,7 +500,7 @@ namespace mg5amcCpu
         { 1, -8, 10, 1, 64, -8 },
         { 10, 1, 1, -8, -8, 64 } }; // 2-D array[6][6]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -556,7 +557,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -615,7 +616,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -678,8 +679,8 @@ namespace mg5amcCpu
       { 1, 1, 1, 1, 1 },
       { 1, 1, 1, -1, -1 },
       { 1, 1, 1, -1, 1 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -720,9 +721,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     const fptype tIPD[2] = { (fptype)m_pars->mdl_MT, (fptype)m_pars->mdl_WT };
     //const cxtype tIPC[0] = { ... }; // nicoup=0
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) ) );
-    //checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ) ); // nicoup=0
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) );
+    //gpuMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
 #else
     memcpy( cIPD, tIPD, 2 * sizeof( fptype ) );
     //memcpy( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
@@ -759,7 +760,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -788,6 +789,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -801,6 +806,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -824,12 +831,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -850,7 +857,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -976,9 +983,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -1002,7 +1009,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -1022,7 +1029,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 256 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -1036,9 +1043,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -1066,7 +1076,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -1276,7 +1286,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/P1_Sigma_sm_gg_ttxg/CPPProcess.h b/epochX/cudacpp/gg_ttg.sa/SubProcesses/P1_Sigma_sm_gg_ttxg/CPPProcess.h
index 37d6ebe981..11f562273e 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/P1_Sigma_sm_gg_ttxg/CPPProcess.h
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/P1_Sigma_sm_gg_ttxg/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -107,7 +107,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -120,7 +120,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -150,7 +150,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/P1_Sigma_sm_gg_ttxg/CudaRuntime.h b/epochX/cudacpp/gg_ttg.sa/SubProcesses/P1_Sigma_sm_gg_ttxg/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/P1_Sigma_sm_gg_ttxg/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/P1_Sigma_sm_gg_ttxg/GpuAbstraction.h b/epochX/cudacpp/gg_ttg.sa/SubProcesses/P1_Sigma_sm_gg_ttxg/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/P1_Sigma_sm_gg_ttxg/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/P1_Sigma_sm_gg_ttxg/GpuRuntime.h b/epochX/cudacpp/gg_ttg.sa/SubProcesses/P1_Sigma_sm_gg_ttxg/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/P1_Sigma_sm_gg_ttxg/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/P1_Sigma_sm_gg_ttxg/check_sa.cc b/epochX/cudacpp/gg_ttg.sa/SubProcesses/P1_Sigma_sm_gg_ttxg/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/P1_Sigma_sm_gg_ttxg/check_sa.cc
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/P1_Sigma_sm_gg_ttxg/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/RamboSamplingKernels.cc b/epochX/cudacpp/gg_ttg.sa/SubProcesses/RamboSamplingKernels.cc
index da68aa9255..79abbcc4f8 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/RamboSamplingKernels.cc
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/RamboSamplingKernels.cc
@@ -1,11 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "RamboSamplingKernels.h"
 
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "MemoryAccessMomenta.h"
 #include "MemoryAccessRandomNumbers.h"
 #include "MemoryAccessWeights.h"
@@ -14,7 +14,7 @@
 
 #include <sstream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -92,7 +92,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingKernelDevice::RamboSamplingKernelDevice( const fptype energy,               // input: energy
                                                         const BufferRndNumMomenta& rndmom, // input: random numbers in [0,1]
                                                         BufferMomenta& momenta,            // output: momenta
@@ -135,7 +135,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   getMomentaInitialDevice( const fptype energy,
                            fptype* momenta )
@@ -147,17 +147,17 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   void
   RamboSamplingKernelDevice::getMomentaInitial()
   {
-    getMomentaInitialDevice<<<m_gpublocks, m_gputhreads>>>( m_energy, m_momenta.data() );
+    gpuLaunchKernel( getMomentaInitialDevice, m_gpublocks, m_gputhreads, m_energy, m_momenta.data() );
   }
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   getMomentaFinalDevice( const fptype energy,
                          const fptype* rndmom,
@@ -171,11 +171,11 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   void
   RamboSamplingKernelDevice::getMomentaFinal()
   {
-    getMomentaFinalDevice<<<m_gpublocks, m_gputhreads>>>( m_energy, m_rndmom.data(), m_momenta.data(), m_weights.data() );
+    gpuLaunchKernel( getMomentaFinalDevice, m_gpublocks, m_gputhreads, m_energy, m_rndmom.data(), m_momenta.data(), m_weights.data() );
   }
 #endif
 
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/RamboSamplingKernels.h b/epochX/cudacpp/gg_ttg.sa/SubProcesses/RamboSamplingKernels.h
index 184089efd7..7c214cd74b 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/RamboSamplingKernels.h
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/RamboSamplingKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef RAMBOSAMPLINGKERNELS_H
 #define RAMBOSAMPLINGKERNELS_H 1
@@ -10,7 +10,7 @@
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -93,7 +93,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating RAMBO phase space sampling on a GPU device
   class RamboSamplingKernelDevice final : public SamplingKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/RandomNumberKernels.h b/epochX/cudacpp/gg_ttg.sa/SubProcesses/RandomNumberKernels.h
index 188a72c2c9..21d63beeac 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/RandomNumberKernels.h
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/RandomNumberKernels.h
@@ -1,14 +1,14 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef RANDOMNUMBERKERNELS_H
 #define RANDOMNUMBERKERNELS_H 1
 
 #include "mgOnGpuConfig.h"
 
-// NB This must come AFTER mgOnGpuConfig.h which contains our definition of __global__ when __CUDACC__ is not defined
+// NB This must come AFTER mgOnGpuConfig.h which contains our definition of __global__ when MGONGPUCPP_GPUIMPL is not defined
 #ifndef MGONGPU_HAS_NO_CURAND
 //#include "curand.h"
 struct curandGenerator_st; // forward definition from curand.h
@@ -16,7 +16,7 @@ struct curandGenerator_st; // forward definition from curand.h
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/cudacpp.mk b/epochX/cudacpp/gg_ttg.sa/SubProcesses/cudacpp.mk
index 509307506b..df74dfc284 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/cudacpp.mk
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/cudacpp.mk
@@ -1,7 +1,7 @@
 # Copyright (C) 2020-2023 CERN and UCLouvain.
 # Licensed under the GNU Lesser General Public License (version 3 or later).
 # Created by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-# Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+# Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #=== Determine the name of this makefile (https://ftp.gnu.org/old-gnu/Manuals/make-3.80/html_node/make_17.html)
 #=== NB: use ':=' to ensure that the value of CUDACPP_MAKEFILE is not modified further down after including make_opts
@@ -42,10 +42,10 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure common compiler flags for C++ and CUDA
+#=== Configure common compiler flags for C++ and CUDA/HIP
 
 INCFLAGS = -I.
-OPTFLAGS = -O3 # this ends up in CUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
+OPTFLAGS = -O3 # this ends up in GPUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
 
 # Dependency on src directory
 MG5AMC_COMMONLIB = mg5amc_common
@@ -121,24 +121,45 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure the CUDA compiler
+#=== Configure the GPU compiler (CUDA or HIP)
+
+# FIXME! (AV 24.01.2024)
+# In the current implementation (without separate builds for C++ and CUDA/HIP), we first check for cudacc and hipcc in CUDA_HOME and HIP_HOME.
+# If CUDA_HOME or HIP_HOME are not set, try to determine them from the path to cudacc and hipcc.
+# While convoluted, this is currently necessary to allow disabling CUDA/HIP builds by setting CUDA_HOME or HIP_HOME to invalid paths.
+# This will (probably?) be fixed when separate C++ and CUDA/HIP builds are implemented (PR #775).
 
 # If CXX is not a single word (example "clang++ --gcc-toolchain...") then disable CUDA builds (issue #505)
-# This is because it is impossible to pass this to "CUFLAGS += -ccbin <host-compiler>" below
+# This is because it is impossible to pass this to "GPUFLAGS += -ccbin <host-compiler>" below
 ifneq ($(words $(subst ccache ,,$(CXX))),1) # allow at most "CXX=ccache <host-compiler>" from outside
   $(warning CUDA builds are not supported for multi-word CXX "$(CXX)")
   override CUDA_HOME=disabled
 endif
 
-# If CUDA_HOME is not set, try to set it from the location of nvcc
+# If CUDA_HOME is not set, try to set it from the path to nvcc
 ifndef CUDA_HOME
-  CUDA_HOME = $(patsubst %bin/nvcc,%,$(shell which nvcc 2>/dev/null))
+  CUDA_HOME = $(patsubst %/bin/nvcc,%,$(shell which nvcc 2>/dev/null))
   $(warning CUDA_HOME was not set: using "$(CUDA_HOME)")
 endif
 
-# Set NVCC as $(CUDA_HOME)/bin/nvcc if it exists
+# If HIP_HOME is not set, try to set it from the path to hipcc
+ifndef HIP_HOME
+  HIP_HOME = $(patsubst %/bin/hipcc,%,$(shell which hipcc 2>/dev/null))
+  $(warning HIP_HOME was not set: using "$(HIP_HOME)")
+endif
+
+# FIXME! (AV 24.01.2024)
+# In the current implementation (without separate builds for C++ and CUDA/HIP),
+# builds are performed for HIP only if CUDA is not found in the path.
+# If both CUDA and HIP are installed, HIP builds can be triggered by unsetting CUDA_HOME.
+# This will be fixed when separate C++ and CUDA/HIP builds are implemented (PR #775).
+
+#--- Option 1: CUDA exists -> use CUDA
+
+# Set GPUCC as $(CUDA_HOME)/bin/nvcc if it exists
 ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
-  NVCC = $(CUDA_HOME)/bin/nvcc
+
+  GPUCC = $(CUDA_HOME)/bin/nvcc
   USE_NVTX ?=-DUSE_NVTX
   # See https://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/index.html
   # See https://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/
@@ -148,7 +169,7 @@ ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
   # Examples: use 60 for P100 (Piz Daint), 80 for A100 (Juwels Booster, NVidia raplab/Curiosity).
   MADGRAPH_CUDA_ARCHITECTURE ?= 70
   ###CUARCHFLAGS = -gencode arch=compute_$(MADGRAPH_CUDA_ARCHITECTURE),code=compute_$(MADGRAPH_CUDA_ARCHITECTURE) -gencode arch=compute_$(MADGRAPH_CUDA_ARCHITECTURE),code=sm_$(MADGRAPH_CUDA_ARCHITECTURE) # Older implementation (AV): go back to this one for multi-GPU support #533
-  ###CUARCHFLAGS = --gpu-architecture=compute_$(MADGRAPH_CUDA_ARCHITECTURE) --gpu-code=sm_$(MADGRAPH_CUDA_ARCHITECTURE),compute_$(MADGRAPH_CUDA_ARCHITECTURE) # Newer implementation (SH): cannot use this as-is for multi-GPU support #533
+  ###CUARCHFLAGS = --gpu-architecture=compute_$(MADGRAPH_CUDA_ARCHITECTURE) --gpu-code=sm_$(MADGRAPH_CUDA_ARCHITECTURE),compute_$(MADGRAPH_CUDA_ARCHITECTURE)  # Newer implementation (SH): cannot use this as-is for multi-GPU support #533
   comma:=,
   CUARCHFLAGS = $(foreach arch,$(subst $(comma), ,$(MADGRAPH_CUDA_ARCHITECTURE)),-gencode arch=compute_$(arch),code=compute_$(arch) -gencode arch=compute_$(arch),code=sm_$(arch))
   CUINC = -I$(CUDA_HOME)/include/
@@ -158,41 +179,79 @@ ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
     CURANDLIBFLAGS = -L$(CUDA_HOME)/lib64/ -lcurand # NB: -lcuda is not needed here!
   endif
   CUOPTFLAGS = -lineinfo
-  CUFLAGS = $(foreach opt, $(OPTFLAGS), -Xcompiler $(opt)) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
-  ###CUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
-  ###NVCC_VERSION = $(shell $(NVCC) --version | grep 'Cuda compilation tools' | cut -d' ' -f5 | cut -d, -f1)
-  CUFLAGS += -std=c++17 # need CUDA >= 11.2 (see #333): this is enforced in mgOnGpuConfig.h
+  ###GPUFLAGS = $(OPTFLAGS) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
+  GPUFLAGS = $(foreach opt, $(OPTFLAGS), -Xcompiler $(opt)) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
+  ###GPUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
+  ###GPUCC_VERSION = $(shell $(GPUCC) --version | grep 'Cuda compilation tools' | cut -d' ' -f5 | cut -d, -f1)
+  GPUFLAGS += -std=c++17 # need CUDA >= 11.2 (see #333): this is enforced in mgOnGpuConfig.h
   # Without -maxrregcount: baseline throughput: 6.5E8 (16384 32 12) up to 7.3E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 160 # improves throughput: 6.9E8 (16384 32 12) up to 7.7E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 128 # improves throughput: 7.3E8 (16384 32 12) up to 7.6E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 96 # degrades throughput: 4.1E8 (16384 32 12) up to 4.5E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 64 # degrades throughput: 1.7E8 (16384 32 12) flat at 1.7E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 160 # improves throughput: 6.9E8 (16384 32 12) up to 7.7E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 128 # improves throughput: 7.3E8 (16384 32 12) up to 7.6E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 96 # degrades throughput: 4.1E8 (16384 32 12) up to 4.5E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 64 # degrades throughput: 1.7E8 (16384 32 12) flat at 1.7E8 (65536 128 12)
+  CUBUILDRULEFLAGS = -Xcompiler -fPIC -c
+  CCBUILDRULEFLAGS = -Xcompiler -fPIC -c -x cu
+  CUDATESTFLAGS = -lcuda
+
+  # Set the host C++ compiler for GPUCC via "-ccbin <host-compiler>"
+  # (NB issue #505: this must be a single word, "clang++ --gcc-toolchain..." is not supported)
+  GPUFLAGS += -ccbin $(shell which $(subst ccache ,,$(CXX)))
+
+  # Allow newer (unsupported) C++ compilers with older versions of CUDA if ALLOW_UNSUPPORTED_COMPILER_IN_CUDA is set (#504)
+  ifneq ($(origin ALLOW_UNSUPPORTED_COMPILER_IN_CUDA),undefined)
+  GPUFLAGS += -allow-unsupported-compiler
+  endif
+
 else ifneq ($(origin REQUIRE_CUDA),undefined)
+
   # If REQUIRE_CUDA is set but no cuda is found, stop here (e.g. for CI tests on GPU #443)
-  $(error No cuda installation found (set CUDA_HOME or make nvcc visible in PATH))
+  $(error No cuda installation found (set CUDA_HOME or make GPUCC visible in PATH))
+
+#--- Option 2: CUDA does not exist, HIP exists -> use HIP
+
+# Set GPUCC as $(HIP_HOME)/bin/hipcc if it exists
+else ifneq ($(wildcard $(HIP_HOME)/bin/hipcc),)
+
+  GPUCC = $(HIP_HOME)/bin/hipcc
+  #USE_NVTX ?=-DUSE_NVTX # should maybe find something equivalent to this in HIP?
+  HIPARCHFLAGS = -target x86_64-linux-gnu --offload-arch=gfx90a
+  HIPINC = -I$(HIP_HOME)/include/
+  # Note: -DHIP_FAST_MATH is equivalent to -use_fast_math in HIP 
+  # (but only for single precision line 208: https://rocm-developer-tools.github.io/HIP/hcc__detail_2math__functions_8h_source.html)
+  # Note: CUOPTFLAGS should not be used for HIP, it had been added here but was then removed (#808)
+  GPUFLAGS = $(OPTFLAGS) $(INCFLAGS) $(HIPINC) $(HIPARCHFLAGS) -DHIP_FAST_MATH -DHIP_PLATFORM=amd -fPIC
+  ###GPUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
+  GPUFLAGS += -std=c++17
+  ###GPUFLAGS+= --maxrregcount 255 # (AV: is this option valid on HIP and meaningful on AMD GPUs?)
+  CUBUILDRULEFLAGS = -fPIC -c
+  CCBUILDRULEFLAGS = -fPIC -c -x hip
+
+else ifneq ($(origin REQUIRE_HIP),undefined)
+
+  # If REQUIRE_HIP is set but no HIP is found, stop here (e.g. for CI tests on GPU #443)
+  $(error No hip installation found (set HIP_HOME or make GPUCC visible in PATH))
+
+#--- Option 3: CUDA does not exist, HIP does not exist -> switch off both CUDA and HIP
+
 else
-  # No cuda. Switch cuda compilation off and go to common random numbers in C++
+
+  # No cudacc and no hipcc: switch CUDA and HIP compilation off and go to common random numbers in C++
   $(warning CUDA_HOME is not set or is invalid: export CUDA_HOME to compile with cuda)
-  override NVCC=
+  $(warning HIP_HOME is not set or is invalid: export HIP_HOME to compile with hip)
+  override GPUCC=
   override USE_NVTX=
   override CUINC=
   override CURANDLIBFLAGS=
-endif
-export NVCC
-export CUFLAGS
 
-# Set the host C++ compiler for nvcc via "-ccbin <host-compiler>"
-# (NB issue #505: this must be a single word, "clang++ --gcc-toolchain..." is not supported)
-CUFLAGS += -ccbin $(shell which $(subst ccache ,,$(CXX)))
-
-# Allow newer (unsupported) C++ compilers with older versions of CUDA if ALLOW_UNSUPPORTED_COMPILER_IN_CUDA is set (#504)
-ifneq ($(origin ALLOW_UNSUPPORTED_COMPILER_IN_CUDA),undefined)
-CUFLAGS += -allow-unsupported-compiler
 endif
 
+# Export GPUCC (so that it can also be used in cudacpp_src.mk?)
+export GPUCC
+export GPUFLAGS
+
 #-------------------------------------------------------------------------------
 
-#=== Configure ccache for C++ and CUDA builds
+#=== Configure ccache for C++ and CUDA/HIP builds
 
 # Enable ccache if USECCACHE=1
 ifeq ($(USECCACHE)$(shell echo $(CXX) | grep ccache),1)
@@ -201,15 +260,15 @@ endif
 #ifeq ($(USECCACHE)$(shell echo $(AR) | grep ccache),1)
 #  override AR:=ccache $(AR)
 #endif
-ifneq ($(NVCC),)
-  ifeq ($(USECCACHE)$(shell echo $(NVCC) | grep ccache),1)
-    override NVCC:=ccache $(NVCC)
+ifneq ($(GPUCC),)
+  ifeq ($(USECCACHE)$(shell echo $(GPUCC) | grep ccache),1)
+    override GPUCC:=ccache $(GPUCC)
   endif
 endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure PowerPC-specific compiler flags for C++ and CUDA
+#=== Configure PowerPC-specific compiler flags for C++ and CUDA/HIP
 
 # PowerPC-specific CXX compiler flags (being reviewed)
 ifeq ($(UNAME_P),ppc64le)
@@ -225,9 +284,9 @@ else
   ######CXXFLAGS+= -fno-semantic-interposition # no benefit (neither alone, nor combined with -flto)
 endif
 
-# PowerPC-specific CUDA compiler flags (to be reviewed!)
+# PowerPC-specific CUDA/HIP compiler flags (to be reviewed!)
 ifeq ($(UNAME_P),ppc64le)
-  CUFLAGS+= -Xcompiler -mno-float128
+  GPUFLAGS+= -Xcompiler -mno-float128
 endif
 
 #-------------------------------------------------------------------------------
@@ -235,9 +294,11 @@ endif
 #=== Configure defaults and check if user-defined choices exist for OMPFLAGS, AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
 
 # Set the default OMPFLAGS choice
-ifneq ($(shell $(CXX) --version | egrep '^Intel'),)
+ifneq ($(findstring hipcc,$(GPUCC)),)
+override OMPFLAGS = # disable OpenMP MT when using hipcc #802
+else ifneq ($(shell $(CXX) --version | egrep '^Intel'),)
 override OMPFLAGS = -fopenmp
-###override OMPFLAGS = # disable OpenMP MT on Intel (was ok without nvcc but not ok with nvcc before #578)
+###override OMPFLAGS = # disable OpenMP MT on Intel (was ok without GPUCC but not ok with GPUCC before #578)
 else ifneq ($(shell $(CXX) --version | egrep '^(clang)'),)
 override OMPFLAGS = -fopenmp
 ###override OMPFLAGS = # disable OpenMP MT on clang (was not ok without or with nvcc before #578)
@@ -293,7 +354,10 @@ endif
 
 # Set the default RNDGEN (random number generator) choice
 ifeq ($(RNDGEN),)
-  ifeq ($(NVCC),)
+  ifeq ($(GPUCC),)
+    override RNDGEN = hasNoCurand
+  # Edgecase for HIP compilation
+  else ifeq ($(findstring hipcc,$(GPUCC)),hipcc)
     override RNDGEN = hasNoCurand
   else ifeq ($(RNDGEN),)
     override RNDGEN = hasCurand
@@ -310,7 +374,7 @@ export OMPFLAGS
 
 #-------------------------------------------------------------------------------
 
-#=== Set the CUDA/C++ compiler flags appropriate to user-defined choices of AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
+#=== Set the CUDA/HIP/C++ compiler flags appropriate to user-defined choices of AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
 
 # Set the build flags appropriate to OMPFLAGS
 $(info OMPFLAGS=$(OMPFLAGS))
@@ -368,13 +432,13 @@ CXXFLAGS+= $(AVXFLAGS)
 $(info FPTYPE=$(FPTYPE))
 ifeq ($(FPTYPE),d)
   CXXFLAGS += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
-  CUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
+  GPUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
 else ifeq ($(FPTYPE),f)
   CXXFLAGS += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
-  CUFLAGS  += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
+  GPUFLAGS  += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
 else ifeq ($(FPTYPE),m)
   CXXFLAGS += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
-  CUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
+  GPUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
 else
   $(error Unknown FPTYPE='$(FPTYPE)': only 'd', 'f' and 'm' are supported)
 endif
@@ -383,7 +447,7 @@ endif
 $(info HELINL=$(HELINL))
 ifeq ($(HELINL),1)
   CXXFLAGS += -DMGONGPU_INLINE_HELAMPS
-  CUFLAGS  += -DMGONGPU_INLINE_HELAMPS
+  GPUFLAGS  += -DMGONGPU_INLINE_HELAMPS
 else ifneq ($(HELINL),0)
   $(error Unknown HELINL='$(HELINL)': only '0' and '1' are supported)
 endif
@@ -392,7 +456,7 @@ endif
 $(info HRDCOD=$(HRDCOD))
 ifeq ($(HRDCOD),1)
   CXXFLAGS += -DMGONGPU_HARDCODE_PARAM
-  CUFLAGS  += -DMGONGPU_HARDCODE_PARAM
+  GPUFLAGS  += -DMGONGPU_HARDCODE_PARAM
 else ifneq ($(HRDCOD),0)
   $(error Unknown HRDCOD='$(HRDCOD)': only '0' and '1' are supported)
 endif
@@ -444,11 +508,11 @@ ifeq ($(UNAME_S),Darwin)
   override CULIBFLAGSRPATH2 =
 else
   # RPATH to cuda/cpp libs when linking executables
-  override CXXLIBFLAGSRPATH = -Wl,-rpath,$(LIBDIRRPATH)
-  override CULIBFLAGSRPATH = -Xlinker -rpath,$(LIBDIRRPATH)
+  override CXXLIBFLAGSRPATH = -Wl,-rpath=$(LIBDIRRPATH)
+  override CULIBFLAGSRPATH = -Xlinker -rpath=$(LIBDIRRPATH)
   # RPATH to common lib when linking cuda/cpp libs
-  override CXXLIBFLAGSRPATH2 = -Wl,-rpath,'$$ORIGIN'
-  override CULIBFLAGSRPATH2 = -Xlinker -rpath,'$$ORIGIN'
+  override CXXLIBFLAGSRPATH2 = -Wl,-rpath='$$ORIGIN'
+  override CULIBFLAGSRPATH2 = -Xlinker -rpath='$$ORIGIN'
 endif
 
 # Setting LD_LIBRARY_PATH or DYLD_LIBRARY_PATH in the RUNTIME is no longer necessary (neither on Linux nor on Mac)
@@ -461,7 +525,7 @@ override RUNTIME =
 cxx_main=$(BUILDDIR)/check.exe
 fcxx_main=$(BUILDDIR)/fcheck.exe
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 cu_main=$(BUILDDIR)/gcheck.exe
 fcu_main=$(BUILDDIR)/fgcheck.exe
 else
@@ -492,14 +556,15 @@ $(BUILDDIR)/.build.$(TAG):
 	@touch $(BUILDDIR)/.build.$(TAG)
 
 # Generic target and build rules: objects from CUDA compilation
-ifneq ($(NVCC),)
+# NB: CCBUILDRULEFLAGS includes "-x cu" for nvcc and "-x hip" for hipcc (#810)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/%.o : %.cu *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $(CUBUILDRULEFLAGS) $< -o $@
 
 $(BUILDDIR)/%_cu.o : %.cc *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c -x cu $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $(CCBUILDRULEFLAGS) $< -o $@
 endif
 
 # Generic target and build rules: objects from C++ compilation
@@ -509,11 +574,14 @@ $(BUILDDIR)/%.o : %.cc *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	$(CXX) $(CPPFLAGS) $(CXXFLAGS) -fPIC -c $< -o $@
 
 # Apply special build flags only to CrossSectionKernel.cc and gCrossSectionKernel.cu (no fast math, see #117 and #516)
+# Added edgecase for HIP compilation
 ifeq ($(shell $(CXX) --version | grep ^nvc++),)
 $(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS := $(filter-out -ffast-math,$(CXXFLAGS))
 $(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS += -fno-fast-math
-ifneq ($(NVCC),)
-$(BUILDDIR)/gCrossSectionKernels.o: CUFLAGS += -Xcompiler -fno-fast-math
+ifeq ($(findstring nvcc,$(GPUCC)),nvcc)
+  $(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -Xcompiler -fno-fast-math
+else
+  $(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -fno-fast-math
 endif
 endif
 
@@ -530,10 +598,10 @@ ifeq ($(RNDGEN),hasCurand)
 $(BUILDDIR)/CurandRandomNumberKernel.o: CXXFLAGS += $(CUINC)
 endif
 
-# Avoid "warning: builtin __has_trivial_... is deprecated; use __is_trivially_... instead" in nvcc with icx2023 (#592)
+# Avoid "warning: builtin __has_trivial_... is deprecated; use __is_trivially_... instead" in GPUCC with icx2023 (#592)
 ifneq ($(shell $(CXX) --version | egrep '^(Intel)'),)
-ifneq ($(NVCC),)
-CUFLAGS += -Xcompiler -Wno-deprecated-builtins
+ifneq ($(GPUCC),)
+GPUFLAGS += -Wno-deprecated-builtins
 endif
 endif
 
@@ -541,8 +609,8 @@ endif
 # This patch does remove the warning, but I prefer to keep it disabled for the moment...
 ###ifneq ($(shell $(CXX) --version | egrep '^(clang|Apple clang|Intel)'),)
 ###$(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS += -Wno-overriding-t-option
-###ifneq ($(NVCC),)
-###$(BUILDDIR)/gCrossSectionKernels.o: CUFLAGS += -Xcompiler -Wno-overriding-t-option
+###ifneq ($(GPUCC),)
+###$(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -Xcompiler -Wno-overriding-t-option
 ###endif
 ###endif
 
@@ -569,7 +637,7 @@ MG5AMC_CXXLIB = mg5amc_$(processid_short)_cpp
 cxx_objects_lib=$(BUILDDIR)/CPPProcess.o $(BUILDDIR)/MatrixElementKernels.o $(BUILDDIR)/BridgeKernels.o $(BUILDDIR)/CrossSectionKernels.o
 cxx_objects_exe=$(BUILDDIR)/CommonRandomNumberKernel.o $(BUILDDIR)/RamboSamplingKernels.o
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 MG5AMC_CULIB = mg5amc_$(processid_short)_cuda
 cu_objects_lib=$(BUILDDIR)/gCPPProcess.o $(BUILDDIR)/gMatrixElementKernels.o $(BUILDDIR)/gBridgeKernels.o $(BUILDDIR)/gCrossSectionKernels.o
 cu_objects_exe=$(BUILDDIR)/gCommonRandomNumberKernel.o $(BUILDDIR)/gRamboSamplingKernels.o
@@ -581,11 +649,17 @@ $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: cxx_objects_lib += $(BUILDDIR)/fbridge.o
 $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib)
 	$(CXX) -shared -o $@ $(cxx_objects_lib) $(CXXLIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: $(BUILDDIR)/fbridge_cu.o
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: cu_objects_lib += $(BUILDDIR)/fbridge_cu.o
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cu_objects_lib)
-	$(NVCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+# Bypass std::filesystem completely to ease portability on LUMI #803
+#ifneq ($(findstring hipcc,$(GPUCC)),)
+#	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB) -lstdc++fs
+#else
+#	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+#endif
 endif
 
 #-------------------------------------------------------------------------------
@@ -602,16 +676,16 @@ $(cxx_main): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PAT
 $(cxx_main): $(BUILDDIR)/check_sa.o $(LIBDIR)/lib$(MG5AMC_CXXLIB).so $(cxx_objects_exe) $(BUILDDIR)/CurandRandomNumberKernel.o
 	$(CXX) -o $@ $(BUILDDIR)/check_sa.o $(OMPFLAGS) -ldl -pthread $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe) $(BUILDDIR)/CurandRandomNumberKernel.o $(CURANDLIBFLAGS)
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(cu_main): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(cu_main): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(cu_main): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(cu_main): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 else ifneq ($(shell $(CXX) --version | grep ^nvc++),) # support nvc++ #531
 $(cu_main): LIBFLAGS += -L$(patsubst %bin/nvc++,%lib,$(subst ccache ,,$(CXX))) -lnvhpcatm -lnvcpumath -lnvc
 endif
 $(cu_main): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(cu_main): $(BUILDDIR)/gcheck_sa.o $(LIBDIR)/lib$(MG5AMC_CULIB).so $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o
-	$(NVCC) -o $@ $(BUILDDIR)/gcheck_sa.o $(CUARCHFLAGS) $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o $(CURANDLIBFLAGS)
+	$(GPUCC) -o $@ $(BUILDDIR)/gcheck_sa.o $(CUARCHFLAGS) $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o $(CURANDLIBFLAGS)
 endif
 
 #-------------------------------------------------------------------------------
@@ -635,19 +709,27 @@ $(fcxx_main): LIBFLAGS += -L$(shell dirname $(shell $(FC) --print-file-name libg
 endif
 $(fcxx_main): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(fcxx_main): $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler.o $(LIBDIR)/lib$(MG5AMC_CXXLIB).so $(cxx_objects_exe)
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(BUILDDIR)/fcheck_sa.o $(OMPFLAGS) $(BUILDDIR)/fsampler.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe) -lstdc++
+else
 	$(CXX) -o $@ $(BUILDDIR)/fcheck_sa.o $(OMPFLAGS) $(BUILDDIR)/fsampler.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe)
+endif
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(fcu_main): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(fcu_main): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(fcu_main): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(fcu_main): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 endif
 ifeq ($(UNAME_S),Darwin)
 $(fcu_main): LIBFLAGS += -L$(shell dirname $(shell $(FC) --print-file-name libgfortran.dylib)) # add path to libgfortran on Mac #375
 endif
 $(fcu_main): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(fcu_main): $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBDIR)/lib$(MG5AMC_CULIB).so $(cu_objects_exe)
-	$(NVCC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe)
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) -lstdc++ -L$(shell dirname $(shell $(GPUCC) -print-prog-name=clang))/../../lib -lamdhip64
+else
+	$(GPUCC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe)
+endif
 endif
 
 #-------------------------------------------------------------------------------
@@ -659,7 +741,7 @@ $(BUILDDIR)/testxxx.o: testxxx_cc_ref.txt
 $(testmain): $(BUILDDIR)/testxxx.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/testxxx.o # Comment out this line to skip the C++ test of xxx functions
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/testxxx_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/testxxx_cu.o: INCFLAGS += $(GTESTINC)
 $(BUILDDIR)/testxxx_cu.o: testxxx_cc_ref.txt
@@ -672,7 +754,7 @@ $(BUILDDIR)/testmisc.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/testmisc.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/testmisc.o # Comment out this line to skip the C++ miscellaneous tests
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/testmisc_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/testmisc_cu.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/testmisc_cu.o
@@ -684,12 +766,12 @@ $(BUILDDIR)/runTest.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/runTest.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/runTest.o
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/runTest_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/runTest_cu.o: INCFLAGS += $(GTESTINC)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(testmain): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(testmain): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(testmain): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(testmain): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 else ifneq ($(shell $(CXX) --version | grep ^nvc++),) # support nvc++ #531
 $(testmain): LIBFLAGS += -L$(patsubst %bin/nvc++,%lib,$(subst ccache ,,$(CXX))) -lnvhpcatm -lnvcpumath -lnvc
 endif
@@ -713,14 +795,23 @@ $(testmain): LIBFLAGS += -lgomp
 endif
 endif
 
-ifeq ($(NVCC),) # link only runTest.o
+# Bypass std::filesystem completely to ease portability on LUMI #803
+#ifneq ($(findstring hipcc,$(GPUCC)),)
+#$(testmain): LIBFLAGS += -lstdc++fs
+#endif
+
+ifeq ($(GPUCC),) # link only runTest.o
 $(testmain): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(testmain): $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib) $(cxx_objects_exe) $(GTESTLIBS)
 	$(CXX) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) -ldl -pthread $(LIBFLAGS)
 else # link both runTest.o and runTest_cu.o
 $(testmain): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(testmain): $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) $(GTESTLIBS)
-	$(NVCC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) -lcuda
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) $(CUDATESTFLAGS) -lstdc++ -lpthread  -L$(shell dirname $(shell $(GPUCC) -print-prog-name=clang))/../../lib -lamdhip64
+else
+	$(GPUCC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) $(CUDATESTFLAGS)
+endif
 endif
 
 # Use target gtestlibs to build only googletest
@@ -829,9 +920,9 @@ ifeq ($(USECCACHE),1)
 	ccache --version | head -1
 endif
 	@echo ""
-	@echo NVCC=$(NVCC)
-ifneq ($(NVCC),)
-	$(NVCC) --version
+	@echo GPUCC=$(GPUCC)
+ifneq ($(GPUCC),)
+	$(GPUCC) --version
 endif
 	@echo ""
 	@echo CXX=$(CXX)
@@ -850,7 +941,7 @@ endif
 
 # Target: check (run the C++ test executable)
 # [NB THIS IS WHAT IS USED IN THE GITHUB CI!]
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 check: runTest cmpFcheck cmpFGcheck
 else
 check: runTest cmpFcheck
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/fbridge.cc b/epochX/cudacpp/gg_ttg.sa/SubProcesses/fbridge.cc
index 2d2b36d560..22ce3f5115 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/fbridge.cc
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/fbridge.cc
@@ -1,11 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Roiser (Oct 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "Bridge.h"
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 
 extern "C"
 {
@@ -22,7 +22,7 @@ extern "C"
    * Using the same Fortran MadEvent code, linking to the hetrerogeneous library would allow access to both CPU and GPU implementations.
    * The specific heterogeneous configuration (how many GPUs, how many threads on each CPU, etc) could be loaded in CUDA/C++ from a data file.
    */
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -46,8 +46,8 @@ extern "C"
    */
   void fbridgecreate_( CppObjectInFortran** ppbridge, const int* pnevtF, const int* pnparF, const int* pnp4F )
   {
-#ifdef __CUDACC__
-    CudaRuntime::setUp();
+#ifdef MGONGPUCPP_GPUIMPL
+    GpuRuntime::setUp();
 #endif
     // (NB: CPPProcess::initProc no longer needs to be executed here because it is called in the Bridge constructor)
     // FIXME: disable OMP in Bridge when called from Fortran
@@ -65,8 +65,8 @@ extern "C"
     Bridge<FORTRANFPTYPE>* pbridge = dynamic_cast<Bridge<FORTRANFPTYPE>*>( *ppbridge );
     if( pbridge == 0 ) throw std::runtime_error( "fbridgedelete_: invalid Bridge address" );
     delete pbridge;
-#ifdef __CUDACC__
-    CudaRuntime::tearDown();
+#ifdef MGONGPUCPP_GPUIMPL
+    GpuRuntime::tearDown();
 #endif
   }
 
@@ -96,7 +96,7 @@ extern "C"
   {
     Bridge<FORTRANFPTYPE>* pbridge = dynamic_cast<Bridge<FORTRANFPTYPE>*>( *ppbridge );
     if( pbridge == 0 ) throw std::runtime_error( "fbridgesequence_: invalid Bridge address" );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Use the device/GPU implementation in the CUDA library
     // (there is also a host implementation in this library)
     pbridge->gpu_sequence( momenta, gs, rndhel, rndcol, *pchannelId, mes, selhel, selcol );
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/fsampler.cc b/epochX/cudacpp/gg_ttg.sa/SubProcesses/fsampler.cc
index 2fb445372d..3743934f41 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/fsampler.cc
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/fsampler.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Feb 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "mgOnGpuConfig.h"
 
@@ -13,7 +13,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -40,7 +40,7 @@ namespace mg5amcCpu
   private:
     const int m_nevt; // The number of events in each iteration
     int m_iiter;      // The iteration counter (for random number seeding)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     HostBufferRndNumMomenta m_hstRndmom; // Memory buffers for random numbers
     HostBufferMomenta m_hstMomenta;      // Memory buffers for momenta
     HostBufferWeights m_hstWeights;      // Memory buffers for sampling weights
@@ -105,7 +105,7 @@ namespace mg5amcCpu
 
 extern "C"
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/runTest.cc b/epochX/cudacpp/gg_ttg.sa/SubProcesses/runTest.cc
index d4a760a71b..de327f2321 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/runTest.cc
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/runTest.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Hageboeck (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*xxx to run only testxxx.cc tests
 //----------------------------------------------------------------------------
@@ -18,7 +18,7 @@
 #include "RandomNumberKernels.h"
 #include "epoch_process_id.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using namespace mg5amcGpu;
 #else
 using namespace mg5amcCpu;
@@ -35,7 +35,7 @@ struct CUDA_CPU_TestBase : public TestDriverBase
     : TestDriverBase( npar, refFileName ) {}
 };
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 struct CPUTest : public CUDA_CPU_TestBase
 {
   // Struct data members (process, and memory structures for random numbers, momenta, matrix elements and weights on host and device)
@@ -119,7 +119,7 @@ struct CPUTest : public CUDA_CPU_TestBase
 };
 #endif
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 struct CUDATest : public CUDA_CPU_TestBase
 {
   // Reset the device when our test goes out of scope. Note that this should happen after
@@ -128,7 +128,7 @@ struct CUDATest : public CUDA_CPU_TestBase
   {
     ~DeviceReset()
     {
-      checkCuda( cudaDeviceReset() ); // this is needed by cuda-memcheck --leak-check full
+      checkGpu( gpuDeviceReset() ); // this is needed by cuda-memcheck --leak-check full
     }
   } deviceResetter;
 
@@ -256,7 +256,7 @@ INSTANTIATE_TEST_SUITE_P( prefix, \
                           test_suite_name, \
                           testing::Values( new CUDATest( MG_EPOCH_REFERENCE_FILE_NAME ) ) );
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 MG_INSTANTIATE_TEST_SUITE_GPU( XTESTID_GPU( MG_EPOCH_PROCESS_ID ), MadgraphTest );
 #else
 MG_INSTANTIATE_TEST_SUITE_CPU( XTESTID_CPU( MG_EPOCH_PROCESS_ID ), MadgraphTest );
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/testmisc.cc b/epochX/cudacpp/gg_ttg.sa/SubProcesses/testmisc.cc
index 895d6eeb56..ba9e59a8a3 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/testmisc.cc
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/testmisc.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*misc to run only testmisc.cc tests
 //----------------------------------------------------------------------------
@@ -17,7 +17,7 @@
 #include <sstream>
 #include <typeinfo>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #define TESTID( s ) s##_GPU_MISC
 #else
 #define TESTID( s ) s##_CPU_MISC
@@ -26,7 +26,7 @@
 #define XTESTID( s ) TESTID( s )
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -59,7 +59,7 @@ namespace mg5amcCpu
 
 TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testmisc )
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_ttg.sa/SubProcesses/testxxx.cc b/epochX/cudacpp/gg_ttg.sa/SubProcesses/testxxx.cc
index 3361fe5aa9..e5167de00c 100644
--- a/epochX/cudacpp/gg_ttg.sa/SubProcesses/testxxx.cc
+++ b/epochX/cudacpp/gg_ttg.sa/SubProcesses/testxxx.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Apr 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*xxx to run only testxxx.cc tests
 //----------------------------------------------------------------------------
@@ -24,7 +24,7 @@
 #include <iomanip>
 #include <iostream>
 #include <vector>
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #define TESTID( s ) s##_GPU_XXX
 #else
 #define TESTID( s ) s##_CPU_XXX
@@ -32,7 +32,7 @@
 
 #define XTESTID( s ) TESTID( s )
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -42,7 +42,7 @@ namespace mg5amcCpu
   int FPEhandlerIevt = -1;
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU): '" << FPEhandlerMessage << "' ievt=" << FPEhandlerIevt << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU neppV=" << neppV << "): '" << FPEhandlerMessage << "' ievt=" << FPEhandlerIevt << std::endl;
@@ -53,7 +53,7 @@ namespace mg5amcCpu
 
 TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -77,7 +77,7 @@ TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
   assert( nevt % neppM == 0 ); // nevt must be a multiple of neppM
   assert( nevt % neppV == 0 ); // nevt must be a multiple of neppV
   // Fill in the input momenta
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   mg5amcGpu::PinnedHostBufferMomenta hstMomenta( nevt ); // AOSOA[npagM][npar=4][np4=4][neppM]
 #else
   mg5amcCpu::HostBufferMomenta hstMomenta( nevt ); // AOSOA[npagM][npar=4][np4=4][neppM]
@@ -322,7 +322,7 @@ TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
   {
     for( int ievt = 0; ievt < nevt; ievt++ )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       using namespace mg5amcGpu;
 #else
       using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_ttg.sa/src/HelAmps_sm.h b/epochX/cudacpp/gg_ttg.sa/src/HelAmps_sm.h
index 361b488401..0dd0f3ebba 100644
--- a/epochX/cudacpp/gg_ttg.sa/src/HelAmps_sm.h
+++ b/epochX/cudacpp/gg_ttg.sa/src/HelAmps_sm.h
@@ -5,7 +5,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -28,7 +28,7 @@
 //#include <iomanip>
 //#include <iostream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttg.sa/src/Parameters_sm.cc b/epochX/cudacpp/gg_ttg.sa/src/Parameters_sm.cc
index 64fc3fea62..067445b198 100644
--- a/epochX/cudacpp/gg_ttg.sa/src/Parameters_sm.cc
+++ b/epochX/cudacpp/gg_ttg.sa/src/Parameters_sm.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -17,7 +17,7 @@
 #include <iomanip>
 #include <iostream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using namespace mg5amcGpu;
 #else
 using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_ttg.sa/src/Parameters_sm.h b/epochX/cudacpp/gg_ttg.sa/src/Parameters_sm.h
index b6568d3761..9581d66e0e 100644
--- a/epochX/cudacpp/gg_ttg.sa/src/Parameters_sm.h
+++ b/epochX/cudacpp/gg_ttg.sa/src/Parameters_sm.h
@@ -27,7 +27,7 @@
 #include "read_slha.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -93,7 +93,7 @@ namespace mg5amcCpu
 #include <limits>
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -218,7 +218,7 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -239,7 +239,7 @@ namespace mg5amcCpu
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wunused-variable"  // e.g. <<warning: unused variable ‘mdl_G__exp__2’ [-Wunused-variable]>>
 #pragma GCC diagnostic ignored "-Wunused-parameter" // e.g. <<warning: unused parameter ‘G’ [-Wunused-parameter]>>
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #pragma nv_diagnostic push
 #pragma nv_diag_suppress 177 // e.g. <<warning #177-D: variable "mdl_G__exp__2" was declared but never referenced>>
 #endif
@@ -267,7 +267,7 @@ namespace mg5amcCpu
       // End SM implementation - no special handling of vectors of floats as in EFT (#439)
       return out;
     }
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #pragma GCC diagnostic pop
 #pragma nv_diagnostic pop
 #endif
diff --git a/epochX/cudacpp/gg_ttg.sa/src/cudacpp_src.mk b/epochX/cudacpp/gg_ttg.sa/src/cudacpp_src.mk
index d4cc628aec..b2b9da5288 100644
--- a/epochX/cudacpp/gg_ttg.sa/src/cudacpp_src.mk
+++ b/epochX/cudacpp/gg_ttg.sa/src/cudacpp_src.mk
@@ -19,7 +19,7 @@ SHELL := /bin/bash
 #=== Configure common compiler flags for CUDA and C++
 
 INCFLAGS = -I.
-OPTFLAGS = -O3 # this ends up in CUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
+OPTFLAGS = -O3 # this ends up in GPUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
 
 #-------------------------------------------------------------------------------
 
@@ -45,13 +45,13 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure the CUDA compiler (note: NVCC is already exported including ccache)
+#=== Configure the CUDA compiler (note: GPUCC is already exported including ccache)
 
-###$(info NVCC=$(NVCC))
+###$(info GPUCC=$(GPUCC))
 
 #-------------------------------------------------------------------------------
 
-#=== Configure ccache for C++ builds (note: NVCC is already exported including ccache)
+#=== Configure ccache for C++ builds (note: GPUCC is already exported including ccache)
 
 # Enable ccache if USECCACHE=1
 ifeq ($(USECCACHE)$(shell echo $(CXX) | grep ccache),1)
@@ -92,6 +92,13 @@ endif
 ###$(info OMPFLAGS=$(OMPFLAGS))
 CXXFLAGS += $(OMPFLAGS)
 
+# Add correct flags for nvcc (-x cu) and hipcc (-x hip) for GPU code (see #810)
+ifeq ($(findstring nvcc,$(GPUCC)),nvcc)
+  GPUFLAGS += -Xcompiler -fPIC -c -x cu
+else ifeq ($(findstring hipcc,$(GPUCC)),hipcc)
+  GPUFLAGS += -fPIC -c -x hip
+endif
+
 # Set the build flags appropriate to each AVX choice (example: "make AVX=none")
 # [NB MGONGPU_PVW512 is needed because "-mprefer-vector-width=256" is not exposed in a macro]
 # [See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96476]
@@ -253,20 +260,20 @@ $(BUILDDIR)/%.o : %.cc *.h $(BUILDDIR)/.build.$(TAG)
 # Generic target and build rules: objects from CUDA compilation
 $(BUILDDIR)/%_cu.o : %.cc *.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c -x cu $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $< -o $@
 
 #-------------------------------------------------------------------------------
 
 cxx_objects=$(addprefix $(BUILDDIR)/, Parameters_sm.o read_slha.o)
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 cu_objects=$(addprefix $(BUILDDIR)/, Parameters_sm_cu.o)
 endif
 
 # Target (and build rules): common (src) library
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so : $(cxx_objects) $(cu_objects)
 	@if [ ! -d $(LIBDIR) ]; then echo "mkdir -p $(LIBDIR)"; mkdir -p $(LIBDIR); fi
-	$(NVCC) -shared -o $@ $(cxx_objects) $(cu_objects) $(LDFLAGS)
+	$(GPUCC) -shared -o $@ $(cxx_objects) $(cu_objects) $(LDFLAGS)
 else
 $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so : $(cxx_objects)
 	@if [ ! -d $(LIBDIR) ]; then echo "mkdir -p $(LIBDIR)"; mkdir -p $(LIBDIR); fi
diff --git a/epochX/cudacpp/gg_ttg.sa/src/mgOnGpuConfig.h b/epochX/cudacpp/gg_ttg.sa/src/mgOnGpuConfig.h
index b247654dcf..475749ca7c 100644
--- a/epochX/cudacpp/gg_ttg.sa/src/mgOnGpuConfig.h
+++ b/epochX/cudacpp/gg_ttg.sa/src/mgOnGpuConfig.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jul 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUCONFIG_H
 #define MGONGPUCONFIG_H 1
@@ -10,12 +10,26 @@
 // There are two different code bases for standalone_cudacpp (without multichannel) and madevent+cudacpp (with multichannel)
 #undef MGONGPU_SUPPORTS_MULTICHANNEL
 
+// Is this a GPU (CUDA, HIP) or CPU implementation?
+#ifdef __CUDACC__
+#define MGONGPUCPP_GPUIMPL cuda
+#elif defined __HIPCC__
+#define MGONGPUCPP_GPUIMPL hip
+#include "hip/hip_runtime.h" // needed for blockDim, blockIdx, threadIdx: better in mgOnGpuConfig.h than in GpuAbstraction.h
+#else
+#undef MGONGPUCPP_GPUIMPL
+#endif
+
 // ** NB1 Throughputs (e.g. 6.8E8) are events/sec for "./gcheck.exe -p 65536 128 12"
 // ** NB2 Baseline on b7g47n0004 fluctuates (probably depends on load on other VMs)
 
 // Choose if curand is supported for generating random numbers
+// For HIP, by default, do not use curand (common random numbers will be used instead)
 // For both CUDA and C++, by default, do not inline, but allow this macro to be set from outside with e.g. -DMGONGPU_HAS_NO_CURAND
-// (there exist CUDA installations, e.g. using the HPC package, which do not include curand - see PR #784)
+// (there exist CUDA installations, e.g. using the HPC package, which do not include curand - see PR #784 and #785)
+#if defined __HIPCC__
+#define MGONGPU_HAS_NO_CURAND 1
+#else
 //#ifdef __CUDACC__
 //#undef MGONGPU_HAS_NO_CURAND // default
 ////#define MGONGPU_HAS_NO_CURAND 1
@@ -23,6 +37,7 @@
 //#undef MGONGPU_HAS_NO_CURAND // default
 ////#define MGONGPU_HAS_NO_CURAND 1
 //#endif
+#endif
 
 // Choose floating point precision (for everything but color algebra #537)
 // If one of these macros has been set from outside with e.g. -DMGONGPU_FPTYPE_FLOAT, nothing happens (issue #167)
@@ -54,23 +69,31 @@
 //#undef MGONGPU_HARDCODE_PARAM // default
 ////#define MGONGPU_HARDCODE_PARAM 1
 
-// Complex type in c++: std::complex or cxsmpl (CHOOSE ONLY ONE)
-#ifndef __CUDACC__
-//#define MGONGPU_CPPCXTYPE_STDCOMPLEX 1 // ~8 percent slower on float, same on double (5.1E6/double, 9.4E6/float)
-#define MGONGPU_CPPCXTYPE_CXSMPL 1 // new default (5.1E6/double, 10.2E6/float)
-#endif
-
-// Complex type in cuda: thrust or cucomplex or cxsmpl (CHOOSE ONLY ONE)
+// Complex type in CUDA: thrust or cucomplex or cxsmpl (CHOOSE ONLY ONE)
+// (NB THIS IS MGONGPU_*CU*CXTYPE_xxx)
 #ifdef __CUDACC__
 #define MGONGPU_CUCXTYPE_THRUST 1 // default (~1.15E9/double, ~3.2E9/float)
 //#define MGONGPU_CUCXTYPE_CUCOMPLEX 1 // ~10 percent slower (1.03E9/double, ~2.8E9/float)
 //#define MGONGPU_CUCXTYPE_CXSMPL 1 // ~10 percent slower (1.00E9/double, ~2.9E9/float)
+
+// Complex type in HIP: cxsmpl (ONLY ONE OPTION POSSIBLE? #810)
+// (NB THIS IS MGONGPU_*HIP*CXTYPE_xxx)
+#elif defined __HIPCC__
+#define MGONGPU_HIPCXTYPE_CXSMPL 1 // default for HIP
+
+// Complex type in C++: std::complex or cxsmpl (CHOOSE ONLY ONE)
+// (NB THIS IS MGONGPU_*CPP*CXTYPE_xxx)
+#else
+//#define MGONGPU_CPPCXTYPE_STDCOMPLEX 1 // ~8 percent slower on float, same on double (5.1E6/double, 9.4E6/float)
+#define MGONGPU_CPPCXTYPE_CXSMPL 1 // new default (5.1E6/double, 10.2E6/float)
 #endif
 
-// Cuda nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
+// CUDA nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
 #ifdef __CUDACC__
-#undef MGONGPU_NSIGHT_DEBUG // default
+#undef MGONGPU_NSIGHT_DEBUG // default in CUDA
 //#define MGONGPU_NSIGHT_DEBUG 1
+#else
+#undef MGONGPU_NSIGHT_DEBUG // only option in HIP or C++
 #endif
 
 // SANITY CHECKS (floating point precision for everything but color algebra #537)
@@ -86,17 +109,21 @@
 #error You cannot use double precision for color algebra and single precision elsewhere
 #endif
 
-// SANITY CHECKS (c++ complex number implementation)
-#ifndef __CUDACC__
-#if defined MGONGPU_CPPCXTYPE_STDCOMPLEX and defined MGONGPU_CPPCXTYPE_CXSMPL
-#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CPPCXTYPE_STDCOMPLEX or MGONGPU_CPPCXTYPE_CXSMPL
+// SANITY CHECKS (CUDA complex number implementation)
+#ifdef __CUDACC__
+#if defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CUCOMPLEX
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX for CUDA
+#elif defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CXSMPL for CUDA
+#elif defined MGONGPU_CUCXTYPE_CUCOMPLEX and defined MGONGPU_CUCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE OF MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL for CUDA
 #endif
 #endif
 
-// SANITY CHECKS (cuda complex number implementation)
-#ifdef __CUDACC__
-#if defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CUCOMPLEX and defined MGONGPU_CUCXTYPE_CXSMPL
-#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL
+// SANITY CHECKS (C++ complex number implementation)
+#ifndef MGONGPUCPP_GPUIMPL
+#if defined MGONGPU_CPPCXTYPE_STDCOMPLEX and defined MGONGPU_CPPCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CPPCXTYPE_STDCOMPLEX or MGONGPU_CPPCXTYPE_CXSMPL for C++
 #endif
 #endif
 
@@ -134,7 +161,7 @@ namespace mgOnGpu
   // Alignment requirement for using reinterpret_cast with SIMD vectorized code
   // (using reinterpret_cast with non aligned memory may lead to segmentation faults!)
   // Only needed for C++ code but can be enforced also in NVCC builds of C++ code using CUDA>=11.2 and C++17 (#318, #319, #333)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   constexpr int cppAlign = 64; // alignment requirement for SIMD vectorization (64-byte i.e. 512-bit)
 #endif
 
@@ -145,7 +172,7 @@ using mgOnGpu::fptype;
 using mgOnGpu::fptype2;
 
 // C++ SIMD vectorization width (this will be used to set neppV)
-#ifdef __CUDACC__ // CUDA implementation has no SIMD
+#ifdef MGONGPUCPP_GPUIMPL // CUDA and HIP implementations have no SIMD
 #undef MGONGPU_CPPSIMD
 #elif defined __AVX512VL__ && defined MGONGPU_PVW512 // C++ "512z" AVX512 with 512 width (512-bit ie 64-byte): 8 (DOUBLE) or 16 (FLOAT)
 #ifdef MGONGPU_FPTYPE_DOUBLE
@@ -175,9 +202,9 @@ using mgOnGpu::fptype2;
 #undef MGONGPU_CPPSIMD
 #endif
 
-// Cuda nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
+// CUDA nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
 // Arguments (not used so far): text is __FUNCTION__, code is 0 (start) or 1 (end)
-#if defined __CUDACC__ && defined MGONGPU_NSIGHT_DEBUG /* clang-format off */
+#if defined __CUDA__ && defined MGONGPU_NSIGHT_DEBUG /* clang-format off */
 #define mgDebugDeclare() __shared__ float mgDebugCounter[mgOnGpu::ntpbMAX];
 #define mgDebugInitialise() { mgDebugCounter[threadIdx.x] = 0; }
 #define mgDebug( code, text ) { mgDebugCounter[threadIdx.x] += 1; }
@@ -189,8 +216,8 @@ using mgOnGpu::fptype2;
 #define mgDebugFinalise() { /*noop*/ }
 #endif /* clang-format on */
 
-// Define empty CUDA declaration specifiers for C++
-#ifndef __CUDACC__
+// Define empty CUDA/HIP declaration specifiers for C++
+#ifndef MGONGPUCPP_GPUIMPL
 #define __global__
 #define __host__
 #define __device__
diff --git a/epochX/cudacpp/gg_ttg.sa/src/mgOnGpuCxtypes.h b/epochX/cudacpp/gg_ttg.sa/src/mgOnGpuCxtypes.h
index ca9a9f00c0..7ede1dbfae 100644
--- a/epochX/cudacpp/gg_ttg.sa/src/mgOnGpuCxtypes.h
+++ b/epochX/cudacpp/gg_ttg.sa/src/mgOnGpuCxtypes.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022, based on earlier work by D. Smith) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUCXTYPES_H
 #define MGONGPUCXTYPES_H 1
@@ -19,7 +19,7 @@
 #include <complex>
 
 // Complex type in cuda: thrust or cucomplex or cxsmpl
-#ifdef __CUDACC__
+#ifdef __CUDACC__ // this must be __CUDAC__ (not MGONGPUCPP_GPUIMPL)
 #if defined MGONGPU_CUCXTYPE_THRUST
 #pragma clang diagnostic push
 #pragma clang diagnostic ignored "-Wtautological-compare" // for icpx2021/clang13 (https://stackoverflow.com/a/15864661)
@@ -30,8 +30,13 @@
 #elif not defined MGONGPU_CUCXTYPE_CXSMPL
 #error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL
 #endif
+// Complex type in HIP: cxsmpl
+#elif defined __HIPCC__
+#if not defined MGONGPU_HIPCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_HIPCXTYPE_CXSMPL
+#endif
 #else
-// Complex type in c++: std::complex or cxsmpl
+// Complex type in c++ or HIP: std::complex or cxsmpl
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
 #include <cmath>
 #elif not defined MGONGPU_CPPCXTYPE_CXSMPL
@@ -82,7 +87,7 @@ namespace mgOnGpu /* clang-format off */
 using mgOnGpu::cxsmpl;
 
 // Printout to stream for user defined types
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -92,7 +97,7 @@ namespace mg5amcCpu
   inline __host__ std::ostream&
   operator<<( std::ostream& out, const cxsmpl<FP>& c )
   {
-    out << std::complex( c.real(), c.imag() );
+    out << std::complex<FP>( c.real(), c.imag() );
     return out;
   }
 
@@ -215,14 +220,14 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
   // --- Type definitions (complex type: cxtype)
-#ifdef __CUDACC__ // cuda
+#ifdef __CUDACC__ // this must be __CUDAC__ (not MGONGPUCPP_GPUIMPL)
 #if defined MGONGPU_CUCXTYPE_THRUST
   typedef thrust::complex<fptype> cxtype;
 #elif defined MGONGPU_CUCXTYPE_CUCOMPLEX
@@ -255,13 +260,13 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
-#if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+#if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
 
   //------------------------------
   // CUDA or C++ - using cxsmpl
@@ -303,11 +308,11 @@ namespace mg5amcCpu
     return cxmake( c.real(), c.imag() );
   }
 
-#endif // #if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+#endif // #if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
 
   //==========================================================================
 
-#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_THRUST // cuda + thrust
+#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_THRUST // cuda + thrust (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // CUDA - using thrust::complex
@@ -347,7 +352,7 @@ namespace mg5amcCpu
 
   //==========================================================================
 
-#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_CUCOMPLEX // cuda + cucomplex
+#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_CUCOMPLEX // cuda + cucomplex (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // CUDA - using cuComplex
@@ -566,7 +571,7 @@ namespace mg5amcCpu
 
   //==========================================================================
 
-#if not defined __CUDACC__ and defined MGONGPU_CPPCXTYPE_STDCOMPLEX // c++ + stdcomplex
+#if not defined __CUDACC__ and defined MGONGPU_CPPCXTYPE_STDCOMPLEX // c++/hip + stdcomplex (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // C++ - using std::complex
@@ -633,7 +638,7 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttg.sa/src/mgOnGpuFptypes.h b/epochX/cudacpp/gg_ttg.sa/src/mgOnGpuFptypes.h
index 905c97d700..fa3a02664b 100644
--- a/epochX/cudacpp/gg_ttg.sa/src/mgOnGpuFptypes.h
+++ b/epochX/cudacpp/gg_ttg.sa/src/mgOnGpuFptypes.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUFPTYPES_H
 #define MGONGPUFPTYPES_H 1
@@ -12,7 +12,7 @@
 #include <cmath>
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL // cuda
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -20,7 +20,7 @@ namespace mg5amcCpu
 {
   //==========================================================================
 
-#ifdef __CUDACC__ // cuda
+#ifdef MGONGPUCPP_GPUIMPL // cuda
 
   //------------------------------
   // Floating point types - Cuda
@@ -64,11 +64,11 @@ namespace mg5amcCpu
 #endif
   }
 
-#endif // #ifdef __CUDACC__
+#endif // #ifdef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
   //------------------------------
   // Floating point types - C++
@@ -92,7 +92,7 @@ namespace mg5amcCpu
     return std::sqrt( f );
   }
 
-#endif // #ifndef __CUDACC__
+#endif // #ifndef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
diff --git a/epochX/cudacpp/gg_ttg.sa/src/mgOnGpuVectors.h b/epochX/cudacpp/gg_ttg.sa/src/mgOnGpuVectors.h
index e1299ba81e..cdae04326b 100644
--- a/epochX/cudacpp/gg_ttg.sa/src/mgOnGpuVectors.h
+++ b/epochX/cudacpp/gg_ttg.sa/src/mgOnGpuVectors.h
@@ -32,7 +32,7 @@
 #endif
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -131,7 +131,7 @@ namespace mg5amcCpu
 #endif
 #endif
 
-#else // i.e #ifndef MGONGPU_CPPSIMD (this includes #ifdef __CUDACC__)
+#else // i.e #ifndef MGONGPU_CPPSIMD (this includes #ifdef MGONGPUCPP_GPUIMPL)
 
   const int neppV = 1;
 
@@ -153,13 +153,13 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
   // Printout to stream for user defined types
 
@@ -805,11 +805,11 @@ namespace mg5amcCpu
 
 #endif // #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
 
-#endif // #ifndef __CUDACC__
+#endif // #ifndef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
   //------------------------------
   // Vector types - CUDA
@@ -853,12 +853,12 @@ namespace mg5amcCpu
     return mask;
   }
 
-#endif // #ifdef __CUDACC__
+#endif // #ifdef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
   // Scalar-or-vector types: scalar in CUDA, vector or scalar in C++
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   typedef bool bool_sv;
   typedef fptype fptype_sv;
   typedef fptype2 fptype2_sv;
@@ -879,7 +879,7 @@ namespace mg5amcCpu
 #endif
 
   // Scalar-or-vector zeros: scalar in CUDA, vector or scalar in C++
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   inline __host__ __device__ cxtype cxzero_sv(){ return cxtype( 0, 0 ); }
 #elif defined MGONGPU_CPPSIMD
   inline cxtype_v cxzero_sv() { return cxtype_v(); } // RRRR=0000 IIII=0000
diff --git a/epochX/cudacpp/gg_ttg.sa/src/rambo.h b/epochX/cudacpp/gg_ttg.sa/src/rambo.h
index e02ea52496..cd7e1008ea 100644
--- a/epochX/cudacpp/gg_ttg.sa/src/rambo.h
+++ b/epochX/cudacpp/gg_ttg.sa/src/rambo.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -18,7 +18,7 @@
 #include <iostream>
 
 // Simplified rambo version for 2 to N (with N>=2) processes with massless particles
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +83,7 @@ namespace mg5amcCpu
       static bool first = true;
       if( first )
       {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
         if constexpr( M_ACCESS::isOnDevice() ) // avoid
         {
           const int ievt0 = 0;
@@ -166,7 +166,7 @@ namespace mg5amcCpu
     wt = po2log;
     if( nparf != 2 ) wt = ( 2. * nparf - 4. ) * log( energy ) + z[nparf - 1];
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     // issue warnings if weight is too small or too large
     static int iwarn[5] = { 0, 0, 0, 0, 0 };
     if( wt < -180. )
diff --git a/epochX/cudacpp/gg_ttg.sa/src/read_slha.cc b/epochX/cudacpp/gg_ttg.sa/src/read_slha.cc
index 055b19a779..f8e46f2e66 100644
--- a/epochX/cudacpp/gg_ttg.sa/src/read_slha.cc
+++ b/epochX/cudacpp/gg_ttg.sa/src/read_slha.cc
@@ -11,7 +11,11 @@
 
 #include <algorithm>
 #include <cstdlib>
-#include <filesystem>
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
 #include <fstream>
 #include <iostream>
 
@@ -60,7 +64,14 @@ SLHAReader::read_slha_file( std::string file_name, bool verbose )
     {
       std::cout << "WARNING! Card file '" << file_name << "' does not exist:"
                 << " look for the file in directory $" << envpath << "='" << getenv( envpath ) << "'" << std::endl;
+      /*
+#ifdef __HIPCC__
+      const std::string file_name2 = std::experimental::filesystem::path( getenv( envpath ) ) / std::experimental::filesystem::path( file_name ).filename();
+#else
       const std::string file_name2 = std::filesystem::path( getenv( envpath ) ) / std::filesystem::path( file_name ).filename();
+#endif
+      */
+      const std::string file_name2 = std::string( getenv( envpath ) ) + "/" + file_name; // bypass std::filesystem #803
       param_card.open( file_name2.c_str(), std::ifstream::in );
       if( param_card.good() )
       {
diff --git a/epochX/cudacpp/gg_ttgg.mad/CODEGEN_mad_gg_ttgg_log.txt b/epochX/cudacpp/gg_ttgg.mad/CODEGEN_mad_gg_ttgg_log.txt
index 2c2fae1608..0da89f1729 100644
--- a/epochX/cudacpp/gg_ttgg.mad/CODEGEN_mad_gg_ttgg_log.txt
+++ b/epochX/cudacpp/gg_ttgg.mad/CODEGEN_mad_gg_ttgg_log.txt
@@ -52,7 +52,7 @@ Note that you can still compile and run aMC@NLO with the built-in PDFs
 
 Using default text editor "vi". Set another one in ./input/mg5_configuration.txt
 Using default eps viewer "evince". Set another one in ./input/mg5_configuration.txt
-No valid web browser found. Please set in ./input/mg5_configuration.txt
+Using default web browser "firefox". Set another one in ./input/mg5_configuration.txt
 import /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_mad_gg_ttgg.mg
 The import format was not given, so we guess it as command
 set stdout_level DEBUG
@@ -62,7 +62,7 @@ generate g g > t t~ g g
 No model currently active, so we import the Standard Model
 INFO: load particles 
 INFO: load vertices 
-DEBUG: model prefixing  takes 0.0057299137115478516 
+DEBUG: model prefixing  takes 0.005736112594604492 
 INFO: Restrict model sm with file models/sm/restrict_default.dat . 
 DEBUG: Simplifying conditional expressions 
 DEBUG: remove interactions: u s w+ at order: QED=1 
@@ -155,7 +155,7 @@ INFO: Please specify coupling orders to bypass this step.
 INFO: Trying coupling order WEIGHTED<=4: WEIGTHED IS QCD+2*QED 
 INFO: Trying process: g g > t t~ g g WEIGHTED<=4 @1  
 INFO: Process has 123 diagrams 
-1 processes with 123 diagrams generated in 0.163 s
+1 processes with 123 diagrams generated in 0.160 s
 Total: 1 processes with 123 diagrams
 output madevent ../TMPOUT/CODEGEN_mad_gg_ttgg --hel_recycling=False --vector_size=32 --me_exporter=standalone_cudacpp
 Load PLUGIN.CUDACPP_OUTPUT
@@ -175,7 +175,7 @@ INFO: Generating Helas calls for process: g g > t t~ g g WEIGHTED<=4 @1
 INFO: Processing color information for process: g g > t t~ g g @1 
 INFO: Creating files in directory P1_gg_ttxgg 
 DEBUG:  kwargs[prefix] = 0 [model_handling.py at line 1058] 
-DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7f37353fffa0> [export_v4.py at line 6262] 
+DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7f0169c41c70> [export_v4.py at line 6262] 
 INFO: Creating files in directory . 
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.h
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.cc
@@ -190,15 +190,15 @@ INFO: Created files CPPProcess.h and CPPProcess.cc in directory ./.
 DEBUG:  vector, subproc_group,self.opt['vector_size'] =  32 True 32 [export_v4.py at line 1872] 
 INFO: Generating Feynman diagrams for Process: g g > t t~ g g WEIGHTED<=4 @1 
 INFO: Finding symmetric diagrams for subprocess group gg_ttxgg 
-Generated helas calls for 1 subprocesses (123 diagrams) in 0.433 s
-Wrote files for 222 helas calls in 0.711 s
+Generated helas calls for 1 subprocesses (123 diagrams) in 0.423 s
+Wrote files for 222 helas calls in 0.683 s
 ALOHA: aloha starts to compute helicity amplitudes
 ALOHA: aloha creates VVV1 routines
 ALOHA: aloha creates FFV1 routines
 ALOHA: aloha creates VVVV1 routines
 ALOHA: aloha creates VVVV3 routines
 ALOHA: aloha creates VVVV4 routines
-ALOHA: aloha creates 5 routines in  0.336 s
+ALOHA: aloha creates 5 routines in  0.328 s
 DEBUG:  Entering PLUGIN_ProcessExporter.convert_model (create the model) [output.py at line 202] 
 ALOHA: aloha starts to compute helicity amplitudes
 ALOHA: aloha creates VVV1 routines
@@ -206,7 +206,7 @@ ALOHA: aloha creates FFV1 routines
 ALOHA: aloha creates VVVV1 routines
 ALOHA: aloha creates VVVV3 routines
 ALOHA: aloha creates VVVV4 routines
-ALOHA: aloha creates 10 routines in  0.327 s
+ALOHA: aloha creates 10 routines in  0.310 s
 <class 'aloha.create_aloha.AbstractRoutine'> VVV1
 <class 'aloha.create_aloha.AbstractRoutine'> VVV1
 <class 'aloha.create_aloha.AbstractRoutine'> FFV1
@@ -255,9 +255,9 @@ Type "launch" to generate events from this process, or see
 Run "open index.html" to see more information about this process.
 quit
 
-real	0m3.329s
-user	0m3.091s
-sys	0m0.226s
+real	0m3.254s
+user	0m3.011s
+sys	0m0.236s
 Code generation completed in 4 seconds
 ************************************************************
 *                                                          *
@@ -284,7 +284,7 @@ INFO: load configuration from /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/mg5amc
 INFO: load configuration from /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_mad_gg_ttgg/Cards/me5_configuration.txt  
 Using default text editor "vi". Set another one in ./input/mg5_configuration.txt
 Using default eps viewer "evince". Set another one in ./input/mg5_configuration.txt
-No valid web browser found. Please set in ./input/mg5_configuration.txt
+Using default web browser "firefox". Set another one in ./input/mg5_configuration.txt
 treatcards run
 quit
 INFO:  
@@ -314,7 +314,7 @@ INFO: load configuration from /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/mg5amc
 INFO: load configuration from /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_mad_gg_ttgg/Cards/me5_configuration.txt  
 Using default text editor "vi". Set another one in ./input/mg5_configuration.txt
 Using default eps viewer "evince". Set another one in ./input/mg5_configuration.txt
-No valid web browser found. Please set in ./input/mg5_configuration.txt
+Using default web browser "firefox". Set another one in ./input/mg5_configuration.txt
 treatcards param
 quit
 INFO:  
diff --git a/epochX/cudacpp/gg_ttgg.mad/COPYRIGHT b/epochX/cudacpp/gg_ttgg.mad/COPYRIGHT
index a134b5fef9..9036d9260a 100644
--- a/epochX/cudacpp/gg_ttgg.mad/COPYRIGHT
+++ b/epochX/cudacpp/gg_ttgg.mad/COPYRIGHT
@@ -15,6 +15,7 @@ The full development team currently includes the following authors :
   Stephan Hageboeck (CERN)
   Olivier Mattelaer (Universite Catholique de Louvain, original author)
   Stefan Roiser (CERN, original author)
+  Jorgen Teig (CERN)
   Andrea Valassi (CERN, original author)
   Zenny Wettersten (CERN)
 See https://github.com/madgraph5/madgraph4gpu for more details. For the full
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/Bridge.h b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/Bridge.h
index bf8b5e024d..f9ed70dfde 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/Bridge.h
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/Bridge.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Roiser (Nov 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef BRIDGE_H
 #define BRIDGE_H 1
@@ -14,16 +14,23 @@
 #include "MemoryAccessMomenta.h"  // for MemoryAccessMomenta::neppM
 #include "MemoryBuffers.h"        // for HostBufferMomenta, DeviceBufferMomenta etc
 
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
+
+#include <sys/stat.h> // bypass std::filesystem #803
+
 #include <algorithm>
 #include <cassert>
 #include <cmath>
 #include <cstring>
-#include <filesystem>
 #include <iostream>
 #include <memory>
 #include <type_traits>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +90,7 @@ namespace mg5amcCpu
     Bridge& operator=( const Bridge& ) = delete;
     Bridge& operator=( Bridge&& ) = delete;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     /**
      * Set the gpublocks and gputhreads for the gpusequence - throws if evnt != gpublocks*gputhreads
      * (this is needed for BridgeKernel tests rather than for actual production use in Fortran)
@@ -150,7 +157,7 @@ namespace mg5amcCpu
     unsigned int m_nevt; // number of events
     int m_nGoodHel;      // the number of good helicities (-1 initially when they have not yet been calculated)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     int m_gputhreads; // number of gpu threads (default set from number of events, can be modified)
     int m_gpublocks;  // number of gpu blocks (default set from number of events, can be modified)
     DeviceBuffer<FORTRANFPTYPE, sizePerEventMomenta> m_devMomentaF;
@@ -187,12 +194,12 @@ namespace mg5amcCpu
   // Forward declare transposition methods
   //
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
   template<typename Tin, typename Tout>
   __global__ void dev_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt );
 
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
 
   template<typename Tin, typename Tout>
   void hst_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt );
@@ -209,7 +216,7 @@ namespace mg5amcCpu
   Bridge<FORTRANFPTYPE>::Bridge( unsigned int nevtF, unsigned int nparF, unsigned int np4F )
     : m_nevt( nevtF )
     , m_nGoodHel( -1 )
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     , m_gputhreads( 256 )                  // default number of gpu threads
     , m_gpublocks( m_nevt / m_gputhreads ) // this ensures m_nevt <= m_gpublocks*m_gputhreads
     , m_devMomentaF( m_nevt )
@@ -233,7 +240,7 @@ namespace mg5amcCpu
   {
     if( nparF != CPPProcess::npar ) throw std::runtime_error( "Bridge constructor: npar mismatch" );
     if( np4F != CPPProcess::np4 ) throw std::runtime_error( "Bridge constructor: np4 mismatch" );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( ( m_nevt < s_gputhreadsmin ) || ( m_nevt % s_gputhreadsmin != 0 ) )
       throw std::runtime_error( "Bridge constructor: nevt should be a multiple of " + std::to_string( s_gputhreadsmin ) );
     while( m_nevt != m_gpublocks * m_gputhreads )
@@ -249,20 +256,28 @@ namespace mg5amcCpu
 #else
     std::cout << "WARNING! Instantiate host Bridge (nevt=" << m_nevt << ")" << std::endl;
     m_pmek.reset( new MatrixElementKernelHost( m_hstMomentaC, m_hstGs, m_hstRndHel, m_hstRndCol, m_hstMEs, m_hstSelHel, m_hstSelCol, m_nevt ) );
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
     // Create a process object, read param card and set parameters
     // FIXME: the process instance can happily go out of scope because it is only needed to read parameters?
     // FIXME: the CPPProcess should really be a singleton? what if fbridgecreate is called from several Fortran threads?
     CPPProcess process( /*verbose=*/false );
     std::string paramCard = "../../Cards/param_card.dat";
-    if( !std::filesystem::exists( paramCard ) )
-    {
-      paramCard = "../" + paramCard;
-    }
+    /*
+#ifdef __HIPCC__
+    if( !std::experimental::filesystem::exists( paramCard ) ) paramCard = "../" + paramCard;
+#else
+    if( !std::filesystem::exists( paramCard ) ) paramCard = "../" + paramCard;
+#endif
+    */
+    //struct stat dummybuffer; // bypass std::filesystem #803
+    //if( !( stat( paramCard.c_str(), &dummyBuffer ) == 0 ) ) paramCard = "../" + paramCard; //
+    auto fileExists = []( std::string& fileName )
+    { struct stat buffer; return stat( fileName.c_str(), &buffer ) == 0; };
+    if( !fileExists( paramCard ) ) paramCard = "../" + paramCard; // bypass std::filesystem #803
     process.initProc( paramCard );
   }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::set_gpugrid( const int gpublocks, const int gputhreads )
   {
@@ -276,7 +291,7 @@ namespace mg5amcCpu
   }
 #endif
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::gpu_sequence( const FORTRANFPTYPE* momenta,
                                             const FORTRANFPTYPE* gs,
@@ -291,14 +306,14 @@ namespace mg5amcCpu
     constexpr int neppM = MemoryAccessMomenta::neppM;
     if constexpr( neppM == 1 && std::is_same_v<FORTRANFPTYPE, fptype> )
     {
-      checkCuda( cudaMemcpy( m_devMomentaC.data(), momenta, m_devMomentaC.bytes(), cudaMemcpyHostToDevice ) );
+      gpuMemcpy( m_devMomentaC.data(), momenta, m_devMomentaC.bytes(), gpuMemcpyHostToDevice );
     }
     else
     {
-      checkCuda( cudaMemcpy( m_devMomentaF.data(), momenta, m_devMomentaF.bytes(), cudaMemcpyHostToDevice ) );
+      gpuMemcpy( m_devMomentaF.data(), momenta, m_devMomentaF.bytes(), gpuMemcpyHostToDevice );
       const int thrPerEvt = CPPProcess::npar * CPPProcess::np4; // AV: transpose alg does 1 element per thread (NOT 1 event per thread)
       //const int thrPerEvt = 1; // AV: try new alg with 1 event per thread... this seems slower
-      dev_transposeMomentaF2C<<<m_gpublocks * thrPerEvt, m_gputhreads>>>( m_devMomentaF.data(), m_devMomentaC.data(), m_nevt );
+      gpuLaunchKernel( dev_transposeMomentaF2C, m_gpublocks * thrPerEvt, m_gputhreads, m_devMomentaF.data(), m_devMomentaC.data(), m_nevt );
     }
     if constexpr( std::is_same_v<FORTRANFPTYPE, fptype> )
     {
@@ -341,7 +356,7 @@ namespace mg5amcCpu
   }
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::cpu_sequence( const FORTRANFPTYPE* momenta,
                                             const FORTRANFPTYPE* gs,
@@ -396,7 +411,7 @@ namespace mg5amcCpu
   // - C++ array: momenta[npagM][npar][np4][neppM] with nevt=npagM*neppM (AOSOA)
   //
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename Tin, typename Tout>
   __global__ void dev_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt )
   {
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/BridgeKernels.cc b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/BridgeKernels.cc
index d58066c9c1..eaf4037a24 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/BridgeKernels.cc
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/BridgeKernels.cc
@@ -1,17 +1,18 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "BridgeKernels.h"
 
+#include "GpuAbstraction.h"
 #include "MemoryAccessMomenta.h"
 
 #include <sstream>
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -45,7 +46,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu
 {
 
@@ -96,7 +97,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/BridgeKernels.h b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/BridgeKernels.h
index 15eb4bff4d..3efef8ce97 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/BridgeKernels.h
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/BridgeKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef BRIDGEKERNELS_H
 #define BRIDGEKERNELS_H 1
@@ -12,7 +12,7 @@
 #include "MatrixElementKernels.h"
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -49,7 +49,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A Bridge wrapper class encapsulating matrix element calculations on a CPU host
   class BridgeKernelHost final : public BridgeKernelBase
   {
@@ -89,7 +89,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A Bridge wrapper class encapsulating matrix element calculations on a GPU device
   class BridgeKernelDevice : public BridgeKernelBase
   {
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/CommonRandomNumberKernel.cc b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/CommonRandomNumberKernel.cc
index 985b39f576..010bc4cbd0 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/CommonRandomNumberKernel.cc
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/CommonRandomNumberKernel.cc
@@ -1,15 +1,16 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "CommonRandomNumbers.h"
+#include "GpuAbstraction.h"
 #include "MemoryBuffers.h"
 #include "RandomNumberKernels.h"
 
 #include <cassert>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/CrossSectionKernels.cc b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/CrossSectionKernels.cc
index 0b355a3c8d..c15b39844d 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/CrossSectionKernels.cc
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/CrossSectionKernels.cc
@@ -1,10 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "CrossSectionKernels.h"
 
+#include "GpuAbstraction.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessWeights.h"
 #include "MemoryBuffers.h"
@@ -77,7 +78,7 @@ debug_me_is_abnormal( const fptype& me, size_t ievtALL )
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -185,7 +186,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/CrossSectionKernels.h b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/CrossSectionKernels.h
index 7933ca4bbf..4d9659e04e 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/CrossSectionKernels.h
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/CrossSectionKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef CROSSSECTIONKERNELS_H
 #define CROSSSECTIONKERNELS_H 1
@@ -13,7 +13,7 @@
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -96,7 +96,7 @@ namespace mg5amcCpu
   //--------------------------------------------------------------------------
 
   /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating the calculation of event statistics on a GPU device
   class CrossSectionKernelDevice : public CrossSectionKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/CudaRuntime.h b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/CudaRuntime.h
deleted file mode 100644
index 64ce52f4b3..0000000000
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/CudaRuntime.h
+++ /dev/null
@@ -1,85 +0,0 @@
-// Copyright (C) 2020-2023 CERN and UCLouvain.
-// Licensed under the GNU Lesser General Public License (version 3 or later).
-// Created by: S. Roiser (Jul 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
-
-#ifndef MG5AMC_CUDARUNTIME_H
-#define MG5AMC_CUDARUNTIME_H 1
-
-// MG5AMC on GPU uses the CUDA runtime API, not the lower level CUDA driver API
-// See https://docs.nvidia.com/cuda/cuda-runtime-api/driver-vs-runtime-api.html#driver-vs-runtime-api
-
-#include <cassert>
-#include <iostream>
-
-//--------------------------------------------------------------------------
-
-// See https://stackoverflow.com/a/14038590
-#ifdef __CUDACC__ /* clang-format off */
-#define checkCuda( code ) { assertCuda( code, __FILE__, __LINE__ ); }
-inline void assertCuda( cudaError_t code, const char* file, int line, bool abort = true )
-{
-  if( code != cudaSuccess )
-  {
-    printf( "ERROR! assertCuda: '%s' (%d) in %s:%d\n", cudaGetErrorString( code ), code, file, line );
-    if( abort ) assert( code == cudaSuccess );
-  }
-}
-#endif /* clang-format on */
-
-//--------------------------------------------------------------------------
-
-#ifdef __CUDACC__
-namespace mg5amcGpu
-{
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  // *** FIXME! This will all need to be designed differently when going to multi-GPU nodes! ***
-  struct CudaRuntime final
-  {
-    CudaRuntime( const bool debug = true )
-      : m_debug( debug ) { setUp( m_debug ); }
-    ~CudaRuntime() { tearDown( m_debug ); }
-    CudaRuntime( const CudaRuntime& ) = delete;
-    CudaRuntime( CudaRuntime&& ) = delete;
-    CudaRuntime& operator=( const CudaRuntime& ) = delete;
-    CudaRuntime& operator=( CudaRuntime&& ) = delete;
-    bool m_debug;
-
-    // Set up CUDA application
-    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
-    // Calling cudaSetDevice on startup is useful to properly book-keep the time spent in CUDA initialization
-    static void setUp( const bool debug = true )
-    {
-      // ** NB: it is useful to call cudaSetDevice, or cudaFree, to properly book-keep the time spent in CUDA initialization
-      // ** NB: otherwise, the first CUDA operation (eg a cudaMemcpyToSymbol in CPPProcess ctor) appears to take much longer!
-      /*
-      // [We initially added cudaFree(0) to "ease profile analysis" only because it shows up as a big recognizable block!]
-      // No explicit initialization is needed: https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#initialization
-      // It is not clear what cudaFree(0) does at all: https://stackoverflow.com/questions/69967813/
-      if ( debug ) std::cout << "__CudaRuntime: calling cudaFree(0)" << std::endl;
-      checkCuda( cudaFree( 0 ) ); // SLOW!
-      */
-      // Replace cudaFree(0) by cudaSetDevice(0), even if it is not really needed either
-      // (but see https://developer.nvidia.com/blog/cuda-pro-tip-always-set-current-device-avoid-multithreading-bugs)
-      if( debug ) std::cout << "__CudaRuntime: calling cudaSetDevice(0)" << std::endl;
-      checkCuda( cudaSetDevice( 0 ) ); // SLOW!
-    }
-
-    // Tear down CUDA application (call cudaDeviceReset)
-    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
-    // Calling cudaDeviceReset on shutdown is only needed for checking memory leaks in cuda-memcheck
-    // See https://docs.nvidia.com/cuda/cuda-memcheck/index.html#leak-checking
-    static void tearDown( const bool debug = true )
-    {
-      if( debug ) std::cout << "__CudaRuntime: calling cudaDeviceReset()" << std::endl;
-      checkCuda( cudaDeviceReset() );
-    }
-  };
-
-}
-#endif
-
-//--------------------------------------------------------------------------
-
-#endif // MG5AMC_CUDARUNTIME_H
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/CurandRandomNumberKernel.cc b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/CurandRandomNumberKernel.cc
index eb56333b03..08a16f6f2c 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/CurandRandomNumberKernel.cc
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/CurandRandomNumberKernel.cc
@@ -1,9 +1,9 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "MemoryBuffers.h"
 #include "RandomNumberKernels.h"
 
@@ -22,7 +22,7 @@ inline void assertCurand( curandStatus_t code, const char *file, int line, bool
 }
 #endif /* clang-format on */
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -36,7 +36,7 @@ namespace mg5amcCpu
   {
     if( m_isOnDevice )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if( !m_rnarray.isOnDevice() )
         throw std::runtime_error( "CurandRandomNumberKernel on device with a host random number array" );
 #else
@@ -114,7 +114,7 @@ namespace mg5amcCpu
     /*
     printf( "\nCurandRandomNumberKernel::generateRnarray size = %d\n", (int)m_rnarray.size() );
     fptype* data = m_rnarray.data();
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( m_rnarray.isOnDevice() )
     {
       data = new fptype[m_rnarray.size()]();
@@ -123,7 +123,7 @@ namespace mg5amcCpu
 #endif
     for( int i = 0; i < ( (int)m_rnarray.size() / 4 ); i++ )
       printf( "[%4d] %f %f %f %f\n", i * 4, data[i * 4], data[i * 4 + 2], data[i * 4 + 2], data[i * 4 + 3] );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( m_rnarray.isOnDevice() ) delete[] data;
 #endif
     */
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/EventStatistics.h b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/EventStatistics.h
index 48b51e0a49..b425a5bade 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/EventStatistics.h
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/EventStatistics.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef EventStatistics_H
 #define EventStatistics_H 1
@@ -16,7 +16,7 @@
 #include <limits>
 #include <string>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/GpuAbstraction.h b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/GpuAbstraction.h
new file mode 100644
index 0000000000..9c467b1e04
--- /dev/null
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/GpuAbstraction.h
@@ -0,0 +1,69 @@
+// Copyright (C) 2020-2023 CERN and UCLouvain.
+// Licensed under the GNU Lesser General Public License (version 3 or later).
+// Created by: J. Teig (Jul 2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+
+#ifndef MG5AMC_GPUABSTRACTION_H
+#define MG5AMC_GPUABSTRACTION_H 1
+
+#include <cassert>
+
+//--------------------------------------------------------------------------
+
+#ifdef __CUDACC__
+
+#define gpuError_t cudaError_t
+#define gpuPeekAtLastError cudaPeekAtLastError
+#define gpuGetErrorString cudaGetErrorString
+#define gpuSuccess cudaSuccess
+
+#define gpuMallocHost( ptr, size ) checkGpu( cudaMallocHost( ptr, size ) )
+#define gpuMalloc( ptr, size ) checkGpu( cudaMalloc( ptr, size ) )
+
+#define gpuMemcpy( dstData, srcData, srcBytes, func ) checkGpu( cudaMemcpy( dstData, srcData, srcBytes, func ) )
+#define gpuMemcpyHostToDevice cudaMemcpyHostToDevice
+#define gpuMemcpyDeviceToHost cudaMemcpyDeviceToHost
+#define gpuMemcpyToSymbol( type1, type2, size ) checkGpu( cudaMemcpyToSymbol( type1, type2, size ) )
+
+#define gpuFree( ptr ) checkGpu( cudaFree( ptr ) )
+#define gpuFreeHost( ptr ) checkGpu( cudaFreeHost( ptr ) )
+
+#define gpuSetDevice cudaSetDevice
+#define gpuDeviceSynchronize cudaDeviceSynchronize
+#define gpuDeviceReset cudaDeviceReset
+
+#define gpuLaunchKernel( kernel, blocks, threads, ... ) kernel<<<blocks, threads>>>( __VA_ARGS__ )
+#define gpuLaunchKernelSharedMem( kernel, blocks, threads, sharedMem, ... ) kernel<<<blocks, threads, sharedMem>>>( __VA_ARGS__ )
+
+//--------------------------------------------------------------------------
+
+#elif defined __HIPCC__
+
+#define gpuError_t hipError_t
+#define gpuPeekAtLastError hipPeekAtLastError
+#define gpuGetErrorString hipGetErrorString
+#define gpuSuccess hipSuccess
+
+#define gpuMallocHost( ptr, size ) checkGpu( hipHostMalloc( ptr, size ) ) // HostMalloc better
+#define gpuMalloc( ptr, size ) checkGpu( hipMalloc( ptr, size ) )
+
+#define gpuMemcpy( dstData, srcData, srcBytes, func ) checkGpu( hipMemcpy( dstData, srcData, srcBytes, func ) )
+#define gpuMemcpyHostToDevice hipMemcpyHostToDevice
+#define gpuMemcpyDeviceToHost hipMemcpyDeviceToHost
+#define gpuMemcpyToSymbol( type1, type2, size ) checkGpu( hipMemcpyToSymbol( type1, type2, size ) )
+
+#define gpuFree( ptr ) checkGpu( hipFree( ptr ) )
+#define gpuFreeHost( ptr ) checkGpu( hipHostFree( ptr ) )
+
+#define gpuSetDevice hipSetDevice
+#define gpuDeviceSynchronize hipDeviceSynchronize
+#define gpuDeviceReset hipDeviceReset
+
+#define gpuLaunchKernel( kernel, blocks, threads, ... ) kernel<<<blocks, threads>>>( __VA_ARGS__ )
+#define gpuLaunchKernelSharedMem( kernel, blocks, threads, sharedMem, ... ) kernel<<<blocks, threads, sharedMem>>>( __VA_ARGS__ )
+
+//--------------------------------------------------------------------------
+
+#endif
+
+#endif // MG5AMC_GPUABSTRACTION_H
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/GpuRuntime.h b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/GpuRuntime.h
new file mode 100644
index 0000000000..93579ef08b
--- /dev/null
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/GpuRuntime.h
@@ -0,0 +1,85 @@
+// Copyright (C) 2020-2023 CERN and UCLouvain.
+// Licensed under the GNU Lesser General Public License (version 3 or later).
+// Created by: J. Teig (Jun 2023, based on earlier work by S. Roiser) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+
+#ifndef MG5AMC_GPURUNTIME_H
+#define MG5AMC_GPURUNTIME_H 1
+
+// MG5AMC on GPU uses the CUDA runtime API, not the lower level CUDA driver API
+// See https://docs.nvidia.com/cuda/cuda-runtime-api/driver-vs-runtime-api.html#driver-vs-runtime-api
+
+#include "GpuAbstraction.h"
+
+#include <iostream>
+
+//--------------------------------------------------------------------------
+
+// See https://stackoverflow.com/a/14038590
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
+#define checkGpu( code ) { assertGpu( code, __FILE__, __LINE__ ); }
+inline void assertGpu( gpuError_t code, const char* file, int line, bool abort = true )
+{
+  if( code != gpuSuccess )
+  {
+    printf( "ERROR! assertGpu: '%s' (%d) in %s:%d\n", gpuGetErrorString( code ), code, file, line );
+    if( abort ) assert( code == gpuSuccess );
+  }
+}
+#endif /* clang-format on */
+
+//--------------------------------------------------------------------------
+
+#ifdef MGONGPUCPP_GPUIMPL
+namespace mg5amcGpu
+{
+  // Instantiate a GpuRuntime at the beginnining of the application's main to
+  // invoke gpuSetDevice(0) in the constructor and book a gpuDeviceReset() call in the destructor
+  // *** FIXME! This will all need to be designed differently when going to multi-GPU nodes! ***
+  struct GpuRuntime final
+  {
+    GpuRuntime( const bool debug = true )
+      : m_debug( debug ) { setUp( m_debug ); }
+    ~GpuRuntime() { tearDown( m_debug ); }
+    GpuRuntime( const GpuRuntime& ) = delete;
+    GpuRuntime( GpuRuntime&& ) = delete;
+    GpuRuntime& operator=( const GpuRuntime& ) = delete;
+    GpuRuntime& operator=( GpuRuntime&& ) = delete;
+    bool m_debug;
+
+    // Set up CUDA application
+    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
+    // Calling cudaSetDevice on startup is useful to properly book-keep the time spent in CUDA initialization
+    static void setUp( const bool debug = true )
+    {
+      // ** NB: it is useful to call cudaSetDevice, or cudaFree, to properly book-keep the time spent in CUDA initialization
+      // ** NB: otherwise, the first CUDA operation (eg a cudaMemcpyToSymbol in CPPProcess ctor) appears to take much longer!
+      /*
+      // [We initially added cudaFree(0) to "ease profile analysis" only because it shows up as a big recognizable block!]
+      // No explicit initialization is needed: https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#initialization
+      // It is not clear what cudaFree(0) does at all: https://stackoverflow.com/questions/69967813/
+      if ( debug ) std::cout << "__CudaRuntime: calling cudaFree(0)" << std::endl;
+      checkCuda( cudaFree( 0 ) ); // SLOW!
+      */
+      // Replace cudaFree(0) by cudaSetDevice(0), even if it is not really needed either
+      // (but see https://developer.nvidia.com/blog/cuda-pro-tip-always-set-current-device-avoid-multithreading-bugs)
+      if( debug ) std::cout << "__GpuRuntime: calling GpuSetDevice(0)" << std::endl;
+      checkGpu( gpuSetDevice( 0 ) ); // SLOW!
+    }
+
+    // Tear down CUDA application (call cudaDeviceReset)
+    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
+    // Calling cudaDeviceReset on shutdown is only needed for checking memory leaks in cuda-memcheck
+    // See https://docs.nvidia.com/cuda/cuda-memcheck/index.html#leak-checking
+    static void tearDown( const bool debug = true )
+    {
+      if( debug ) std::cout << "__GpuRuntime: calling GpuDeviceReset()" << std::endl;
+      checkGpu( gpuDeviceReset() );
+    }
+  };
+}
+#endif
+
+//--------------------------------------------------------------------------
+
+#endif // MG5AMC_GPURUNTIME_H
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MadgraphTest.h b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MadgraphTest.h
index ef40624c88..6054185300 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MadgraphTest.h
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MadgraphTest.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Hageboeck (Dec 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MADGRAPHTEST_H_
 #define MADGRAPHTEST_H_ 1
@@ -14,7 +14,11 @@
 
 #include <array>
 #include <cmath>
-#include <filesystem>
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
 #include <fstream>
 #include <iomanip>
 #include <memory>
@@ -22,7 +26,7 @@
 #include <string>
 #include <vector>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using mg5amcGpu::CPPProcess;
 #else
 using mg5amcCpu::CPPProcess;
@@ -201,7 +205,7 @@ class MadgraphTest : public testing::TestWithParam<TestDriverBase*>
 
 // Since we link both the CPU-only and GPU tests into the same executable, we prevent
 // a multiply defined symbol by only compiling this in the non-CUDA phase:
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
 /// Compare momenta and matrix elements.
 /// This uses an implementation of TestDriverBase to run a madgraph workflow,
@@ -219,7 +223,14 @@ TEST_P( MadgraphTest, CompareMomentaAndME )
   const char* dumpEventsC = getenv( "CUDACPP_RUNTEST_DUMPEVENTS" );
   const bool dumpEvents = ( dumpEventsC != 0 ) && ( std::string( dumpEventsC ) != "" );
   const std::string refFileName = testDriver->getRefFileName();
+  /*
+#ifdef __HIPCC__
+  const std::string dumpFileName = std::experimental::filesystem::path( refFileName ).filename();
+#else
   const std::string dumpFileName = std::filesystem::path( refFileName ).filename();
+#endif
+  */
+  const std::string dumpFileName = refFileName; // bypass std::filesystem #803
   std::ofstream dumpFile;
   if( dumpEvents )
   {
@@ -307,6 +318,6 @@ TEST_P( MadgraphTest, CompareMomentaAndME )
   }
 }
 
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
 
 #endif /* MADGRAPHTEST_H_ */
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MatrixElementKernels.cc b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MatrixElementKernels.cc
index 74b5239ebf..81699dfea9 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MatrixElementKernels.cc
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MatrixElementKernels.cc
@@ -1,12 +1,12 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "MatrixElementKernels.h"
 
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h" // Includes the abstraction for Nvidia/AMD compilation
 #include "MemoryAccessMomenta.h"
 #include "MemoryBuffers.h"
 
@@ -14,7 +14,7 @@
 
 //============================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu
 {
 
@@ -150,7 +150,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
@@ -209,13 +209,13 @@ namespace mg5amcGpu
     PinnedHostBufferHelicityMask hstIsGoodHel( ncomb );
     DeviceBufferHelicityMask devIsGoodHel( ncomb );
     // ... 0d1. Compute good helicity mask on the device
-    computeDependentCouplings<<<m_gpublocks, m_gputhreads>>>( m_gs.data(), m_couplings.data() );
+    gpuLaunchKernel( computeDependentCouplings, m_gpublocks, m_gputhreads, m_gs.data(), m_couplings.data() );
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-    sigmaKin_getGoodHel<<<m_gpublocks, m_gputhreads>>>( m_momenta.data(), m_couplings.data(), m_matrixElements.data(), m_numerators.data(), m_denominators.data(), devIsGoodHel.data() );
+    gpuLaunchKernel( sigmaKin_getGoodHel, m_gpublocks, m_gputhreads, m_momenta.data(), m_couplings.data(), m_matrixElements.data(), m_numerators.data(), m_denominators.data(), devIsGoodHel.data() );
 #else
-    sigmaKin_getGoodHel<<<m_gpublocks, m_gputhreads>>>( m_momenta.data(), m_couplings.data(), m_matrixElements.data(), devIsGoodHel.data() );
+    gpuLaunchKernel( sigmaKin_getGoodHel, m_gpublocks, m_gputhreads, m_momenta.data(), m_couplings.data(), m_matrixElements.data(), devIsGoodHel.data() );
 #endif
-    checkCuda( cudaPeekAtLastError() );
+    checkGpu( gpuPeekAtLastError() );
     // ... 0d2. Copy back good helicity mask to the host
     copyHostFromDevice( hstIsGoodHel, devIsGoodHel );
     // ... 0d3. Copy back good helicity list to constant memory on the device
@@ -226,19 +226,19 @@ namespace mg5amcGpu
 
   void MatrixElementKernelDevice::computeMatrixElements( const unsigned int channelId )
   {
-    computeDependentCouplings<<<m_gpublocks, m_gputhreads>>>( m_gs.data(), m_couplings.data() );
+    gpuLaunchKernel( computeDependentCouplings, m_gpublocks, m_gputhreads, m_gs.data(), m_couplings.data() );
 #ifndef MGONGPU_NSIGHT_DEBUG
     constexpr unsigned int sharedMemSize = 0;
 #else
     constexpr unsigned int sharedMemSize = ntpbMAX * sizeof( float );
 #endif
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-    sigmaKin<<<m_gpublocks, m_gputhreads, sharedMemSize>>>( m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), channelId, m_numerators.data(), m_denominators.data(), m_selhel.data(), m_selcol.data() );
+    gpuLaunchKernelSharedMem( sigmaKin, m_gpublocks, m_gputhreads, sharedMemSize, m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), channelId, m_numerators.data(), m_denominators.data(), m_selhel.data(), m_selcol.data() );
 #else
-    sigmaKin<<<m_gpublocks, m_gputhreads, sharedMemSize>>>( m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), m_selhel.data(), m_selcol.data() );
+    gpuLaunchKernelSharedMem( sigmaKin, m_gpublocks, m_gputhreads, sharedMemSize, m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), m_selhel.data(), m_selcol.data() );
 #endif
-    checkCuda( cudaPeekAtLastError() );
-    checkCuda( cudaDeviceSynchronize() );
+    checkGpu( gpuPeekAtLastError() );
+    checkGpu( gpuDeviceSynchronize() );
   }
 
   //--------------------------------------------------------------------------
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MatrixElementKernels.h b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MatrixElementKernels.h
index 23e84757a2..72bd8f195b 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MatrixElementKernels.h
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MatrixElementKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MATRIXELEMENTKERNELS_H
 #define MATRIXELEMENTKERNELS_H 1
@@ -10,7 +10,7 @@
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -81,7 +81,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating matrix element calculations on a CPU host
   class MatrixElementKernelHost final : public MatrixElementKernelBase, public NumberOfEvents
   {
@@ -130,7 +130,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating matrix element calculations on a GPU device
   class MatrixElementKernelDevice : public MatrixElementKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessAmplitudes.h b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessAmplitudes.h
index 573b3bbbc9..ffb76e93de 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessAmplitudes.h
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessAmplitudes.h
@@ -15,7 +15,7 @@
 #define MGONGPU_TRIVIAL_AMPLITUDES 1
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessCouplings.h b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessCouplings.h
index 35a3af42e0..3afdf3e554 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessCouplings.h
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessCouplings.h
@@ -15,7 +15,7 @@
 #include "MemoryBuffers.h"       // for HostBufferCouplings::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessCouplingsFixed.h b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessCouplingsFixed.h
index dc0d93afff..ffcdf4dbef 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessCouplingsFixed.h
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessCouplingsFixed.h
@@ -14,7 +14,7 @@
 //#include "MemoryAccessHelpers.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessDenominators.h b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessDenominators.h
index 3bce635718..66f2d32a6b 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessDenominators.h
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessDenominators.h
@@ -10,7 +10,7 @@
 #include "MemoryAccessGs.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessGs.h b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessGs.h
index 31311aa375..4c726b30f3 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessGs.h
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessGs.h
@@ -13,7 +13,7 @@
 #include "MemoryBuffers.h" // for HostBufferMatrixElements::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessHelpers.h b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessHelpers.h
index c82a6c7635..db73e4e064 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessHelpers.h
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessHelpers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessHelpers_H
 #define MemoryAccessHelpers_H 1
@@ -105,7 +105,7 @@ class KernelAccessHelper : public MemoryAccessHelper<T>
     }
     else
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
       //printf( "kernelAccessRecord: ievt=%d threadId=%d\n", ievt, threadIdx.x );
       return T::ieventAccessRecord( buffer, ievt ); // NB fptype and fptype_sv coincide for CUDA
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessMatrixElements.h b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessMatrixElements.h
index f32e6fea5b..3741011971 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessMatrixElements.h
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessMatrixElements.h
@@ -13,7 +13,7 @@
 #include "MemoryBuffers.h" // for HostBufferMatrixElements::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessMomenta.h b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessMomenta.h
index 29266de32c..3be229d392 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessMomenta.h
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessMomenta.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessMomenta_H
 #define MemoryAccessMomenta_H 1
@@ -13,7 +13,7 @@
 #include "MemoryAccessVectors.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -30,7 +30,7 @@ namespace mg5amcCpu
 
     // Number of Events Per Page in the momenta AOSOA memory buffer layout
     // (these are all best kept as a compile-time constants: see issue #23)
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
     // -----------------------------------------------------------------------------------------------
     // --- GPUs: neppM is best set to a power of 2 times the number of fptype's in a 32-byte cacheline
     // --- This is relevant to ensure coalesced access to momenta in global memory
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessNumerators.h b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessNumerators.h
index b152183b28..18991f4fa6 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessNumerators.h
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessNumerators.h
@@ -10,7 +10,7 @@
 #include "MemoryAccessGs.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessRandomNumbers.h b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessRandomNumbers.h
index e2988d39f3..40cb089135 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessRandomNumbers.h
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessRandomNumbers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessRandomNumbers_H
 #define MemoryAccessRandomNumbers_H 1
@@ -11,7 +11,7 @@
 #include "CPPProcess.h"
 #include "MemoryAccessHelpers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using mg5amcGpu::CPPProcess;
 #else
 using mg5amcCpu::CPPProcess;
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessVectors.h b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessVectors.h
index e9b197368e..08faccff0f 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessVectors.h
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessVectors.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessVectors_H
 #define MemoryAccessVectors_H 1
@@ -10,7 +10,7 @@
 
 #include "mgOnGpuVectors.h"
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu // this is only needed for CPU SIMD vectorization
 {
 
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessWavefunctions.h b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessWavefunctions.h
index 5428aaf933..33bef4559e 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessWavefunctions.h
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryAccessWavefunctions.h
@@ -15,7 +15,7 @@
 #define MGONGPU_TRIVIAL_WAVEFUNCTIONS 1
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryBuffers.h b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryBuffers.h
index 3093e6ed18..7756a71621 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryBuffers.h
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/MemoryBuffers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021, based on earlier work by S. Hageboeck) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryBuffers_H
 #define MemoryBuffers_H 1
@@ -11,12 +11,12 @@
 #include "mgOnGpuCxtypes.h"
 
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "Parameters_sm.h"
 
 #include <sstream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -87,7 +87,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   constexpr bool HostBufferALIGNED = false;   // ismisaligned=false
   constexpr bool HostBufferMISALIGNED = true; // ismisaligned=true
 
@@ -119,7 +119,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA pinned host buffer
   template<typename T>
   class PinnedHostBufferBase : public BufferBase<T>
@@ -128,18 +128,18 @@ namespace mg5amcCpu
     PinnedHostBufferBase( const size_t size )
       : BufferBase<T>( size, false )
     {
-      checkCuda( cudaMallocHost( &( this->m_data ), this->bytes() ) );
+      gpuMallocHost( &( this->m_data ), this->bytes() );
     }
     virtual ~PinnedHostBufferBase()
     {
-      checkCuda( cudaFreeHost( this->m_data ) );
+      gpuFreeHost( this->m_data );
     }
   };
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA device buffer
   template<typename T>
   class DeviceBufferBase : public BufferBase<T>
@@ -148,18 +148,18 @@ namespace mg5amcCpu
     DeviceBufferBase( const size_t size )
       : BufferBase<T>( size, true )
     {
-      checkCuda( cudaMalloc( &( this->m_data ), this->bytes() ) );
+      gpuMalloc( &( this->m_data ), this->bytes() );
     }
     virtual ~DeviceBufferBase()
     {
-      checkCuda( cudaFree( this->m_data ) );
+      gpuFree( this->m_data );
     }
   };
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for a given number of events
   template<typename T, size_t sizePerEvent, bool ismisaligned>
   class HostBuffer : public HostBufferBase<T, ismisaligned>, virtual private NumberOfEvents
@@ -175,7 +175,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA pinned host buffer for a given number of events
   template<typename T, size_t sizePerEvent>
   class PinnedHostBuffer : public PinnedHostBufferBase<T>, virtual private NumberOfEvents
@@ -191,7 +191,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA device buffer for a given number of events
   template<typename T, size_t sizePerEvent>
   class DeviceBuffer : public DeviceBufferBase<T>, virtual private NumberOfEvents
@@ -213,7 +213,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for momenta random numbers
   constexpr size_t sizePerEventRndNumMomenta = MemoryBuffers::np4 * MemoryBuffers::nparf;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for momenta random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumMomenta, HostBufferALIGNED> HostBufferRndNumMomenta;
 #else
@@ -232,7 +232,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer with ONE fptype per event
   constexpr size_t sizePerEventOneFp = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer with ONE fptype per event
   typedef HostBuffer<fptype, sizePerEventOneFp, HostBufferALIGNED> HostBufferOneFp;
 #else
@@ -257,7 +257,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for Gs
   constexpr size_t sizePerEventGs = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventGs, HostBufferALIGNED> HostBufferGs;
 #else
@@ -276,7 +276,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for numerators
   constexpr size_t sizePerEventNumerators = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventNumerators, HostBufferALIGNED> HostBufferNumerators;
 #else
@@ -296,7 +296,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for denominators
   constexpr size_t sizePerEventDenominators = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventDenominators, HostBufferALIGNED> HostBufferDenominators;
 #else
@@ -315,7 +315,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for random numbers
   constexpr size_t sizePerEventCouplings = MemoryBuffers::ndcoup * MemoryBuffers::nx2;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventCouplings, HostBufferALIGNED> HostBufferCouplings;
 #else
@@ -333,7 +333,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for momenta
   constexpr size_t sizePerEventMomenta = MemoryBuffers::np4 * MemoryBuffers::npar;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for momenta
   typedef HostBuffer<fptype, sizePerEventMomenta, HostBufferALIGNED> HostBufferMomenta;
   //typedef HostBuffer<fptype, sizePerEventMomenta, HostBufferMISALIGNED> HostBufferMomenta; // TEST MISALIGNMENT!
@@ -352,7 +352,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for sampling weights
   constexpr size_t sizePerEventWeights = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for sampling weights
   typedef HostBuffer<fptype, sizePerEventWeights, HostBufferALIGNED> HostBufferWeights;
 #else
@@ -370,7 +370,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for matrix elements
   constexpr size_t sizePerEventMatrixElements = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for matrix elements
   typedef HostBuffer<fptype, sizePerEventMatrixElements, HostBufferALIGNED> HostBufferMatrixElements;
 #else
@@ -385,7 +385,7 @@ namespace mg5amcCpu
   // A base class encapsulating a memory buffer for the helicity mask
   typedef BufferBase<bool> BufferHelicityMask;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for the helicity mask
   typedef HostBufferBase<bool, HostBufferALIGNED> HostBufferHelicityMask;
 #else
@@ -403,7 +403,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for wavefunctions
   constexpr size_t sizePerEventWavefunctions = MemoryBuffers::nw6 * MemoryBuffers::nx2;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for wavefunctions
   typedef HostBuffer<fptype, sizePerEventWavefunctions, HostBufferALIGNED> HostBufferWavefunctions;
 #else
@@ -421,7 +421,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for helicity random numbers
   constexpr size_t sizePerEventRndNumHelicity = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for helicity random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumHelicity, HostBufferALIGNED> HostBufferRndNumHelicity;
 #else
@@ -439,7 +439,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for color random numbers
   constexpr size_t sizePerEventRndNumColor = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for color random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumColor, HostBufferALIGNED> HostBufferRndNumColor;
 #else
@@ -457,7 +457,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for helicity selection
   constexpr size_t sizePerEventSelectedHelicity = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for helicity selection
   typedef HostBuffer<int, sizePerEventSelectedHelicity, HostBufferALIGNED> HostBufferSelectedHelicity;
 #else
@@ -475,7 +475,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for color selection
   constexpr size_t sizePerEventSelectedColor = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for color selection
   typedef HostBuffer<int, sizePerEventSelectedColor, HostBufferALIGNED> HostBufferSelectedColor;
 #else
@@ -487,7 +487,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<class Tdst, class Tsrc>
   void copyDeviceFromHost( Tdst& dst, const Tsrc& src ) // keep the same order of arguments as in memcpy
   {
@@ -504,13 +504,13 @@ namespace mg5amcCpu
       throw std::runtime_error( sstr.str() );
     }
     // NB (PR #45): cudaMemcpy involves an intermediate memcpy to pinned memory if host array is a not a pinned host array
-    checkCuda( cudaMemcpy( dst.data(), src.data(), src.bytes(), cudaMemcpyHostToDevice ) );
+    gpuMemcpy( dst.data(), src.data(), src.bytes(), gpuMemcpyHostToDevice );
   }
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<class Tdst, class Tsrc>
   void copyHostFromDevice( Tdst& dst, const Tsrc& src ) // keep the same order of arguments as in memcpy
   {
@@ -527,7 +527,7 @@ namespace mg5amcCpu
       throw std::runtime_error( sstr.str() );
     }
     // NB (PR #45): cudaMemcpy involves an intermediate memcpy to pinned memory if host array is a not a pinned host array
-    checkCuda( cudaMemcpy( dst.data(), src.data(), src.bytes(), cudaMemcpyDeviceToHost ) );
+    gpuMemcpy( dst.data(), src.data(), src.bytes(), gpuMemcpyDeviceToHost );
   }
 #endif
 
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/CPPProcess.cc b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/CPPProcess.cc
index 19bc1e7973..0bb184f0e6 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/CPPProcess.cc
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_sm.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -46,7 +45,7 @@
 // Class member functions for calculating the matrix elements for
 // Process: g g > t t~ g g WEIGHTED<=4 @1
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -80,7 +79,7 @@ namespace mg5amcCpu
   __device__ const fptype cIPD[2] = { (fptype)Parameters_sm::mdl_MT, (fptype)Parameters_sm::mdl_WT };
   __device__ const fptype* cIPC = nullptr; // unused as nicoup=0
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ fptype cIPD[2];
   __device__ __constant__ fptype* cIPC = nullptr; // unused as nicoup=0
 #else
@@ -90,7 +89,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -118,13 +117,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -151,7 +150,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -187,7 +186,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -200,8 +199,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -2417,7 +2418,7 @@ namespace mg5amcCpu
         { 62, 71, -10, 80, -1, 8, -28, 62, 62, -10, -10, -1, -1, 8, -10, -1, -64, 8, 8, -64, 80, 8, 512, -64 },
         { -28, 62, 62, -10, -10, -1, 62, 71, -10, 80, -1, 8, -10, -1, -1, 8, 8, -64, 80, 8, 8, -64, -64, 512 } }; // 2-D array[24][24]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -2474,7 +2475,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -2533,7 +2534,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -2628,8 +2629,8 @@ namespace mg5amcCpu
       { 1, 1, 1, -1, -1, 1 },
       { 1, 1, 1, -1, 1, -1 },
       { 1, 1, 1, -1, 1, 1 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -2671,9 +2672,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     const fptype tIPD[2] = { (fptype)m_pars->mdl_MT, (fptype)m_pars->mdl_WT };
     //const cxtype tIPC[0] = { ... }; // nicoup=0
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) ) );
-    //checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ) ); // nicoup=0
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) );
+    //gpuMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
 #else
     memcpy( cIPD, tIPD, 2 * sizeof( fptype ) );
     //memcpy( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
@@ -2711,7 +2712,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -2740,6 +2741,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -2753,6 +2758,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -2776,12 +2783,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -2802,7 +2809,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -2928,9 +2935,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -2954,7 +2961,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -2974,7 +2981,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 512 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -2988,9 +2995,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -3018,7 +3028,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -3228,7 +3238,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/CPPProcess.h b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/CPPProcess.h
index 04f7c62976..deb1358992 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/CPPProcess.h
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -107,7 +107,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -120,7 +120,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -150,7 +150,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/CudaRuntime.h b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/GpuAbstraction.h b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/GpuRuntime.h b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/check_sa.cc b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/check_sa.cc
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/RamboSamplingKernels.cc b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/RamboSamplingKernels.cc
index da68aa9255..79abbcc4f8 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/RamboSamplingKernels.cc
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/RamboSamplingKernels.cc
@@ -1,11 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "RamboSamplingKernels.h"
 
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "MemoryAccessMomenta.h"
 #include "MemoryAccessRandomNumbers.h"
 #include "MemoryAccessWeights.h"
@@ -14,7 +14,7 @@
 
 #include <sstream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -92,7 +92,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingKernelDevice::RamboSamplingKernelDevice( const fptype energy,               // input: energy
                                                         const BufferRndNumMomenta& rndmom, // input: random numbers in [0,1]
                                                         BufferMomenta& momenta,            // output: momenta
@@ -135,7 +135,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   getMomentaInitialDevice( const fptype energy,
                            fptype* momenta )
@@ -147,17 +147,17 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   void
   RamboSamplingKernelDevice::getMomentaInitial()
   {
-    getMomentaInitialDevice<<<m_gpublocks, m_gputhreads>>>( m_energy, m_momenta.data() );
+    gpuLaunchKernel( getMomentaInitialDevice, m_gpublocks, m_gputhreads, m_energy, m_momenta.data() );
   }
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   getMomentaFinalDevice( const fptype energy,
                          const fptype* rndmom,
@@ -171,11 +171,11 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   void
   RamboSamplingKernelDevice::getMomentaFinal()
   {
-    getMomentaFinalDevice<<<m_gpublocks, m_gputhreads>>>( m_energy, m_rndmom.data(), m_momenta.data(), m_weights.data() );
+    gpuLaunchKernel( getMomentaFinalDevice, m_gpublocks, m_gputhreads, m_energy, m_rndmom.data(), m_momenta.data(), m_weights.data() );
   }
 #endif
 
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/RamboSamplingKernels.h b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/RamboSamplingKernels.h
index 184089efd7..7c214cd74b 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/RamboSamplingKernels.h
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/RamboSamplingKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef RAMBOSAMPLINGKERNELS_H
 #define RAMBOSAMPLINGKERNELS_H 1
@@ -10,7 +10,7 @@
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -93,7 +93,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating RAMBO phase space sampling on a GPU device
   class RamboSamplingKernelDevice final : public SamplingKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/RandomNumberKernels.h b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/RandomNumberKernels.h
index 188a72c2c9..21d63beeac 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/RandomNumberKernels.h
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/RandomNumberKernels.h
@@ -1,14 +1,14 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef RANDOMNUMBERKERNELS_H
 #define RANDOMNUMBERKERNELS_H 1
 
 #include "mgOnGpuConfig.h"
 
-// NB This must come AFTER mgOnGpuConfig.h which contains our definition of __global__ when __CUDACC__ is not defined
+// NB This must come AFTER mgOnGpuConfig.h which contains our definition of __global__ when MGONGPUCPP_GPUIMPL is not defined
 #ifndef MGONGPU_HAS_NO_CURAND
 //#include "curand.h"
 struct curandGenerator_st; // forward definition from curand.h
@@ -16,7 +16,7 @@ struct curandGenerator_st; // forward definition from curand.h
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/cudacpp.mk b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/cudacpp.mk
index 509307506b..df74dfc284 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/cudacpp.mk
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/cudacpp.mk
@@ -1,7 +1,7 @@
 # Copyright (C) 2020-2023 CERN and UCLouvain.
 # Licensed under the GNU Lesser General Public License (version 3 or later).
 # Created by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-# Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+# Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #=== Determine the name of this makefile (https://ftp.gnu.org/old-gnu/Manuals/make-3.80/html_node/make_17.html)
 #=== NB: use ':=' to ensure that the value of CUDACPP_MAKEFILE is not modified further down after including make_opts
@@ -42,10 +42,10 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure common compiler flags for C++ and CUDA
+#=== Configure common compiler flags for C++ and CUDA/HIP
 
 INCFLAGS = -I.
-OPTFLAGS = -O3 # this ends up in CUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
+OPTFLAGS = -O3 # this ends up in GPUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
 
 # Dependency on src directory
 MG5AMC_COMMONLIB = mg5amc_common
@@ -121,24 +121,45 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure the CUDA compiler
+#=== Configure the GPU compiler (CUDA or HIP)
+
+# FIXME! (AV 24.01.2024)
+# In the current implementation (without separate builds for C++ and CUDA/HIP), we first check for cudacc and hipcc in CUDA_HOME and HIP_HOME.
+# If CUDA_HOME or HIP_HOME are not set, try to determine them from the path to cudacc and hipcc.
+# While convoluted, this is currently necessary to allow disabling CUDA/HIP builds by setting CUDA_HOME or HIP_HOME to invalid paths.
+# This will (probably?) be fixed when separate C++ and CUDA/HIP builds are implemented (PR #775).
 
 # If CXX is not a single word (example "clang++ --gcc-toolchain...") then disable CUDA builds (issue #505)
-# This is because it is impossible to pass this to "CUFLAGS += -ccbin <host-compiler>" below
+# This is because it is impossible to pass this to "GPUFLAGS += -ccbin <host-compiler>" below
 ifneq ($(words $(subst ccache ,,$(CXX))),1) # allow at most "CXX=ccache <host-compiler>" from outside
   $(warning CUDA builds are not supported for multi-word CXX "$(CXX)")
   override CUDA_HOME=disabled
 endif
 
-# If CUDA_HOME is not set, try to set it from the location of nvcc
+# If CUDA_HOME is not set, try to set it from the path to nvcc
 ifndef CUDA_HOME
-  CUDA_HOME = $(patsubst %bin/nvcc,%,$(shell which nvcc 2>/dev/null))
+  CUDA_HOME = $(patsubst %/bin/nvcc,%,$(shell which nvcc 2>/dev/null))
   $(warning CUDA_HOME was not set: using "$(CUDA_HOME)")
 endif
 
-# Set NVCC as $(CUDA_HOME)/bin/nvcc if it exists
+# If HIP_HOME is not set, try to set it from the path to hipcc
+ifndef HIP_HOME
+  HIP_HOME = $(patsubst %/bin/hipcc,%,$(shell which hipcc 2>/dev/null))
+  $(warning HIP_HOME was not set: using "$(HIP_HOME)")
+endif
+
+# FIXME! (AV 24.01.2024)
+# In the current implementation (without separate builds for C++ and CUDA/HIP),
+# builds are performed for HIP only if CUDA is not found in the path.
+# If both CUDA and HIP are installed, HIP builds can be triggered by unsetting CUDA_HOME.
+# This will be fixed when separate C++ and CUDA/HIP builds are implemented (PR #775).
+
+#--- Option 1: CUDA exists -> use CUDA
+
+# Set GPUCC as $(CUDA_HOME)/bin/nvcc if it exists
 ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
-  NVCC = $(CUDA_HOME)/bin/nvcc
+
+  GPUCC = $(CUDA_HOME)/bin/nvcc
   USE_NVTX ?=-DUSE_NVTX
   # See https://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/index.html
   # See https://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/
@@ -148,7 +169,7 @@ ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
   # Examples: use 60 for P100 (Piz Daint), 80 for A100 (Juwels Booster, NVidia raplab/Curiosity).
   MADGRAPH_CUDA_ARCHITECTURE ?= 70
   ###CUARCHFLAGS = -gencode arch=compute_$(MADGRAPH_CUDA_ARCHITECTURE),code=compute_$(MADGRAPH_CUDA_ARCHITECTURE) -gencode arch=compute_$(MADGRAPH_CUDA_ARCHITECTURE),code=sm_$(MADGRAPH_CUDA_ARCHITECTURE) # Older implementation (AV): go back to this one for multi-GPU support #533
-  ###CUARCHFLAGS = --gpu-architecture=compute_$(MADGRAPH_CUDA_ARCHITECTURE) --gpu-code=sm_$(MADGRAPH_CUDA_ARCHITECTURE),compute_$(MADGRAPH_CUDA_ARCHITECTURE) # Newer implementation (SH): cannot use this as-is for multi-GPU support #533
+  ###CUARCHFLAGS = --gpu-architecture=compute_$(MADGRAPH_CUDA_ARCHITECTURE) --gpu-code=sm_$(MADGRAPH_CUDA_ARCHITECTURE),compute_$(MADGRAPH_CUDA_ARCHITECTURE)  # Newer implementation (SH): cannot use this as-is for multi-GPU support #533
   comma:=,
   CUARCHFLAGS = $(foreach arch,$(subst $(comma), ,$(MADGRAPH_CUDA_ARCHITECTURE)),-gencode arch=compute_$(arch),code=compute_$(arch) -gencode arch=compute_$(arch),code=sm_$(arch))
   CUINC = -I$(CUDA_HOME)/include/
@@ -158,41 +179,79 @@ ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
     CURANDLIBFLAGS = -L$(CUDA_HOME)/lib64/ -lcurand # NB: -lcuda is not needed here!
   endif
   CUOPTFLAGS = -lineinfo
-  CUFLAGS = $(foreach opt, $(OPTFLAGS), -Xcompiler $(opt)) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
-  ###CUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
-  ###NVCC_VERSION = $(shell $(NVCC) --version | grep 'Cuda compilation tools' | cut -d' ' -f5 | cut -d, -f1)
-  CUFLAGS += -std=c++17 # need CUDA >= 11.2 (see #333): this is enforced in mgOnGpuConfig.h
+  ###GPUFLAGS = $(OPTFLAGS) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
+  GPUFLAGS = $(foreach opt, $(OPTFLAGS), -Xcompiler $(opt)) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
+  ###GPUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
+  ###GPUCC_VERSION = $(shell $(GPUCC) --version | grep 'Cuda compilation tools' | cut -d' ' -f5 | cut -d, -f1)
+  GPUFLAGS += -std=c++17 # need CUDA >= 11.2 (see #333): this is enforced in mgOnGpuConfig.h
   # Without -maxrregcount: baseline throughput: 6.5E8 (16384 32 12) up to 7.3E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 160 # improves throughput: 6.9E8 (16384 32 12) up to 7.7E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 128 # improves throughput: 7.3E8 (16384 32 12) up to 7.6E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 96 # degrades throughput: 4.1E8 (16384 32 12) up to 4.5E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 64 # degrades throughput: 1.7E8 (16384 32 12) flat at 1.7E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 160 # improves throughput: 6.9E8 (16384 32 12) up to 7.7E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 128 # improves throughput: 7.3E8 (16384 32 12) up to 7.6E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 96 # degrades throughput: 4.1E8 (16384 32 12) up to 4.5E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 64 # degrades throughput: 1.7E8 (16384 32 12) flat at 1.7E8 (65536 128 12)
+  CUBUILDRULEFLAGS = -Xcompiler -fPIC -c
+  CCBUILDRULEFLAGS = -Xcompiler -fPIC -c -x cu
+  CUDATESTFLAGS = -lcuda
+
+  # Set the host C++ compiler for GPUCC via "-ccbin <host-compiler>"
+  # (NB issue #505: this must be a single word, "clang++ --gcc-toolchain..." is not supported)
+  GPUFLAGS += -ccbin $(shell which $(subst ccache ,,$(CXX)))
+
+  # Allow newer (unsupported) C++ compilers with older versions of CUDA if ALLOW_UNSUPPORTED_COMPILER_IN_CUDA is set (#504)
+  ifneq ($(origin ALLOW_UNSUPPORTED_COMPILER_IN_CUDA),undefined)
+  GPUFLAGS += -allow-unsupported-compiler
+  endif
+
 else ifneq ($(origin REQUIRE_CUDA),undefined)
+
   # If REQUIRE_CUDA is set but no cuda is found, stop here (e.g. for CI tests on GPU #443)
-  $(error No cuda installation found (set CUDA_HOME or make nvcc visible in PATH))
+  $(error No cuda installation found (set CUDA_HOME or make GPUCC visible in PATH))
+
+#--- Option 2: CUDA does not exist, HIP exists -> use HIP
+
+# Set GPUCC as $(HIP_HOME)/bin/hipcc if it exists
+else ifneq ($(wildcard $(HIP_HOME)/bin/hipcc),)
+
+  GPUCC = $(HIP_HOME)/bin/hipcc
+  #USE_NVTX ?=-DUSE_NVTX # should maybe find something equivalent to this in HIP?
+  HIPARCHFLAGS = -target x86_64-linux-gnu --offload-arch=gfx90a
+  HIPINC = -I$(HIP_HOME)/include/
+  # Note: -DHIP_FAST_MATH is equivalent to -use_fast_math in HIP 
+  # (but only for single precision line 208: https://rocm-developer-tools.github.io/HIP/hcc__detail_2math__functions_8h_source.html)
+  # Note: CUOPTFLAGS should not be used for HIP, it had been added here but was then removed (#808)
+  GPUFLAGS = $(OPTFLAGS) $(INCFLAGS) $(HIPINC) $(HIPARCHFLAGS) -DHIP_FAST_MATH -DHIP_PLATFORM=amd -fPIC
+  ###GPUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
+  GPUFLAGS += -std=c++17
+  ###GPUFLAGS+= --maxrregcount 255 # (AV: is this option valid on HIP and meaningful on AMD GPUs?)
+  CUBUILDRULEFLAGS = -fPIC -c
+  CCBUILDRULEFLAGS = -fPIC -c -x hip
+
+else ifneq ($(origin REQUIRE_HIP),undefined)
+
+  # If REQUIRE_HIP is set but no HIP is found, stop here (e.g. for CI tests on GPU #443)
+  $(error No hip installation found (set HIP_HOME or make GPUCC visible in PATH))
+
+#--- Option 3: CUDA does not exist, HIP does not exist -> switch off both CUDA and HIP
+
 else
-  # No cuda. Switch cuda compilation off and go to common random numbers in C++
+
+  # No cudacc and no hipcc: switch CUDA and HIP compilation off and go to common random numbers in C++
   $(warning CUDA_HOME is not set or is invalid: export CUDA_HOME to compile with cuda)
-  override NVCC=
+  $(warning HIP_HOME is not set or is invalid: export HIP_HOME to compile with hip)
+  override GPUCC=
   override USE_NVTX=
   override CUINC=
   override CURANDLIBFLAGS=
-endif
-export NVCC
-export CUFLAGS
 
-# Set the host C++ compiler for nvcc via "-ccbin <host-compiler>"
-# (NB issue #505: this must be a single word, "clang++ --gcc-toolchain..." is not supported)
-CUFLAGS += -ccbin $(shell which $(subst ccache ,,$(CXX)))
-
-# Allow newer (unsupported) C++ compilers with older versions of CUDA if ALLOW_UNSUPPORTED_COMPILER_IN_CUDA is set (#504)
-ifneq ($(origin ALLOW_UNSUPPORTED_COMPILER_IN_CUDA),undefined)
-CUFLAGS += -allow-unsupported-compiler
 endif
 
+# Export GPUCC (so that it can also be used in cudacpp_src.mk?)
+export GPUCC
+export GPUFLAGS
+
 #-------------------------------------------------------------------------------
 
-#=== Configure ccache for C++ and CUDA builds
+#=== Configure ccache for C++ and CUDA/HIP builds
 
 # Enable ccache if USECCACHE=1
 ifeq ($(USECCACHE)$(shell echo $(CXX) | grep ccache),1)
@@ -201,15 +260,15 @@ endif
 #ifeq ($(USECCACHE)$(shell echo $(AR) | grep ccache),1)
 #  override AR:=ccache $(AR)
 #endif
-ifneq ($(NVCC),)
-  ifeq ($(USECCACHE)$(shell echo $(NVCC) | grep ccache),1)
-    override NVCC:=ccache $(NVCC)
+ifneq ($(GPUCC),)
+  ifeq ($(USECCACHE)$(shell echo $(GPUCC) | grep ccache),1)
+    override GPUCC:=ccache $(GPUCC)
   endif
 endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure PowerPC-specific compiler flags for C++ and CUDA
+#=== Configure PowerPC-specific compiler flags for C++ and CUDA/HIP
 
 # PowerPC-specific CXX compiler flags (being reviewed)
 ifeq ($(UNAME_P),ppc64le)
@@ -225,9 +284,9 @@ else
   ######CXXFLAGS+= -fno-semantic-interposition # no benefit (neither alone, nor combined with -flto)
 endif
 
-# PowerPC-specific CUDA compiler flags (to be reviewed!)
+# PowerPC-specific CUDA/HIP compiler flags (to be reviewed!)
 ifeq ($(UNAME_P),ppc64le)
-  CUFLAGS+= -Xcompiler -mno-float128
+  GPUFLAGS+= -Xcompiler -mno-float128
 endif
 
 #-------------------------------------------------------------------------------
@@ -235,9 +294,11 @@ endif
 #=== Configure defaults and check if user-defined choices exist for OMPFLAGS, AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
 
 # Set the default OMPFLAGS choice
-ifneq ($(shell $(CXX) --version | egrep '^Intel'),)
+ifneq ($(findstring hipcc,$(GPUCC)),)
+override OMPFLAGS = # disable OpenMP MT when using hipcc #802
+else ifneq ($(shell $(CXX) --version | egrep '^Intel'),)
 override OMPFLAGS = -fopenmp
-###override OMPFLAGS = # disable OpenMP MT on Intel (was ok without nvcc but not ok with nvcc before #578)
+###override OMPFLAGS = # disable OpenMP MT on Intel (was ok without GPUCC but not ok with GPUCC before #578)
 else ifneq ($(shell $(CXX) --version | egrep '^(clang)'),)
 override OMPFLAGS = -fopenmp
 ###override OMPFLAGS = # disable OpenMP MT on clang (was not ok without or with nvcc before #578)
@@ -293,7 +354,10 @@ endif
 
 # Set the default RNDGEN (random number generator) choice
 ifeq ($(RNDGEN),)
-  ifeq ($(NVCC),)
+  ifeq ($(GPUCC),)
+    override RNDGEN = hasNoCurand
+  # Edgecase for HIP compilation
+  else ifeq ($(findstring hipcc,$(GPUCC)),hipcc)
     override RNDGEN = hasNoCurand
   else ifeq ($(RNDGEN),)
     override RNDGEN = hasCurand
@@ -310,7 +374,7 @@ export OMPFLAGS
 
 #-------------------------------------------------------------------------------
 
-#=== Set the CUDA/C++ compiler flags appropriate to user-defined choices of AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
+#=== Set the CUDA/HIP/C++ compiler flags appropriate to user-defined choices of AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
 
 # Set the build flags appropriate to OMPFLAGS
 $(info OMPFLAGS=$(OMPFLAGS))
@@ -368,13 +432,13 @@ CXXFLAGS+= $(AVXFLAGS)
 $(info FPTYPE=$(FPTYPE))
 ifeq ($(FPTYPE),d)
   CXXFLAGS += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
-  CUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
+  GPUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
 else ifeq ($(FPTYPE),f)
   CXXFLAGS += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
-  CUFLAGS  += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
+  GPUFLAGS  += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
 else ifeq ($(FPTYPE),m)
   CXXFLAGS += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
-  CUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
+  GPUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
 else
   $(error Unknown FPTYPE='$(FPTYPE)': only 'd', 'f' and 'm' are supported)
 endif
@@ -383,7 +447,7 @@ endif
 $(info HELINL=$(HELINL))
 ifeq ($(HELINL),1)
   CXXFLAGS += -DMGONGPU_INLINE_HELAMPS
-  CUFLAGS  += -DMGONGPU_INLINE_HELAMPS
+  GPUFLAGS  += -DMGONGPU_INLINE_HELAMPS
 else ifneq ($(HELINL),0)
   $(error Unknown HELINL='$(HELINL)': only '0' and '1' are supported)
 endif
@@ -392,7 +456,7 @@ endif
 $(info HRDCOD=$(HRDCOD))
 ifeq ($(HRDCOD),1)
   CXXFLAGS += -DMGONGPU_HARDCODE_PARAM
-  CUFLAGS  += -DMGONGPU_HARDCODE_PARAM
+  GPUFLAGS  += -DMGONGPU_HARDCODE_PARAM
 else ifneq ($(HRDCOD),0)
   $(error Unknown HRDCOD='$(HRDCOD)': only '0' and '1' are supported)
 endif
@@ -444,11 +508,11 @@ ifeq ($(UNAME_S),Darwin)
   override CULIBFLAGSRPATH2 =
 else
   # RPATH to cuda/cpp libs when linking executables
-  override CXXLIBFLAGSRPATH = -Wl,-rpath,$(LIBDIRRPATH)
-  override CULIBFLAGSRPATH = -Xlinker -rpath,$(LIBDIRRPATH)
+  override CXXLIBFLAGSRPATH = -Wl,-rpath=$(LIBDIRRPATH)
+  override CULIBFLAGSRPATH = -Xlinker -rpath=$(LIBDIRRPATH)
   # RPATH to common lib when linking cuda/cpp libs
-  override CXXLIBFLAGSRPATH2 = -Wl,-rpath,'$$ORIGIN'
-  override CULIBFLAGSRPATH2 = -Xlinker -rpath,'$$ORIGIN'
+  override CXXLIBFLAGSRPATH2 = -Wl,-rpath='$$ORIGIN'
+  override CULIBFLAGSRPATH2 = -Xlinker -rpath='$$ORIGIN'
 endif
 
 # Setting LD_LIBRARY_PATH or DYLD_LIBRARY_PATH in the RUNTIME is no longer necessary (neither on Linux nor on Mac)
@@ -461,7 +525,7 @@ override RUNTIME =
 cxx_main=$(BUILDDIR)/check.exe
 fcxx_main=$(BUILDDIR)/fcheck.exe
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 cu_main=$(BUILDDIR)/gcheck.exe
 fcu_main=$(BUILDDIR)/fgcheck.exe
 else
@@ -492,14 +556,15 @@ $(BUILDDIR)/.build.$(TAG):
 	@touch $(BUILDDIR)/.build.$(TAG)
 
 # Generic target and build rules: objects from CUDA compilation
-ifneq ($(NVCC),)
+# NB: CCBUILDRULEFLAGS includes "-x cu" for nvcc and "-x hip" for hipcc (#810)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/%.o : %.cu *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $(CUBUILDRULEFLAGS) $< -o $@
 
 $(BUILDDIR)/%_cu.o : %.cc *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c -x cu $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $(CCBUILDRULEFLAGS) $< -o $@
 endif
 
 # Generic target and build rules: objects from C++ compilation
@@ -509,11 +574,14 @@ $(BUILDDIR)/%.o : %.cc *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	$(CXX) $(CPPFLAGS) $(CXXFLAGS) -fPIC -c $< -o $@
 
 # Apply special build flags only to CrossSectionKernel.cc and gCrossSectionKernel.cu (no fast math, see #117 and #516)
+# Added edgecase for HIP compilation
 ifeq ($(shell $(CXX) --version | grep ^nvc++),)
 $(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS := $(filter-out -ffast-math,$(CXXFLAGS))
 $(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS += -fno-fast-math
-ifneq ($(NVCC),)
-$(BUILDDIR)/gCrossSectionKernels.o: CUFLAGS += -Xcompiler -fno-fast-math
+ifeq ($(findstring nvcc,$(GPUCC)),nvcc)
+  $(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -Xcompiler -fno-fast-math
+else
+  $(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -fno-fast-math
 endif
 endif
 
@@ -530,10 +598,10 @@ ifeq ($(RNDGEN),hasCurand)
 $(BUILDDIR)/CurandRandomNumberKernel.o: CXXFLAGS += $(CUINC)
 endif
 
-# Avoid "warning: builtin __has_trivial_... is deprecated; use __is_trivially_... instead" in nvcc with icx2023 (#592)
+# Avoid "warning: builtin __has_trivial_... is deprecated; use __is_trivially_... instead" in GPUCC with icx2023 (#592)
 ifneq ($(shell $(CXX) --version | egrep '^(Intel)'),)
-ifneq ($(NVCC),)
-CUFLAGS += -Xcompiler -Wno-deprecated-builtins
+ifneq ($(GPUCC),)
+GPUFLAGS += -Wno-deprecated-builtins
 endif
 endif
 
@@ -541,8 +609,8 @@ endif
 # This patch does remove the warning, but I prefer to keep it disabled for the moment...
 ###ifneq ($(shell $(CXX) --version | egrep '^(clang|Apple clang|Intel)'),)
 ###$(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS += -Wno-overriding-t-option
-###ifneq ($(NVCC),)
-###$(BUILDDIR)/gCrossSectionKernels.o: CUFLAGS += -Xcompiler -Wno-overriding-t-option
+###ifneq ($(GPUCC),)
+###$(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -Xcompiler -Wno-overriding-t-option
 ###endif
 ###endif
 
@@ -569,7 +637,7 @@ MG5AMC_CXXLIB = mg5amc_$(processid_short)_cpp
 cxx_objects_lib=$(BUILDDIR)/CPPProcess.o $(BUILDDIR)/MatrixElementKernels.o $(BUILDDIR)/BridgeKernels.o $(BUILDDIR)/CrossSectionKernels.o
 cxx_objects_exe=$(BUILDDIR)/CommonRandomNumberKernel.o $(BUILDDIR)/RamboSamplingKernels.o
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 MG5AMC_CULIB = mg5amc_$(processid_short)_cuda
 cu_objects_lib=$(BUILDDIR)/gCPPProcess.o $(BUILDDIR)/gMatrixElementKernels.o $(BUILDDIR)/gBridgeKernels.o $(BUILDDIR)/gCrossSectionKernels.o
 cu_objects_exe=$(BUILDDIR)/gCommonRandomNumberKernel.o $(BUILDDIR)/gRamboSamplingKernels.o
@@ -581,11 +649,17 @@ $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: cxx_objects_lib += $(BUILDDIR)/fbridge.o
 $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib)
 	$(CXX) -shared -o $@ $(cxx_objects_lib) $(CXXLIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: $(BUILDDIR)/fbridge_cu.o
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: cu_objects_lib += $(BUILDDIR)/fbridge_cu.o
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cu_objects_lib)
-	$(NVCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+# Bypass std::filesystem completely to ease portability on LUMI #803
+#ifneq ($(findstring hipcc,$(GPUCC)),)
+#	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB) -lstdc++fs
+#else
+#	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+#endif
 endif
 
 #-------------------------------------------------------------------------------
@@ -602,16 +676,16 @@ $(cxx_main): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PAT
 $(cxx_main): $(BUILDDIR)/check_sa.o $(LIBDIR)/lib$(MG5AMC_CXXLIB).so $(cxx_objects_exe) $(BUILDDIR)/CurandRandomNumberKernel.o
 	$(CXX) -o $@ $(BUILDDIR)/check_sa.o $(OMPFLAGS) -ldl -pthread $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe) $(BUILDDIR)/CurandRandomNumberKernel.o $(CURANDLIBFLAGS)
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(cu_main): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(cu_main): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(cu_main): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(cu_main): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 else ifneq ($(shell $(CXX) --version | grep ^nvc++),) # support nvc++ #531
 $(cu_main): LIBFLAGS += -L$(patsubst %bin/nvc++,%lib,$(subst ccache ,,$(CXX))) -lnvhpcatm -lnvcpumath -lnvc
 endif
 $(cu_main): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(cu_main): $(BUILDDIR)/gcheck_sa.o $(LIBDIR)/lib$(MG5AMC_CULIB).so $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o
-	$(NVCC) -o $@ $(BUILDDIR)/gcheck_sa.o $(CUARCHFLAGS) $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o $(CURANDLIBFLAGS)
+	$(GPUCC) -o $@ $(BUILDDIR)/gcheck_sa.o $(CUARCHFLAGS) $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o $(CURANDLIBFLAGS)
 endif
 
 #-------------------------------------------------------------------------------
@@ -635,19 +709,27 @@ $(fcxx_main): LIBFLAGS += -L$(shell dirname $(shell $(FC) --print-file-name libg
 endif
 $(fcxx_main): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(fcxx_main): $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler.o $(LIBDIR)/lib$(MG5AMC_CXXLIB).so $(cxx_objects_exe)
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(BUILDDIR)/fcheck_sa.o $(OMPFLAGS) $(BUILDDIR)/fsampler.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe) -lstdc++
+else
 	$(CXX) -o $@ $(BUILDDIR)/fcheck_sa.o $(OMPFLAGS) $(BUILDDIR)/fsampler.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe)
+endif
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(fcu_main): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(fcu_main): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(fcu_main): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(fcu_main): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 endif
 ifeq ($(UNAME_S),Darwin)
 $(fcu_main): LIBFLAGS += -L$(shell dirname $(shell $(FC) --print-file-name libgfortran.dylib)) # add path to libgfortran on Mac #375
 endif
 $(fcu_main): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(fcu_main): $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBDIR)/lib$(MG5AMC_CULIB).so $(cu_objects_exe)
-	$(NVCC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe)
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) -lstdc++ -L$(shell dirname $(shell $(GPUCC) -print-prog-name=clang))/../../lib -lamdhip64
+else
+	$(GPUCC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe)
+endif
 endif
 
 #-------------------------------------------------------------------------------
@@ -659,7 +741,7 @@ $(BUILDDIR)/testxxx.o: testxxx_cc_ref.txt
 $(testmain): $(BUILDDIR)/testxxx.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/testxxx.o # Comment out this line to skip the C++ test of xxx functions
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/testxxx_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/testxxx_cu.o: INCFLAGS += $(GTESTINC)
 $(BUILDDIR)/testxxx_cu.o: testxxx_cc_ref.txt
@@ -672,7 +754,7 @@ $(BUILDDIR)/testmisc.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/testmisc.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/testmisc.o # Comment out this line to skip the C++ miscellaneous tests
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/testmisc_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/testmisc_cu.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/testmisc_cu.o
@@ -684,12 +766,12 @@ $(BUILDDIR)/runTest.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/runTest.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/runTest.o
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/runTest_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/runTest_cu.o: INCFLAGS += $(GTESTINC)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(testmain): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(testmain): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(testmain): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(testmain): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 else ifneq ($(shell $(CXX) --version | grep ^nvc++),) # support nvc++ #531
 $(testmain): LIBFLAGS += -L$(patsubst %bin/nvc++,%lib,$(subst ccache ,,$(CXX))) -lnvhpcatm -lnvcpumath -lnvc
 endif
@@ -713,14 +795,23 @@ $(testmain): LIBFLAGS += -lgomp
 endif
 endif
 
-ifeq ($(NVCC),) # link only runTest.o
+# Bypass std::filesystem completely to ease portability on LUMI #803
+#ifneq ($(findstring hipcc,$(GPUCC)),)
+#$(testmain): LIBFLAGS += -lstdc++fs
+#endif
+
+ifeq ($(GPUCC),) # link only runTest.o
 $(testmain): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(testmain): $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib) $(cxx_objects_exe) $(GTESTLIBS)
 	$(CXX) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) -ldl -pthread $(LIBFLAGS)
 else # link both runTest.o and runTest_cu.o
 $(testmain): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(testmain): $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) $(GTESTLIBS)
-	$(NVCC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) -lcuda
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) $(CUDATESTFLAGS) -lstdc++ -lpthread  -L$(shell dirname $(shell $(GPUCC) -print-prog-name=clang))/../../lib -lamdhip64
+else
+	$(GPUCC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) $(CUDATESTFLAGS)
+endif
 endif
 
 # Use target gtestlibs to build only googletest
@@ -829,9 +920,9 @@ ifeq ($(USECCACHE),1)
 	ccache --version | head -1
 endif
 	@echo ""
-	@echo NVCC=$(NVCC)
-ifneq ($(NVCC),)
-	$(NVCC) --version
+	@echo GPUCC=$(GPUCC)
+ifneq ($(GPUCC),)
+	$(GPUCC) --version
 endif
 	@echo ""
 	@echo CXX=$(CXX)
@@ -850,7 +941,7 @@ endif
 
 # Target: check (run the C++ test executable)
 # [NB THIS IS WHAT IS USED IN THE GITHUB CI!]
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 check: runTest cmpFcheck cmpFGcheck
 else
 check: runTest cmpFcheck
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/fbridge.cc b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/fbridge.cc
index 2d2b36d560..22ce3f5115 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/fbridge.cc
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/fbridge.cc
@@ -1,11 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Roiser (Oct 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "Bridge.h"
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 
 extern "C"
 {
@@ -22,7 +22,7 @@ extern "C"
    * Using the same Fortran MadEvent code, linking to the hetrerogeneous library would allow access to both CPU and GPU implementations.
    * The specific heterogeneous configuration (how many GPUs, how many threads on each CPU, etc) could be loaded in CUDA/C++ from a data file.
    */
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -46,8 +46,8 @@ extern "C"
    */
   void fbridgecreate_( CppObjectInFortran** ppbridge, const int* pnevtF, const int* pnparF, const int* pnp4F )
   {
-#ifdef __CUDACC__
-    CudaRuntime::setUp();
+#ifdef MGONGPUCPP_GPUIMPL
+    GpuRuntime::setUp();
 #endif
     // (NB: CPPProcess::initProc no longer needs to be executed here because it is called in the Bridge constructor)
     // FIXME: disable OMP in Bridge when called from Fortran
@@ -65,8 +65,8 @@ extern "C"
     Bridge<FORTRANFPTYPE>* pbridge = dynamic_cast<Bridge<FORTRANFPTYPE>*>( *ppbridge );
     if( pbridge == 0 ) throw std::runtime_error( "fbridgedelete_: invalid Bridge address" );
     delete pbridge;
-#ifdef __CUDACC__
-    CudaRuntime::tearDown();
+#ifdef MGONGPUCPP_GPUIMPL
+    GpuRuntime::tearDown();
 #endif
   }
 
@@ -96,7 +96,7 @@ extern "C"
   {
     Bridge<FORTRANFPTYPE>* pbridge = dynamic_cast<Bridge<FORTRANFPTYPE>*>( *ppbridge );
     if( pbridge == 0 ) throw std::runtime_error( "fbridgesequence_: invalid Bridge address" );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Use the device/GPU implementation in the CUDA library
     // (there is also a host implementation in this library)
     pbridge->gpu_sequence( momenta, gs, rndhel, rndcol, *pchannelId, mes, selhel, selcol );
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/fsampler.cc b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/fsampler.cc
index 2fb445372d..3743934f41 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/fsampler.cc
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/fsampler.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Feb 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "mgOnGpuConfig.h"
 
@@ -13,7 +13,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -40,7 +40,7 @@ namespace mg5amcCpu
   private:
     const int m_nevt; // The number of events in each iteration
     int m_iiter;      // The iteration counter (for random number seeding)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     HostBufferRndNumMomenta m_hstRndmom; // Memory buffers for random numbers
     HostBufferMomenta m_hstMomenta;      // Memory buffers for momenta
     HostBufferWeights m_hstWeights;      // Memory buffers for sampling weights
@@ -105,7 +105,7 @@ namespace mg5amcCpu
 
 extern "C"
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/runTest.cc b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/runTest.cc
index d4a760a71b..de327f2321 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/runTest.cc
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/runTest.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Hageboeck (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*xxx to run only testxxx.cc tests
 //----------------------------------------------------------------------------
@@ -18,7 +18,7 @@
 #include "RandomNumberKernels.h"
 #include "epoch_process_id.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using namespace mg5amcGpu;
 #else
 using namespace mg5amcCpu;
@@ -35,7 +35,7 @@ struct CUDA_CPU_TestBase : public TestDriverBase
     : TestDriverBase( npar, refFileName ) {}
 };
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 struct CPUTest : public CUDA_CPU_TestBase
 {
   // Struct data members (process, and memory structures for random numbers, momenta, matrix elements and weights on host and device)
@@ -119,7 +119,7 @@ struct CPUTest : public CUDA_CPU_TestBase
 };
 #endif
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 struct CUDATest : public CUDA_CPU_TestBase
 {
   // Reset the device when our test goes out of scope. Note that this should happen after
@@ -128,7 +128,7 @@ struct CUDATest : public CUDA_CPU_TestBase
   {
     ~DeviceReset()
     {
-      checkCuda( cudaDeviceReset() ); // this is needed by cuda-memcheck --leak-check full
+      checkGpu( gpuDeviceReset() ); // this is needed by cuda-memcheck --leak-check full
     }
   } deviceResetter;
 
@@ -256,7 +256,7 @@ INSTANTIATE_TEST_SUITE_P( prefix, \
                           test_suite_name, \
                           testing::Values( new CUDATest( MG_EPOCH_REFERENCE_FILE_NAME ) ) );
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 MG_INSTANTIATE_TEST_SUITE_GPU( XTESTID_GPU( MG_EPOCH_PROCESS_ID ), MadgraphTest );
 #else
 MG_INSTANTIATE_TEST_SUITE_CPU( XTESTID_CPU( MG_EPOCH_PROCESS_ID ), MadgraphTest );
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/testmisc.cc b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/testmisc.cc
index 895d6eeb56..ba9e59a8a3 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/testmisc.cc
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/testmisc.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*misc to run only testmisc.cc tests
 //----------------------------------------------------------------------------
@@ -17,7 +17,7 @@
 #include <sstream>
 #include <typeinfo>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #define TESTID( s ) s##_GPU_MISC
 #else
 #define TESTID( s ) s##_CPU_MISC
@@ -26,7 +26,7 @@
 #define XTESTID( s ) TESTID( s )
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -59,7 +59,7 @@ namespace mg5amcCpu
 
 TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testmisc )
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/testxxx.cc b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/testxxx.cc
index 3361fe5aa9..e5167de00c 100644
--- a/epochX/cudacpp/gg_ttgg.mad/SubProcesses/testxxx.cc
+++ b/epochX/cudacpp/gg_ttgg.mad/SubProcesses/testxxx.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Apr 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*xxx to run only testxxx.cc tests
 //----------------------------------------------------------------------------
@@ -24,7 +24,7 @@
 #include <iomanip>
 #include <iostream>
 #include <vector>
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #define TESTID( s ) s##_GPU_XXX
 #else
 #define TESTID( s ) s##_CPU_XXX
@@ -32,7 +32,7 @@
 
 #define XTESTID( s ) TESTID( s )
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -42,7 +42,7 @@ namespace mg5amcCpu
   int FPEhandlerIevt = -1;
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU): '" << FPEhandlerMessage << "' ievt=" << FPEhandlerIevt << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU neppV=" << neppV << "): '" << FPEhandlerMessage << "' ievt=" << FPEhandlerIevt << std::endl;
@@ -53,7 +53,7 @@ namespace mg5amcCpu
 
 TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -77,7 +77,7 @@ TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
   assert( nevt % neppM == 0 ); // nevt must be a multiple of neppM
   assert( nevt % neppV == 0 ); // nevt must be a multiple of neppV
   // Fill in the input momenta
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   mg5amcGpu::PinnedHostBufferMomenta hstMomenta( nevt ); // AOSOA[npagM][npar=4][np4=4][neppM]
 #else
   mg5amcCpu::HostBufferMomenta hstMomenta( nevt ); // AOSOA[npagM][npar=4][np4=4][neppM]
@@ -322,7 +322,7 @@ TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
   {
     for( int ievt = 0; ievt < nevt; ievt++ )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       using namespace mg5amcGpu;
 #else
       using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_ttgg.mad/bin/internal/launch_plugin.py b/epochX/cudacpp/gg_ttgg.mad/bin/internal/launch_plugin.py
index 3b09713e12..90af6b7053 100644
--- a/epochX/cudacpp/gg_ttgg.mad/bin/internal/launch_plugin.py
+++ b/epochX/cudacpp/gg_ttgg.mad/bin/internal/launch_plugin.py
@@ -103,7 +103,7 @@ def default_setup(self):
                        fct_mod=(self.reset_makeopts,(),{}),
                        allowed=['auto', 'none', 'sse4', 'avx2','512y','512z'])
         self.add_param('cudacpp_backend', 'CPP', include=False, hidden=False,
-                       allowed=['Fortan', 'CPP', 'CUDA'])
+                       allowed=['Fortran', 'CPP', 'CUDA'])
         self['vector_size'] = 16 # already setup in default class (just change value)
         self['aloha_flag'] = '--fast-math'
         self['matrix_flag'] = '-O3'
diff --git a/epochX/cudacpp/gg_ttgg.mad/src/HelAmps_sm.h b/epochX/cudacpp/gg_ttgg.mad/src/HelAmps_sm.h
index 8df465ad6d..8b4ad719be 100644
--- a/epochX/cudacpp/gg_ttgg.mad/src/HelAmps_sm.h
+++ b/epochX/cudacpp/gg_ttgg.mad/src/HelAmps_sm.h
@@ -5,7 +5,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -28,7 +28,7 @@
 //#include <iomanip>
 //#include <iostream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttgg.mad/src/Parameters_sm.cc b/epochX/cudacpp/gg_ttgg.mad/src/Parameters_sm.cc
index 64fc3fea62..067445b198 100644
--- a/epochX/cudacpp/gg_ttgg.mad/src/Parameters_sm.cc
+++ b/epochX/cudacpp/gg_ttgg.mad/src/Parameters_sm.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -17,7 +17,7 @@
 #include <iomanip>
 #include <iostream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using namespace mg5amcGpu;
 #else
 using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_ttgg.mad/src/Parameters_sm.h b/epochX/cudacpp/gg_ttgg.mad/src/Parameters_sm.h
index b6568d3761..9581d66e0e 100644
--- a/epochX/cudacpp/gg_ttgg.mad/src/Parameters_sm.h
+++ b/epochX/cudacpp/gg_ttgg.mad/src/Parameters_sm.h
@@ -27,7 +27,7 @@
 #include "read_slha.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -93,7 +93,7 @@ namespace mg5amcCpu
 #include <limits>
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -218,7 +218,7 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -239,7 +239,7 @@ namespace mg5amcCpu
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wunused-variable"  // e.g. <<warning: unused variable ‘mdl_G__exp__2’ [-Wunused-variable]>>
 #pragma GCC diagnostic ignored "-Wunused-parameter" // e.g. <<warning: unused parameter ‘G’ [-Wunused-parameter]>>
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #pragma nv_diagnostic push
 #pragma nv_diag_suppress 177 // e.g. <<warning #177-D: variable "mdl_G__exp__2" was declared but never referenced>>
 #endif
@@ -267,7 +267,7 @@ namespace mg5amcCpu
       // End SM implementation - no special handling of vectors of floats as in EFT (#439)
       return out;
     }
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #pragma GCC diagnostic pop
 #pragma nv_diagnostic pop
 #endif
diff --git a/epochX/cudacpp/gg_ttgg.mad/src/cudacpp_src.mk b/epochX/cudacpp/gg_ttgg.mad/src/cudacpp_src.mk
index d4cc628aec..b2b9da5288 100644
--- a/epochX/cudacpp/gg_ttgg.mad/src/cudacpp_src.mk
+++ b/epochX/cudacpp/gg_ttgg.mad/src/cudacpp_src.mk
@@ -19,7 +19,7 @@ SHELL := /bin/bash
 #=== Configure common compiler flags for CUDA and C++
 
 INCFLAGS = -I.
-OPTFLAGS = -O3 # this ends up in CUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
+OPTFLAGS = -O3 # this ends up in GPUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
 
 #-------------------------------------------------------------------------------
 
@@ -45,13 +45,13 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure the CUDA compiler (note: NVCC is already exported including ccache)
+#=== Configure the CUDA compiler (note: GPUCC is already exported including ccache)
 
-###$(info NVCC=$(NVCC))
+###$(info GPUCC=$(GPUCC))
 
 #-------------------------------------------------------------------------------
 
-#=== Configure ccache for C++ builds (note: NVCC is already exported including ccache)
+#=== Configure ccache for C++ builds (note: GPUCC is already exported including ccache)
 
 # Enable ccache if USECCACHE=1
 ifeq ($(USECCACHE)$(shell echo $(CXX) | grep ccache),1)
@@ -92,6 +92,13 @@ endif
 ###$(info OMPFLAGS=$(OMPFLAGS))
 CXXFLAGS += $(OMPFLAGS)
 
+# Add correct flags for nvcc (-x cu) and hipcc (-x hip) for GPU code (see #810)
+ifeq ($(findstring nvcc,$(GPUCC)),nvcc)
+  GPUFLAGS += -Xcompiler -fPIC -c -x cu
+else ifeq ($(findstring hipcc,$(GPUCC)),hipcc)
+  GPUFLAGS += -fPIC -c -x hip
+endif
+
 # Set the build flags appropriate to each AVX choice (example: "make AVX=none")
 # [NB MGONGPU_PVW512 is needed because "-mprefer-vector-width=256" is not exposed in a macro]
 # [See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96476]
@@ -253,20 +260,20 @@ $(BUILDDIR)/%.o : %.cc *.h $(BUILDDIR)/.build.$(TAG)
 # Generic target and build rules: objects from CUDA compilation
 $(BUILDDIR)/%_cu.o : %.cc *.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c -x cu $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $< -o $@
 
 #-------------------------------------------------------------------------------
 
 cxx_objects=$(addprefix $(BUILDDIR)/, Parameters_sm.o read_slha.o)
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 cu_objects=$(addprefix $(BUILDDIR)/, Parameters_sm_cu.o)
 endif
 
 # Target (and build rules): common (src) library
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so : $(cxx_objects) $(cu_objects)
 	@if [ ! -d $(LIBDIR) ]; then echo "mkdir -p $(LIBDIR)"; mkdir -p $(LIBDIR); fi
-	$(NVCC) -shared -o $@ $(cxx_objects) $(cu_objects) $(LDFLAGS)
+	$(GPUCC) -shared -o $@ $(cxx_objects) $(cu_objects) $(LDFLAGS)
 else
 $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so : $(cxx_objects)
 	@if [ ! -d $(LIBDIR) ]; then echo "mkdir -p $(LIBDIR)"; mkdir -p $(LIBDIR); fi
diff --git a/epochX/cudacpp/gg_ttgg.mad/src/mgOnGpuConfig.h b/epochX/cudacpp/gg_ttgg.mad/src/mgOnGpuConfig.h
index 80032e528b..6bde4466d0 100644
--- a/epochX/cudacpp/gg_ttgg.mad/src/mgOnGpuConfig.h
+++ b/epochX/cudacpp/gg_ttgg.mad/src/mgOnGpuConfig.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jul 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUCONFIG_H
 #define MGONGPUCONFIG_H 1
@@ -10,12 +10,26 @@
 // There are two different code bases for standalone_cudacpp (without multichannel) and madevent+cudacpp (with multichannel)
 #define MGONGPU_SUPPORTS_MULTICHANNEL 1
 
+// Is this a GPU (CUDA, HIP) or CPU implementation?
+#ifdef __CUDACC__
+#define MGONGPUCPP_GPUIMPL cuda
+#elif defined __HIPCC__
+#define MGONGPUCPP_GPUIMPL hip
+#include "hip/hip_runtime.h" // needed for blockDim, blockIdx, threadIdx: better in mgOnGpuConfig.h than in GpuAbstraction.h
+#else
+#undef MGONGPUCPP_GPUIMPL
+#endif
+
 // ** NB1 Throughputs (e.g. 6.8E8) are events/sec for "./gcheck.exe -p 65536 128 12"
 // ** NB2 Baseline on b7g47n0004 fluctuates (probably depends on load on other VMs)
 
 // Choose if curand is supported for generating random numbers
+// For HIP, by default, do not use curand (common random numbers will be used instead)
 // For both CUDA and C++, by default, do not inline, but allow this macro to be set from outside with e.g. -DMGONGPU_HAS_NO_CURAND
-// (there exist CUDA installations, e.g. using the HPC package, which do not include curand - see PR #784)
+// (there exist CUDA installations, e.g. using the HPC package, which do not include curand - see PR #784 and #785)
+#if defined __HIPCC__
+#define MGONGPU_HAS_NO_CURAND 1
+#else
 //#ifdef __CUDACC__
 //#undef MGONGPU_HAS_NO_CURAND // default
 ////#define MGONGPU_HAS_NO_CURAND 1
@@ -23,6 +37,7 @@
 //#undef MGONGPU_HAS_NO_CURAND // default
 ////#define MGONGPU_HAS_NO_CURAND 1
 //#endif
+#endif
 
 // Choose floating point precision (for everything but color algebra #537)
 // If one of these macros has been set from outside with e.g. -DMGONGPU_FPTYPE_FLOAT, nothing happens (issue #167)
@@ -54,23 +69,31 @@
 //#undef MGONGPU_HARDCODE_PARAM // default
 ////#define MGONGPU_HARDCODE_PARAM 1
 
-// Complex type in c++: std::complex or cxsmpl (CHOOSE ONLY ONE)
-#ifndef __CUDACC__
-//#define MGONGPU_CPPCXTYPE_STDCOMPLEX 1 // ~8 percent slower on float, same on double (5.1E6/double, 9.4E6/float)
-#define MGONGPU_CPPCXTYPE_CXSMPL 1 // new default (5.1E6/double, 10.2E6/float)
-#endif
-
-// Complex type in cuda: thrust or cucomplex or cxsmpl (CHOOSE ONLY ONE)
+// Complex type in CUDA: thrust or cucomplex or cxsmpl (CHOOSE ONLY ONE)
+// (NB THIS IS MGONGPU_*CU*CXTYPE_xxx)
 #ifdef __CUDACC__
 #define MGONGPU_CUCXTYPE_THRUST 1 // default (~1.15E9/double, ~3.2E9/float)
 //#define MGONGPU_CUCXTYPE_CUCOMPLEX 1 // ~10 percent slower (1.03E9/double, ~2.8E9/float)
 //#define MGONGPU_CUCXTYPE_CXSMPL 1 // ~10 percent slower (1.00E9/double, ~2.9E9/float)
+
+// Complex type in HIP: cxsmpl (ONLY ONE OPTION POSSIBLE? #810)
+// (NB THIS IS MGONGPU_*HIP*CXTYPE_xxx)
+#elif defined __HIPCC__
+#define MGONGPU_HIPCXTYPE_CXSMPL 1 // default for HIP
+
+// Complex type in C++: std::complex or cxsmpl (CHOOSE ONLY ONE)
+// (NB THIS IS MGONGPU_*CPP*CXTYPE_xxx)
+#else
+//#define MGONGPU_CPPCXTYPE_STDCOMPLEX 1 // ~8 percent slower on float, same on double (5.1E6/double, 9.4E6/float)
+#define MGONGPU_CPPCXTYPE_CXSMPL 1 // new default (5.1E6/double, 10.2E6/float)
 #endif
 
-// Cuda nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
+// CUDA nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
 #ifdef __CUDACC__
-#undef MGONGPU_NSIGHT_DEBUG // default
+#undef MGONGPU_NSIGHT_DEBUG // default in CUDA
 //#define MGONGPU_NSIGHT_DEBUG 1
+#else
+#undef MGONGPU_NSIGHT_DEBUG // only option in HIP or C++
 #endif
 
 // SANITY CHECKS (floating point precision for everything but color algebra #537)
@@ -86,17 +109,21 @@
 #error You cannot use double precision for color algebra and single precision elsewhere
 #endif
 
-// SANITY CHECKS (c++ complex number implementation)
-#ifndef __CUDACC__
-#if defined MGONGPU_CPPCXTYPE_STDCOMPLEX and defined MGONGPU_CPPCXTYPE_CXSMPL
-#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CPPCXTYPE_STDCOMPLEX or MGONGPU_CPPCXTYPE_CXSMPL
+// SANITY CHECKS (CUDA complex number implementation)
+#ifdef __CUDACC__
+#if defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CUCOMPLEX
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX for CUDA
+#elif defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CXSMPL for CUDA
+#elif defined MGONGPU_CUCXTYPE_CUCOMPLEX and defined MGONGPU_CUCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE OF MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL for CUDA
 #endif
 #endif
 
-// SANITY CHECKS (cuda complex number implementation)
-#ifdef __CUDACC__
-#if defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CUCOMPLEX and defined MGONGPU_CUCXTYPE_CXSMPL
-#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL
+// SANITY CHECKS (C++ complex number implementation)
+#ifndef MGONGPUCPP_GPUIMPL
+#if defined MGONGPU_CPPCXTYPE_STDCOMPLEX and defined MGONGPU_CPPCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CPPCXTYPE_STDCOMPLEX or MGONGPU_CPPCXTYPE_CXSMPL for C++
 #endif
 #endif
 
@@ -134,7 +161,7 @@ namespace mgOnGpu
   // Alignment requirement for using reinterpret_cast with SIMD vectorized code
   // (using reinterpret_cast with non aligned memory may lead to segmentation faults!)
   // Only needed for C++ code but can be enforced also in NVCC builds of C++ code using CUDA>=11.2 and C++17 (#318, #319, #333)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   constexpr int cppAlign = 64; // alignment requirement for SIMD vectorization (64-byte i.e. 512-bit)
 #endif
 
@@ -145,7 +172,7 @@ using mgOnGpu::fptype;
 using mgOnGpu::fptype2;
 
 // C++ SIMD vectorization width (this will be used to set neppV)
-#ifdef __CUDACC__ // CUDA implementation has no SIMD
+#ifdef MGONGPUCPP_GPUIMPL // CUDA and HIP implementations have no SIMD
 #undef MGONGPU_CPPSIMD
 #elif defined __AVX512VL__ && defined MGONGPU_PVW512 // C++ "512z" AVX512 with 512 width (512-bit ie 64-byte): 8 (DOUBLE) or 16 (FLOAT)
 #ifdef MGONGPU_FPTYPE_DOUBLE
@@ -175,9 +202,9 @@ using mgOnGpu::fptype2;
 #undef MGONGPU_CPPSIMD
 #endif
 
-// Cuda nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
+// CUDA nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
 // Arguments (not used so far): text is __FUNCTION__, code is 0 (start) or 1 (end)
-#if defined __CUDACC__ && defined MGONGPU_NSIGHT_DEBUG /* clang-format off */
+#if defined __CUDA__ && defined MGONGPU_NSIGHT_DEBUG /* clang-format off */
 #define mgDebugDeclare() __shared__ float mgDebugCounter[mgOnGpu::ntpbMAX];
 #define mgDebugInitialise() { mgDebugCounter[threadIdx.x] = 0; }
 #define mgDebug( code, text ) { mgDebugCounter[threadIdx.x] += 1; }
@@ -189,8 +216,8 @@ using mgOnGpu::fptype2;
 #define mgDebugFinalise() { /*noop*/ }
 #endif /* clang-format on */
 
-// Define empty CUDA declaration specifiers for C++
-#ifndef __CUDACC__
+// Define empty CUDA/HIP declaration specifiers for C++
+#ifndef MGONGPUCPP_GPUIMPL
 #define __global__
 #define __host__
 #define __device__
diff --git a/epochX/cudacpp/gg_ttgg.mad/src/mgOnGpuCxtypes.h b/epochX/cudacpp/gg_ttgg.mad/src/mgOnGpuCxtypes.h
index ca9a9f00c0..7ede1dbfae 100644
--- a/epochX/cudacpp/gg_ttgg.mad/src/mgOnGpuCxtypes.h
+++ b/epochX/cudacpp/gg_ttgg.mad/src/mgOnGpuCxtypes.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022, based on earlier work by D. Smith) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUCXTYPES_H
 #define MGONGPUCXTYPES_H 1
@@ -19,7 +19,7 @@
 #include <complex>
 
 // Complex type in cuda: thrust or cucomplex or cxsmpl
-#ifdef __CUDACC__
+#ifdef __CUDACC__ // this must be __CUDAC__ (not MGONGPUCPP_GPUIMPL)
 #if defined MGONGPU_CUCXTYPE_THRUST
 #pragma clang diagnostic push
 #pragma clang diagnostic ignored "-Wtautological-compare" // for icpx2021/clang13 (https://stackoverflow.com/a/15864661)
@@ -30,8 +30,13 @@
 #elif not defined MGONGPU_CUCXTYPE_CXSMPL
 #error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL
 #endif
+// Complex type in HIP: cxsmpl
+#elif defined __HIPCC__
+#if not defined MGONGPU_HIPCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_HIPCXTYPE_CXSMPL
+#endif
 #else
-// Complex type in c++: std::complex or cxsmpl
+// Complex type in c++ or HIP: std::complex or cxsmpl
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
 #include <cmath>
 #elif not defined MGONGPU_CPPCXTYPE_CXSMPL
@@ -82,7 +87,7 @@ namespace mgOnGpu /* clang-format off */
 using mgOnGpu::cxsmpl;
 
 // Printout to stream for user defined types
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -92,7 +97,7 @@ namespace mg5amcCpu
   inline __host__ std::ostream&
   operator<<( std::ostream& out, const cxsmpl<FP>& c )
   {
-    out << std::complex( c.real(), c.imag() );
+    out << std::complex<FP>( c.real(), c.imag() );
     return out;
   }
 
@@ -215,14 +220,14 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
   // --- Type definitions (complex type: cxtype)
-#ifdef __CUDACC__ // cuda
+#ifdef __CUDACC__ // this must be __CUDAC__ (not MGONGPUCPP_GPUIMPL)
 #if defined MGONGPU_CUCXTYPE_THRUST
   typedef thrust::complex<fptype> cxtype;
 #elif defined MGONGPU_CUCXTYPE_CUCOMPLEX
@@ -255,13 +260,13 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
-#if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+#if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
 
   //------------------------------
   // CUDA or C++ - using cxsmpl
@@ -303,11 +308,11 @@ namespace mg5amcCpu
     return cxmake( c.real(), c.imag() );
   }
 
-#endif // #if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+#endif // #if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
 
   //==========================================================================
 
-#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_THRUST // cuda + thrust
+#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_THRUST // cuda + thrust (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // CUDA - using thrust::complex
@@ -347,7 +352,7 @@ namespace mg5amcCpu
 
   //==========================================================================
 
-#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_CUCOMPLEX // cuda + cucomplex
+#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_CUCOMPLEX // cuda + cucomplex (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // CUDA - using cuComplex
@@ -566,7 +571,7 @@ namespace mg5amcCpu
 
   //==========================================================================
 
-#if not defined __CUDACC__ and defined MGONGPU_CPPCXTYPE_STDCOMPLEX // c++ + stdcomplex
+#if not defined __CUDACC__ and defined MGONGPU_CPPCXTYPE_STDCOMPLEX // c++/hip + stdcomplex (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // C++ - using std::complex
@@ -633,7 +638,7 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttgg.mad/src/mgOnGpuFptypes.h b/epochX/cudacpp/gg_ttgg.mad/src/mgOnGpuFptypes.h
index 905c97d700..fa3a02664b 100644
--- a/epochX/cudacpp/gg_ttgg.mad/src/mgOnGpuFptypes.h
+++ b/epochX/cudacpp/gg_ttgg.mad/src/mgOnGpuFptypes.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUFPTYPES_H
 #define MGONGPUFPTYPES_H 1
@@ -12,7 +12,7 @@
 #include <cmath>
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL // cuda
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -20,7 +20,7 @@ namespace mg5amcCpu
 {
   //==========================================================================
 
-#ifdef __CUDACC__ // cuda
+#ifdef MGONGPUCPP_GPUIMPL // cuda
 
   //------------------------------
   // Floating point types - Cuda
@@ -64,11 +64,11 @@ namespace mg5amcCpu
 #endif
   }
 
-#endif // #ifdef __CUDACC__
+#endif // #ifdef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
   //------------------------------
   // Floating point types - C++
@@ -92,7 +92,7 @@ namespace mg5amcCpu
     return std::sqrt( f );
   }
 
-#endif // #ifndef __CUDACC__
+#endif // #ifndef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
diff --git a/epochX/cudacpp/gg_ttgg.mad/src/mgOnGpuVectors.h b/epochX/cudacpp/gg_ttgg.mad/src/mgOnGpuVectors.h
index e1299ba81e..cdae04326b 100644
--- a/epochX/cudacpp/gg_ttgg.mad/src/mgOnGpuVectors.h
+++ b/epochX/cudacpp/gg_ttgg.mad/src/mgOnGpuVectors.h
@@ -32,7 +32,7 @@
 #endif
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -131,7 +131,7 @@ namespace mg5amcCpu
 #endif
 #endif
 
-#else // i.e #ifndef MGONGPU_CPPSIMD (this includes #ifdef __CUDACC__)
+#else // i.e #ifndef MGONGPU_CPPSIMD (this includes #ifdef MGONGPUCPP_GPUIMPL)
 
   const int neppV = 1;
 
@@ -153,13 +153,13 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
   // Printout to stream for user defined types
 
@@ -805,11 +805,11 @@ namespace mg5amcCpu
 
 #endif // #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
 
-#endif // #ifndef __CUDACC__
+#endif // #ifndef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
   //------------------------------
   // Vector types - CUDA
@@ -853,12 +853,12 @@ namespace mg5amcCpu
     return mask;
   }
 
-#endif // #ifdef __CUDACC__
+#endif // #ifdef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
   // Scalar-or-vector types: scalar in CUDA, vector or scalar in C++
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   typedef bool bool_sv;
   typedef fptype fptype_sv;
   typedef fptype2 fptype2_sv;
@@ -879,7 +879,7 @@ namespace mg5amcCpu
 #endif
 
   // Scalar-or-vector zeros: scalar in CUDA, vector or scalar in C++
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   inline __host__ __device__ cxtype cxzero_sv(){ return cxtype( 0, 0 ); }
 #elif defined MGONGPU_CPPSIMD
   inline cxtype_v cxzero_sv() { return cxtype_v(); } // RRRR=0000 IIII=0000
diff --git a/epochX/cudacpp/gg_ttgg.mad/src/rambo.h b/epochX/cudacpp/gg_ttgg.mad/src/rambo.h
index e02ea52496..cd7e1008ea 100644
--- a/epochX/cudacpp/gg_ttgg.mad/src/rambo.h
+++ b/epochX/cudacpp/gg_ttgg.mad/src/rambo.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -18,7 +18,7 @@
 #include <iostream>
 
 // Simplified rambo version for 2 to N (with N>=2) processes with massless particles
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +83,7 @@ namespace mg5amcCpu
       static bool first = true;
       if( first )
       {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
         if constexpr( M_ACCESS::isOnDevice() ) // avoid
         {
           const int ievt0 = 0;
@@ -166,7 +166,7 @@ namespace mg5amcCpu
     wt = po2log;
     if( nparf != 2 ) wt = ( 2. * nparf - 4. ) * log( energy ) + z[nparf - 1];
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     // issue warnings if weight is too small or too large
     static int iwarn[5] = { 0, 0, 0, 0, 0 };
     if( wt < -180. )
diff --git a/epochX/cudacpp/gg_ttgg.mad/src/read_slha.cc b/epochX/cudacpp/gg_ttgg.mad/src/read_slha.cc
index 055b19a779..f8e46f2e66 100644
--- a/epochX/cudacpp/gg_ttgg.mad/src/read_slha.cc
+++ b/epochX/cudacpp/gg_ttgg.mad/src/read_slha.cc
@@ -11,7 +11,11 @@
 
 #include <algorithm>
 #include <cstdlib>
-#include <filesystem>
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
 #include <fstream>
 #include <iostream>
 
@@ -60,7 +64,14 @@ SLHAReader::read_slha_file( std::string file_name, bool verbose )
     {
       std::cout << "WARNING! Card file '" << file_name << "' does not exist:"
                 << " look for the file in directory $" << envpath << "='" << getenv( envpath ) << "'" << std::endl;
+      /*
+#ifdef __HIPCC__
+      const std::string file_name2 = std::experimental::filesystem::path( getenv( envpath ) ) / std::experimental::filesystem::path( file_name ).filename();
+#else
       const std::string file_name2 = std::filesystem::path( getenv( envpath ) ) / std::filesystem::path( file_name ).filename();
+#endif
+      */
+      const std::string file_name2 = std::string( getenv( envpath ) ) + "/" + file_name; // bypass std::filesystem #803
       param_card.open( file_name2.c_str(), std::ifstream::in );
       if( param_card.good() )
       {
diff --git a/epochX/cudacpp/gg_ttgg.sa/CODEGEN_cudacpp_gg_ttgg_log.txt b/epochX/cudacpp/gg_ttgg.sa/CODEGEN_cudacpp_gg_ttgg_log.txt
index 3c3686e228..9ebee16fdf 100644
--- a/epochX/cudacpp/gg_ttgg.sa/CODEGEN_cudacpp_gg_ttgg_log.txt
+++ b/epochX/cudacpp/gg_ttgg.sa/CODEGEN_cudacpp_gg_ttgg_log.txt
@@ -52,7 +52,7 @@ Note that you can still compile and run aMC@NLO with the built-in PDFs
 
 Using default text editor "vi". Set another one in ./input/mg5_configuration.txt
 Using default eps viewer "evince". Set another one in ./input/mg5_configuration.txt
-No valid web browser found. Please set in ./input/mg5_configuration.txt
+Using default web browser "firefox". Set another one in ./input/mg5_configuration.txt
 import /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_cudacpp_gg_ttgg.mg
 The import format was not given, so we guess it as command
 set stdout_level DEBUG
@@ -62,7 +62,7 @@ generate g g > t t~ g g
 No model currently active, so we import the Standard Model
 INFO: load particles 
 INFO: load vertices 
-DEBUG: model prefixing  takes 0.005596637725830078 
+DEBUG: model prefixing  takes 0.005690574645996094 
 INFO: Restrict model sm with file models/sm/restrict_default.dat . 
 DEBUG: Simplifying conditional expressions 
 DEBUG: remove interactions: u s w+ at order: QED=1 
@@ -155,7 +155,7 @@ INFO: Please specify coupling orders to bypass this step.
 INFO: Trying coupling order WEIGHTED<=4: WEIGTHED IS QCD+2*QED 
 INFO: Trying process: g g > t t~ g g WEIGHTED<=4 @1  
 INFO: Process has 123 diagrams 
-1 processes with 123 diagrams generated in 0.166 s
+1 processes with 123 diagrams generated in 0.157 s
 Total: 1 processes with 123 diagrams
 output standalone_cudacpp ../TMPOUT/CODEGEN_cudacpp_gg_ttgg
 Load PLUGIN.CUDACPP_OUTPUT
@@ -175,7 +175,7 @@ INFO: Creating files in directory /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TM
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_cudacpp_gg_ttgg/SubProcesses/P1_Sigma_sm_gg_ttxgg/./CPPProcess.h
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_cudacpp_gg_ttgg/SubProcesses/P1_Sigma_sm_gg_ttxgg/./CPPProcess.cc
 INFO: Created files CPPProcess.h and CPPProcess.cc in directory /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_cudacpp_gg_ttgg/SubProcesses/P1_Sigma_sm_gg_ttxgg/. 
-Generated helas calls for 1 subprocesses (123 diagrams) in 0.442 s
+Generated helas calls for 1 subprocesses (123 diagrams) in 0.432 s
 DEBUG:  Entering PLUGIN_ProcessExporter.convert_model (create the model) [output.py at line 202] 
 ALOHA: aloha starts to compute helicity amplitudes
 ALOHA: aloha creates VVV1 routines
@@ -183,7 +183,7 @@ ALOHA: aloha creates FFV1 routines
 ALOHA: aloha creates VVVV1 routines
 ALOHA: aloha creates VVVV3 routines
 ALOHA: aloha creates VVVV4 routines
-ALOHA: aloha creates 5 routines in  0.337 s
+ALOHA: aloha creates 5 routines in  0.323 s
 <class 'aloha.create_aloha.AbstractRoutine'> VVV1
 <class 'aloha.create_aloha.AbstractRoutine'> VVV1
 <class 'aloha.create_aloha.AbstractRoutine'> FFV1
@@ -206,7 +206,7 @@ INFO: Created files Parameters_sm.h and Parameters_sm.cc in directory
 INFO: /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_cudacpp_gg_ttgg/src/. and /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_cudacpp_gg_ttgg/src/. 
 quit
 
-real	0m1.506s
-user	0m1.438s
-sys	0m0.059s
+real	0m1.851s
+user	0m1.398s
+sys	0m0.060s
 Code generation completed in 2 seconds
diff --git a/epochX/cudacpp/gg_ttgg.sa/COPYRIGHT b/epochX/cudacpp/gg_ttgg.sa/COPYRIGHT
index a134b5fef9..9036d9260a 100644
--- a/epochX/cudacpp/gg_ttgg.sa/COPYRIGHT
+++ b/epochX/cudacpp/gg_ttgg.sa/COPYRIGHT
@@ -15,6 +15,7 @@ The full development team currently includes the following authors :
   Stephan Hageboeck (CERN)
   Olivier Mattelaer (Universite Catholique de Louvain, original author)
   Stefan Roiser (CERN, original author)
+  Jorgen Teig (CERN)
   Andrea Valassi (CERN, original author)
   Zenny Wettersten (CERN)
 See https://github.com/madgraph5/madgraph4gpu for more details. For the full
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/Bridge.h b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/Bridge.h
index bf8b5e024d..f9ed70dfde 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/Bridge.h
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/Bridge.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Roiser (Nov 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef BRIDGE_H
 #define BRIDGE_H 1
@@ -14,16 +14,23 @@
 #include "MemoryAccessMomenta.h"  // for MemoryAccessMomenta::neppM
 #include "MemoryBuffers.h"        // for HostBufferMomenta, DeviceBufferMomenta etc
 
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
+
+#include <sys/stat.h> // bypass std::filesystem #803
+
 #include <algorithm>
 #include <cassert>
 #include <cmath>
 #include <cstring>
-#include <filesystem>
 #include <iostream>
 #include <memory>
 #include <type_traits>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +90,7 @@ namespace mg5amcCpu
     Bridge& operator=( const Bridge& ) = delete;
     Bridge& operator=( Bridge&& ) = delete;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     /**
      * Set the gpublocks and gputhreads for the gpusequence - throws if evnt != gpublocks*gputhreads
      * (this is needed for BridgeKernel tests rather than for actual production use in Fortran)
@@ -150,7 +157,7 @@ namespace mg5amcCpu
     unsigned int m_nevt; // number of events
     int m_nGoodHel;      // the number of good helicities (-1 initially when they have not yet been calculated)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     int m_gputhreads; // number of gpu threads (default set from number of events, can be modified)
     int m_gpublocks;  // number of gpu blocks (default set from number of events, can be modified)
     DeviceBuffer<FORTRANFPTYPE, sizePerEventMomenta> m_devMomentaF;
@@ -187,12 +194,12 @@ namespace mg5amcCpu
   // Forward declare transposition methods
   //
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
   template<typename Tin, typename Tout>
   __global__ void dev_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt );
 
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
 
   template<typename Tin, typename Tout>
   void hst_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt );
@@ -209,7 +216,7 @@ namespace mg5amcCpu
   Bridge<FORTRANFPTYPE>::Bridge( unsigned int nevtF, unsigned int nparF, unsigned int np4F )
     : m_nevt( nevtF )
     , m_nGoodHel( -1 )
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     , m_gputhreads( 256 )                  // default number of gpu threads
     , m_gpublocks( m_nevt / m_gputhreads ) // this ensures m_nevt <= m_gpublocks*m_gputhreads
     , m_devMomentaF( m_nevt )
@@ -233,7 +240,7 @@ namespace mg5amcCpu
   {
     if( nparF != CPPProcess::npar ) throw std::runtime_error( "Bridge constructor: npar mismatch" );
     if( np4F != CPPProcess::np4 ) throw std::runtime_error( "Bridge constructor: np4 mismatch" );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( ( m_nevt < s_gputhreadsmin ) || ( m_nevt % s_gputhreadsmin != 0 ) )
       throw std::runtime_error( "Bridge constructor: nevt should be a multiple of " + std::to_string( s_gputhreadsmin ) );
     while( m_nevt != m_gpublocks * m_gputhreads )
@@ -249,20 +256,28 @@ namespace mg5amcCpu
 #else
     std::cout << "WARNING! Instantiate host Bridge (nevt=" << m_nevt << ")" << std::endl;
     m_pmek.reset( new MatrixElementKernelHost( m_hstMomentaC, m_hstGs, m_hstRndHel, m_hstRndCol, m_hstMEs, m_hstSelHel, m_hstSelCol, m_nevt ) );
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
     // Create a process object, read param card and set parameters
     // FIXME: the process instance can happily go out of scope because it is only needed to read parameters?
     // FIXME: the CPPProcess should really be a singleton? what if fbridgecreate is called from several Fortran threads?
     CPPProcess process( /*verbose=*/false );
     std::string paramCard = "../../Cards/param_card.dat";
-    if( !std::filesystem::exists( paramCard ) )
-    {
-      paramCard = "../" + paramCard;
-    }
+    /*
+#ifdef __HIPCC__
+    if( !std::experimental::filesystem::exists( paramCard ) ) paramCard = "../" + paramCard;
+#else
+    if( !std::filesystem::exists( paramCard ) ) paramCard = "../" + paramCard;
+#endif
+    */
+    //struct stat dummybuffer; // bypass std::filesystem #803
+    //if( !( stat( paramCard.c_str(), &dummyBuffer ) == 0 ) ) paramCard = "../" + paramCard; //
+    auto fileExists = []( std::string& fileName )
+    { struct stat buffer; return stat( fileName.c_str(), &buffer ) == 0; };
+    if( !fileExists( paramCard ) ) paramCard = "../" + paramCard; // bypass std::filesystem #803
     process.initProc( paramCard );
   }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::set_gpugrid( const int gpublocks, const int gputhreads )
   {
@@ -276,7 +291,7 @@ namespace mg5amcCpu
   }
 #endif
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::gpu_sequence( const FORTRANFPTYPE* momenta,
                                             const FORTRANFPTYPE* gs,
@@ -291,14 +306,14 @@ namespace mg5amcCpu
     constexpr int neppM = MemoryAccessMomenta::neppM;
     if constexpr( neppM == 1 && std::is_same_v<FORTRANFPTYPE, fptype> )
     {
-      checkCuda( cudaMemcpy( m_devMomentaC.data(), momenta, m_devMomentaC.bytes(), cudaMemcpyHostToDevice ) );
+      gpuMemcpy( m_devMomentaC.data(), momenta, m_devMomentaC.bytes(), gpuMemcpyHostToDevice );
     }
     else
     {
-      checkCuda( cudaMemcpy( m_devMomentaF.data(), momenta, m_devMomentaF.bytes(), cudaMemcpyHostToDevice ) );
+      gpuMemcpy( m_devMomentaF.data(), momenta, m_devMomentaF.bytes(), gpuMemcpyHostToDevice );
       const int thrPerEvt = CPPProcess::npar * CPPProcess::np4; // AV: transpose alg does 1 element per thread (NOT 1 event per thread)
       //const int thrPerEvt = 1; // AV: try new alg with 1 event per thread... this seems slower
-      dev_transposeMomentaF2C<<<m_gpublocks * thrPerEvt, m_gputhreads>>>( m_devMomentaF.data(), m_devMomentaC.data(), m_nevt );
+      gpuLaunchKernel( dev_transposeMomentaF2C, m_gpublocks * thrPerEvt, m_gputhreads, m_devMomentaF.data(), m_devMomentaC.data(), m_nevt );
     }
     if constexpr( std::is_same_v<FORTRANFPTYPE, fptype> )
     {
@@ -341,7 +356,7 @@ namespace mg5amcCpu
   }
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::cpu_sequence( const FORTRANFPTYPE* momenta,
                                             const FORTRANFPTYPE* gs,
@@ -396,7 +411,7 @@ namespace mg5amcCpu
   // - C++ array: momenta[npagM][npar][np4][neppM] with nevt=npagM*neppM (AOSOA)
   //
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename Tin, typename Tout>
   __global__ void dev_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt )
   {
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/BridgeKernels.cc b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/BridgeKernels.cc
index d58066c9c1..eaf4037a24 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/BridgeKernels.cc
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/BridgeKernels.cc
@@ -1,17 +1,18 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "BridgeKernels.h"
 
+#include "GpuAbstraction.h"
 #include "MemoryAccessMomenta.h"
 
 #include <sstream>
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -45,7 +46,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu
 {
 
@@ -96,7 +97,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/BridgeKernels.h b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/BridgeKernels.h
index 15eb4bff4d..3efef8ce97 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/BridgeKernels.h
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/BridgeKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef BRIDGEKERNELS_H
 #define BRIDGEKERNELS_H 1
@@ -12,7 +12,7 @@
 #include "MatrixElementKernels.h"
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -49,7 +49,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A Bridge wrapper class encapsulating matrix element calculations on a CPU host
   class BridgeKernelHost final : public BridgeKernelBase
   {
@@ -89,7 +89,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A Bridge wrapper class encapsulating matrix element calculations on a GPU device
   class BridgeKernelDevice : public BridgeKernelBase
   {
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/CommonRandomNumberKernel.cc b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/CommonRandomNumberKernel.cc
index 985b39f576..010bc4cbd0 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/CommonRandomNumberKernel.cc
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/CommonRandomNumberKernel.cc
@@ -1,15 +1,16 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "CommonRandomNumbers.h"
+#include "GpuAbstraction.h"
 #include "MemoryBuffers.h"
 #include "RandomNumberKernels.h"
 
 #include <cassert>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/CrossSectionKernels.cc b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/CrossSectionKernels.cc
index 0b355a3c8d..c15b39844d 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/CrossSectionKernels.cc
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/CrossSectionKernels.cc
@@ -1,10 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "CrossSectionKernels.h"
 
+#include "GpuAbstraction.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessWeights.h"
 #include "MemoryBuffers.h"
@@ -77,7 +78,7 @@ debug_me_is_abnormal( const fptype& me, size_t ievtALL )
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -185,7 +186,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/CrossSectionKernels.h b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/CrossSectionKernels.h
index 7933ca4bbf..4d9659e04e 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/CrossSectionKernels.h
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/CrossSectionKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef CROSSSECTIONKERNELS_H
 #define CROSSSECTIONKERNELS_H 1
@@ -13,7 +13,7 @@
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -96,7 +96,7 @@ namespace mg5amcCpu
   //--------------------------------------------------------------------------
 
   /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating the calculation of event statistics on a GPU device
   class CrossSectionKernelDevice : public CrossSectionKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/CudaRuntime.h b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/CudaRuntime.h
deleted file mode 100644
index 64ce52f4b3..0000000000
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/CudaRuntime.h
+++ /dev/null
@@ -1,85 +0,0 @@
-// Copyright (C) 2020-2023 CERN and UCLouvain.
-// Licensed under the GNU Lesser General Public License (version 3 or later).
-// Created by: S. Roiser (Jul 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
-
-#ifndef MG5AMC_CUDARUNTIME_H
-#define MG5AMC_CUDARUNTIME_H 1
-
-// MG5AMC on GPU uses the CUDA runtime API, not the lower level CUDA driver API
-// See https://docs.nvidia.com/cuda/cuda-runtime-api/driver-vs-runtime-api.html#driver-vs-runtime-api
-
-#include <cassert>
-#include <iostream>
-
-//--------------------------------------------------------------------------
-
-// See https://stackoverflow.com/a/14038590
-#ifdef __CUDACC__ /* clang-format off */
-#define checkCuda( code ) { assertCuda( code, __FILE__, __LINE__ ); }
-inline void assertCuda( cudaError_t code, const char* file, int line, bool abort = true )
-{
-  if( code != cudaSuccess )
-  {
-    printf( "ERROR! assertCuda: '%s' (%d) in %s:%d\n", cudaGetErrorString( code ), code, file, line );
-    if( abort ) assert( code == cudaSuccess );
-  }
-}
-#endif /* clang-format on */
-
-//--------------------------------------------------------------------------
-
-#ifdef __CUDACC__
-namespace mg5amcGpu
-{
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  // *** FIXME! This will all need to be designed differently when going to multi-GPU nodes! ***
-  struct CudaRuntime final
-  {
-    CudaRuntime( const bool debug = true )
-      : m_debug( debug ) { setUp( m_debug ); }
-    ~CudaRuntime() { tearDown( m_debug ); }
-    CudaRuntime( const CudaRuntime& ) = delete;
-    CudaRuntime( CudaRuntime&& ) = delete;
-    CudaRuntime& operator=( const CudaRuntime& ) = delete;
-    CudaRuntime& operator=( CudaRuntime&& ) = delete;
-    bool m_debug;
-
-    // Set up CUDA application
-    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
-    // Calling cudaSetDevice on startup is useful to properly book-keep the time spent in CUDA initialization
-    static void setUp( const bool debug = true )
-    {
-      // ** NB: it is useful to call cudaSetDevice, or cudaFree, to properly book-keep the time spent in CUDA initialization
-      // ** NB: otherwise, the first CUDA operation (eg a cudaMemcpyToSymbol in CPPProcess ctor) appears to take much longer!
-      /*
-      // [We initially added cudaFree(0) to "ease profile analysis" only because it shows up as a big recognizable block!]
-      // No explicit initialization is needed: https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#initialization
-      // It is not clear what cudaFree(0) does at all: https://stackoverflow.com/questions/69967813/
-      if ( debug ) std::cout << "__CudaRuntime: calling cudaFree(0)" << std::endl;
-      checkCuda( cudaFree( 0 ) ); // SLOW!
-      */
-      // Replace cudaFree(0) by cudaSetDevice(0), even if it is not really needed either
-      // (but see https://developer.nvidia.com/blog/cuda-pro-tip-always-set-current-device-avoid-multithreading-bugs)
-      if( debug ) std::cout << "__CudaRuntime: calling cudaSetDevice(0)" << std::endl;
-      checkCuda( cudaSetDevice( 0 ) ); // SLOW!
-    }
-
-    // Tear down CUDA application (call cudaDeviceReset)
-    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
-    // Calling cudaDeviceReset on shutdown is only needed for checking memory leaks in cuda-memcheck
-    // See https://docs.nvidia.com/cuda/cuda-memcheck/index.html#leak-checking
-    static void tearDown( const bool debug = true )
-    {
-      if( debug ) std::cout << "__CudaRuntime: calling cudaDeviceReset()" << std::endl;
-      checkCuda( cudaDeviceReset() );
-    }
-  };
-
-}
-#endif
-
-//--------------------------------------------------------------------------
-
-#endif // MG5AMC_CUDARUNTIME_H
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/CurandRandomNumberKernel.cc b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/CurandRandomNumberKernel.cc
index eb56333b03..08a16f6f2c 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/CurandRandomNumberKernel.cc
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/CurandRandomNumberKernel.cc
@@ -1,9 +1,9 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "MemoryBuffers.h"
 #include "RandomNumberKernels.h"
 
@@ -22,7 +22,7 @@ inline void assertCurand( curandStatus_t code, const char *file, int line, bool
 }
 #endif /* clang-format on */
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -36,7 +36,7 @@ namespace mg5amcCpu
   {
     if( m_isOnDevice )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if( !m_rnarray.isOnDevice() )
         throw std::runtime_error( "CurandRandomNumberKernel on device with a host random number array" );
 #else
@@ -114,7 +114,7 @@ namespace mg5amcCpu
     /*
     printf( "\nCurandRandomNumberKernel::generateRnarray size = %d\n", (int)m_rnarray.size() );
     fptype* data = m_rnarray.data();
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( m_rnarray.isOnDevice() )
     {
       data = new fptype[m_rnarray.size()]();
@@ -123,7 +123,7 @@ namespace mg5amcCpu
 #endif
     for( int i = 0; i < ( (int)m_rnarray.size() / 4 ); i++ )
       printf( "[%4d] %f %f %f %f\n", i * 4, data[i * 4], data[i * 4 + 2], data[i * 4 + 2], data[i * 4 + 3] );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( m_rnarray.isOnDevice() ) delete[] data;
 #endif
     */
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/EventStatistics.h b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/EventStatistics.h
index 48b51e0a49..b425a5bade 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/EventStatistics.h
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/EventStatistics.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef EventStatistics_H
 #define EventStatistics_H 1
@@ -16,7 +16,7 @@
 #include <limits>
 #include <string>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/GpuAbstraction.h b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/GpuAbstraction.h
new file mode 100644
index 0000000000..9c467b1e04
--- /dev/null
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/GpuAbstraction.h
@@ -0,0 +1,69 @@
+// Copyright (C) 2020-2023 CERN and UCLouvain.
+// Licensed under the GNU Lesser General Public License (version 3 or later).
+// Created by: J. Teig (Jul 2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+
+#ifndef MG5AMC_GPUABSTRACTION_H
+#define MG5AMC_GPUABSTRACTION_H 1
+
+#include <cassert>
+
+//--------------------------------------------------------------------------
+
+#ifdef __CUDACC__
+
+#define gpuError_t cudaError_t
+#define gpuPeekAtLastError cudaPeekAtLastError
+#define gpuGetErrorString cudaGetErrorString
+#define gpuSuccess cudaSuccess
+
+#define gpuMallocHost( ptr, size ) checkGpu( cudaMallocHost( ptr, size ) )
+#define gpuMalloc( ptr, size ) checkGpu( cudaMalloc( ptr, size ) )
+
+#define gpuMemcpy( dstData, srcData, srcBytes, func ) checkGpu( cudaMemcpy( dstData, srcData, srcBytes, func ) )
+#define gpuMemcpyHostToDevice cudaMemcpyHostToDevice
+#define gpuMemcpyDeviceToHost cudaMemcpyDeviceToHost
+#define gpuMemcpyToSymbol( type1, type2, size ) checkGpu( cudaMemcpyToSymbol( type1, type2, size ) )
+
+#define gpuFree( ptr ) checkGpu( cudaFree( ptr ) )
+#define gpuFreeHost( ptr ) checkGpu( cudaFreeHost( ptr ) )
+
+#define gpuSetDevice cudaSetDevice
+#define gpuDeviceSynchronize cudaDeviceSynchronize
+#define gpuDeviceReset cudaDeviceReset
+
+#define gpuLaunchKernel( kernel, blocks, threads, ... ) kernel<<<blocks, threads>>>( __VA_ARGS__ )
+#define gpuLaunchKernelSharedMem( kernel, blocks, threads, sharedMem, ... ) kernel<<<blocks, threads, sharedMem>>>( __VA_ARGS__ )
+
+//--------------------------------------------------------------------------
+
+#elif defined __HIPCC__
+
+#define gpuError_t hipError_t
+#define gpuPeekAtLastError hipPeekAtLastError
+#define gpuGetErrorString hipGetErrorString
+#define gpuSuccess hipSuccess
+
+#define gpuMallocHost( ptr, size ) checkGpu( hipHostMalloc( ptr, size ) ) // HostMalloc better
+#define gpuMalloc( ptr, size ) checkGpu( hipMalloc( ptr, size ) )
+
+#define gpuMemcpy( dstData, srcData, srcBytes, func ) checkGpu( hipMemcpy( dstData, srcData, srcBytes, func ) )
+#define gpuMemcpyHostToDevice hipMemcpyHostToDevice
+#define gpuMemcpyDeviceToHost hipMemcpyDeviceToHost
+#define gpuMemcpyToSymbol( type1, type2, size ) checkGpu( hipMemcpyToSymbol( type1, type2, size ) )
+
+#define gpuFree( ptr ) checkGpu( hipFree( ptr ) )
+#define gpuFreeHost( ptr ) checkGpu( hipHostFree( ptr ) )
+
+#define gpuSetDevice hipSetDevice
+#define gpuDeviceSynchronize hipDeviceSynchronize
+#define gpuDeviceReset hipDeviceReset
+
+#define gpuLaunchKernel( kernel, blocks, threads, ... ) kernel<<<blocks, threads>>>( __VA_ARGS__ )
+#define gpuLaunchKernelSharedMem( kernel, blocks, threads, sharedMem, ... ) kernel<<<blocks, threads, sharedMem>>>( __VA_ARGS__ )
+
+//--------------------------------------------------------------------------
+
+#endif
+
+#endif // MG5AMC_GPUABSTRACTION_H
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/GpuRuntime.h b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/GpuRuntime.h
new file mode 100644
index 0000000000..93579ef08b
--- /dev/null
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/GpuRuntime.h
@@ -0,0 +1,85 @@
+// Copyright (C) 2020-2023 CERN and UCLouvain.
+// Licensed under the GNU Lesser General Public License (version 3 or later).
+// Created by: J. Teig (Jun 2023, based on earlier work by S. Roiser) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+
+#ifndef MG5AMC_GPURUNTIME_H
+#define MG5AMC_GPURUNTIME_H 1
+
+// MG5AMC on GPU uses the CUDA runtime API, not the lower level CUDA driver API
+// See https://docs.nvidia.com/cuda/cuda-runtime-api/driver-vs-runtime-api.html#driver-vs-runtime-api
+
+#include "GpuAbstraction.h"
+
+#include <iostream>
+
+//--------------------------------------------------------------------------
+
+// See https://stackoverflow.com/a/14038590
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
+#define checkGpu( code ) { assertGpu( code, __FILE__, __LINE__ ); }
+inline void assertGpu( gpuError_t code, const char* file, int line, bool abort = true )
+{
+  if( code != gpuSuccess )
+  {
+    printf( "ERROR! assertGpu: '%s' (%d) in %s:%d\n", gpuGetErrorString( code ), code, file, line );
+    if( abort ) assert( code == gpuSuccess );
+  }
+}
+#endif /* clang-format on */
+
+//--------------------------------------------------------------------------
+
+#ifdef MGONGPUCPP_GPUIMPL
+namespace mg5amcGpu
+{
+  // Instantiate a GpuRuntime at the beginnining of the application's main to
+  // invoke gpuSetDevice(0) in the constructor and book a gpuDeviceReset() call in the destructor
+  // *** FIXME! This will all need to be designed differently when going to multi-GPU nodes! ***
+  struct GpuRuntime final
+  {
+    GpuRuntime( const bool debug = true )
+      : m_debug( debug ) { setUp( m_debug ); }
+    ~GpuRuntime() { tearDown( m_debug ); }
+    GpuRuntime( const GpuRuntime& ) = delete;
+    GpuRuntime( GpuRuntime&& ) = delete;
+    GpuRuntime& operator=( const GpuRuntime& ) = delete;
+    GpuRuntime& operator=( GpuRuntime&& ) = delete;
+    bool m_debug;
+
+    // Set up CUDA application
+    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
+    // Calling cudaSetDevice on startup is useful to properly book-keep the time spent in CUDA initialization
+    static void setUp( const bool debug = true )
+    {
+      // ** NB: it is useful to call cudaSetDevice, or cudaFree, to properly book-keep the time spent in CUDA initialization
+      // ** NB: otherwise, the first CUDA operation (eg a cudaMemcpyToSymbol in CPPProcess ctor) appears to take much longer!
+      /*
+      // [We initially added cudaFree(0) to "ease profile analysis" only because it shows up as a big recognizable block!]
+      // No explicit initialization is needed: https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#initialization
+      // It is not clear what cudaFree(0) does at all: https://stackoverflow.com/questions/69967813/
+      if ( debug ) std::cout << "__CudaRuntime: calling cudaFree(0)" << std::endl;
+      checkCuda( cudaFree( 0 ) ); // SLOW!
+      */
+      // Replace cudaFree(0) by cudaSetDevice(0), even if it is not really needed either
+      // (but see https://developer.nvidia.com/blog/cuda-pro-tip-always-set-current-device-avoid-multithreading-bugs)
+      if( debug ) std::cout << "__GpuRuntime: calling GpuSetDevice(0)" << std::endl;
+      checkGpu( gpuSetDevice( 0 ) ); // SLOW!
+    }
+
+    // Tear down CUDA application (call cudaDeviceReset)
+    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
+    // Calling cudaDeviceReset on shutdown is only needed for checking memory leaks in cuda-memcheck
+    // See https://docs.nvidia.com/cuda/cuda-memcheck/index.html#leak-checking
+    static void tearDown( const bool debug = true )
+    {
+      if( debug ) std::cout << "__GpuRuntime: calling GpuDeviceReset()" << std::endl;
+      checkGpu( gpuDeviceReset() );
+    }
+  };
+}
+#endif
+
+//--------------------------------------------------------------------------
+
+#endif // MG5AMC_GPURUNTIME_H
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MadgraphTest.h b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MadgraphTest.h
index ef40624c88..6054185300 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MadgraphTest.h
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MadgraphTest.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Hageboeck (Dec 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MADGRAPHTEST_H_
 #define MADGRAPHTEST_H_ 1
@@ -14,7 +14,11 @@
 
 #include <array>
 #include <cmath>
-#include <filesystem>
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
 #include <fstream>
 #include <iomanip>
 #include <memory>
@@ -22,7 +26,7 @@
 #include <string>
 #include <vector>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using mg5amcGpu::CPPProcess;
 #else
 using mg5amcCpu::CPPProcess;
@@ -201,7 +205,7 @@ class MadgraphTest : public testing::TestWithParam<TestDriverBase*>
 
 // Since we link both the CPU-only and GPU tests into the same executable, we prevent
 // a multiply defined symbol by only compiling this in the non-CUDA phase:
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
 /// Compare momenta and matrix elements.
 /// This uses an implementation of TestDriverBase to run a madgraph workflow,
@@ -219,7 +223,14 @@ TEST_P( MadgraphTest, CompareMomentaAndME )
   const char* dumpEventsC = getenv( "CUDACPP_RUNTEST_DUMPEVENTS" );
   const bool dumpEvents = ( dumpEventsC != 0 ) && ( std::string( dumpEventsC ) != "" );
   const std::string refFileName = testDriver->getRefFileName();
+  /*
+#ifdef __HIPCC__
+  const std::string dumpFileName = std::experimental::filesystem::path( refFileName ).filename();
+#else
   const std::string dumpFileName = std::filesystem::path( refFileName ).filename();
+#endif
+  */
+  const std::string dumpFileName = refFileName; // bypass std::filesystem #803
   std::ofstream dumpFile;
   if( dumpEvents )
   {
@@ -307,6 +318,6 @@ TEST_P( MadgraphTest, CompareMomentaAndME )
   }
 }
 
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
 
 #endif /* MADGRAPHTEST_H_ */
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MatrixElementKernels.cc b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MatrixElementKernels.cc
index 74b5239ebf..81699dfea9 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MatrixElementKernels.cc
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MatrixElementKernels.cc
@@ -1,12 +1,12 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "MatrixElementKernels.h"
 
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h" // Includes the abstraction for Nvidia/AMD compilation
 #include "MemoryAccessMomenta.h"
 #include "MemoryBuffers.h"
 
@@ -14,7 +14,7 @@
 
 //============================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu
 {
 
@@ -150,7 +150,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
@@ -209,13 +209,13 @@ namespace mg5amcGpu
     PinnedHostBufferHelicityMask hstIsGoodHel( ncomb );
     DeviceBufferHelicityMask devIsGoodHel( ncomb );
     // ... 0d1. Compute good helicity mask on the device
-    computeDependentCouplings<<<m_gpublocks, m_gputhreads>>>( m_gs.data(), m_couplings.data() );
+    gpuLaunchKernel( computeDependentCouplings, m_gpublocks, m_gputhreads, m_gs.data(), m_couplings.data() );
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-    sigmaKin_getGoodHel<<<m_gpublocks, m_gputhreads>>>( m_momenta.data(), m_couplings.data(), m_matrixElements.data(), m_numerators.data(), m_denominators.data(), devIsGoodHel.data() );
+    gpuLaunchKernel( sigmaKin_getGoodHel, m_gpublocks, m_gputhreads, m_momenta.data(), m_couplings.data(), m_matrixElements.data(), m_numerators.data(), m_denominators.data(), devIsGoodHel.data() );
 #else
-    sigmaKin_getGoodHel<<<m_gpublocks, m_gputhreads>>>( m_momenta.data(), m_couplings.data(), m_matrixElements.data(), devIsGoodHel.data() );
+    gpuLaunchKernel( sigmaKin_getGoodHel, m_gpublocks, m_gputhreads, m_momenta.data(), m_couplings.data(), m_matrixElements.data(), devIsGoodHel.data() );
 #endif
-    checkCuda( cudaPeekAtLastError() );
+    checkGpu( gpuPeekAtLastError() );
     // ... 0d2. Copy back good helicity mask to the host
     copyHostFromDevice( hstIsGoodHel, devIsGoodHel );
     // ... 0d3. Copy back good helicity list to constant memory on the device
@@ -226,19 +226,19 @@ namespace mg5amcGpu
 
   void MatrixElementKernelDevice::computeMatrixElements( const unsigned int channelId )
   {
-    computeDependentCouplings<<<m_gpublocks, m_gputhreads>>>( m_gs.data(), m_couplings.data() );
+    gpuLaunchKernel( computeDependentCouplings, m_gpublocks, m_gputhreads, m_gs.data(), m_couplings.data() );
 #ifndef MGONGPU_NSIGHT_DEBUG
     constexpr unsigned int sharedMemSize = 0;
 #else
     constexpr unsigned int sharedMemSize = ntpbMAX * sizeof( float );
 #endif
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-    sigmaKin<<<m_gpublocks, m_gputhreads, sharedMemSize>>>( m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), channelId, m_numerators.data(), m_denominators.data(), m_selhel.data(), m_selcol.data() );
+    gpuLaunchKernelSharedMem( sigmaKin, m_gpublocks, m_gputhreads, sharedMemSize, m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), channelId, m_numerators.data(), m_denominators.data(), m_selhel.data(), m_selcol.data() );
 #else
-    sigmaKin<<<m_gpublocks, m_gputhreads, sharedMemSize>>>( m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), m_selhel.data(), m_selcol.data() );
+    gpuLaunchKernelSharedMem( sigmaKin, m_gpublocks, m_gputhreads, sharedMemSize, m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), m_selhel.data(), m_selcol.data() );
 #endif
-    checkCuda( cudaPeekAtLastError() );
-    checkCuda( cudaDeviceSynchronize() );
+    checkGpu( gpuPeekAtLastError() );
+    checkGpu( gpuDeviceSynchronize() );
   }
 
   //--------------------------------------------------------------------------
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MatrixElementKernels.h b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MatrixElementKernels.h
index 23e84757a2..72bd8f195b 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MatrixElementKernels.h
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MatrixElementKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MATRIXELEMENTKERNELS_H
 #define MATRIXELEMENTKERNELS_H 1
@@ -10,7 +10,7 @@
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -81,7 +81,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating matrix element calculations on a CPU host
   class MatrixElementKernelHost final : public MatrixElementKernelBase, public NumberOfEvents
   {
@@ -130,7 +130,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating matrix element calculations on a GPU device
   class MatrixElementKernelDevice : public MatrixElementKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessAmplitudes.h b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessAmplitudes.h
index 573b3bbbc9..ffb76e93de 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessAmplitudes.h
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessAmplitudes.h
@@ -15,7 +15,7 @@
 #define MGONGPU_TRIVIAL_AMPLITUDES 1
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessCouplings.h b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessCouplings.h
index 35a3af42e0..3afdf3e554 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessCouplings.h
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessCouplings.h
@@ -15,7 +15,7 @@
 #include "MemoryBuffers.h"       // for HostBufferCouplings::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessCouplingsFixed.h b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessCouplingsFixed.h
index dc0d93afff..ffcdf4dbef 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessCouplingsFixed.h
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessCouplingsFixed.h
@@ -14,7 +14,7 @@
 //#include "MemoryAccessHelpers.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessDenominators.h b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessDenominators.h
index 3bce635718..66f2d32a6b 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessDenominators.h
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessDenominators.h
@@ -10,7 +10,7 @@
 #include "MemoryAccessGs.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessGs.h b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessGs.h
index 31311aa375..4c726b30f3 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessGs.h
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessGs.h
@@ -13,7 +13,7 @@
 #include "MemoryBuffers.h" // for HostBufferMatrixElements::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessHelpers.h b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessHelpers.h
index c82a6c7635..db73e4e064 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessHelpers.h
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessHelpers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessHelpers_H
 #define MemoryAccessHelpers_H 1
@@ -105,7 +105,7 @@ class KernelAccessHelper : public MemoryAccessHelper<T>
     }
     else
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
       //printf( "kernelAccessRecord: ievt=%d threadId=%d\n", ievt, threadIdx.x );
       return T::ieventAccessRecord( buffer, ievt ); // NB fptype and fptype_sv coincide for CUDA
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessMatrixElements.h b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessMatrixElements.h
index f32e6fea5b..3741011971 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessMatrixElements.h
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessMatrixElements.h
@@ -13,7 +13,7 @@
 #include "MemoryBuffers.h" // for HostBufferMatrixElements::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessMomenta.h b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessMomenta.h
index 29266de32c..3be229d392 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessMomenta.h
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessMomenta.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessMomenta_H
 #define MemoryAccessMomenta_H 1
@@ -13,7 +13,7 @@
 #include "MemoryAccessVectors.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -30,7 +30,7 @@ namespace mg5amcCpu
 
     // Number of Events Per Page in the momenta AOSOA memory buffer layout
     // (these are all best kept as a compile-time constants: see issue #23)
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
     // -----------------------------------------------------------------------------------------------
     // --- GPUs: neppM is best set to a power of 2 times the number of fptype's in a 32-byte cacheline
     // --- This is relevant to ensure coalesced access to momenta in global memory
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessNumerators.h b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessNumerators.h
index b152183b28..18991f4fa6 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessNumerators.h
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessNumerators.h
@@ -10,7 +10,7 @@
 #include "MemoryAccessGs.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessRandomNumbers.h b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessRandomNumbers.h
index e2988d39f3..40cb089135 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessRandomNumbers.h
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessRandomNumbers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessRandomNumbers_H
 #define MemoryAccessRandomNumbers_H 1
@@ -11,7 +11,7 @@
 #include "CPPProcess.h"
 #include "MemoryAccessHelpers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using mg5amcGpu::CPPProcess;
 #else
 using mg5amcCpu::CPPProcess;
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessVectors.h b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessVectors.h
index e9b197368e..08faccff0f 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessVectors.h
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessVectors.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessVectors_H
 #define MemoryAccessVectors_H 1
@@ -10,7 +10,7 @@
 
 #include "mgOnGpuVectors.h"
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu // this is only needed for CPU SIMD vectorization
 {
 
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessWavefunctions.h b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessWavefunctions.h
index 5428aaf933..33bef4559e 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessWavefunctions.h
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryAccessWavefunctions.h
@@ -15,7 +15,7 @@
 #define MGONGPU_TRIVIAL_WAVEFUNCTIONS 1
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryBuffers.h b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryBuffers.h
index 3093e6ed18..7756a71621 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryBuffers.h
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/MemoryBuffers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021, based on earlier work by S. Hageboeck) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryBuffers_H
 #define MemoryBuffers_H 1
@@ -11,12 +11,12 @@
 #include "mgOnGpuCxtypes.h"
 
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "Parameters_sm.h"
 
 #include <sstream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -87,7 +87,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   constexpr bool HostBufferALIGNED = false;   // ismisaligned=false
   constexpr bool HostBufferMISALIGNED = true; // ismisaligned=true
 
@@ -119,7 +119,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA pinned host buffer
   template<typename T>
   class PinnedHostBufferBase : public BufferBase<T>
@@ -128,18 +128,18 @@ namespace mg5amcCpu
     PinnedHostBufferBase( const size_t size )
       : BufferBase<T>( size, false )
     {
-      checkCuda( cudaMallocHost( &( this->m_data ), this->bytes() ) );
+      gpuMallocHost( &( this->m_data ), this->bytes() );
     }
     virtual ~PinnedHostBufferBase()
     {
-      checkCuda( cudaFreeHost( this->m_data ) );
+      gpuFreeHost( this->m_data );
     }
   };
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA device buffer
   template<typename T>
   class DeviceBufferBase : public BufferBase<T>
@@ -148,18 +148,18 @@ namespace mg5amcCpu
     DeviceBufferBase( const size_t size )
       : BufferBase<T>( size, true )
     {
-      checkCuda( cudaMalloc( &( this->m_data ), this->bytes() ) );
+      gpuMalloc( &( this->m_data ), this->bytes() );
     }
     virtual ~DeviceBufferBase()
     {
-      checkCuda( cudaFree( this->m_data ) );
+      gpuFree( this->m_data );
     }
   };
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for a given number of events
   template<typename T, size_t sizePerEvent, bool ismisaligned>
   class HostBuffer : public HostBufferBase<T, ismisaligned>, virtual private NumberOfEvents
@@ -175,7 +175,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA pinned host buffer for a given number of events
   template<typename T, size_t sizePerEvent>
   class PinnedHostBuffer : public PinnedHostBufferBase<T>, virtual private NumberOfEvents
@@ -191,7 +191,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA device buffer for a given number of events
   template<typename T, size_t sizePerEvent>
   class DeviceBuffer : public DeviceBufferBase<T>, virtual private NumberOfEvents
@@ -213,7 +213,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for momenta random numbers
   constexpr size_t sizePerEventRndNumMomenta = MemoryBuffers::np4 * MemoryBuffers::nparf;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for momenta random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumMomenta, HostBufferALIGNED> HostBufferRndNumMomenta;
 #else
@@ -232,7 +232,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer with ONE fptype per event
   constexpr size_t sizePerEventOneFp = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer with ONE fptype per event
   typedef HostBuffer<fptype, sizePerEventOneFp, HostBufferALIGNED> HostBufferOneFp;
 #else
@@ -257,7 +257,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for Gs
   constexpr size_t sizePerEventGs = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventGs, HostBufferALIGNED> HostBufferGs;
 #else
@@ -276,7 +276,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for numerators
   constexpr size_t sizePerEventNumerators = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventNumerators, HostBufferALIGNED> HostBufferNumerators;
 #else
@@ -296,7 +296,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for denominators
   constexpr size_t sizePerEventDenominators = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventDenominators, HostBufferALIGNED> HostBufferDenominators;
 #else
@@ -315,7 +315,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for random numbers
   constexpr size_t sizePerEventCouplings = MemoryBuffers::ndcoup * MemoryBuffers::nx2;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventCouplings, HostBufferALIGNED> HostBufferCouplings;
 #else
@@ -333,7 +333,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for momenta
   constexpr size_t sizePerEventMomenta = MemoryBuffers::np4 * MemoryBuffers::npar;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for momenta
   typedef HostBuffer<fptype, sizePerEventMomenta, HostBufferALIGNED> HostBufferMomenta;
   //typedef HostBuffer<fptype, sizePerEventMomenta, HostBufferMISALIGNED> HostBufferMomenta; // TEST MISALIGNMENT!
@@ -352,7 +352,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for sampling weights
   constexpr size_t sizePerEventWeights = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for sampling weights
   typedef HostBuffer<fptype, sizePerEventWeights, HostBufferALIGNED> HostBufferWeights;
 #else
@@ -370,7 +370,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for matrix elements
   constexpr size_t sizePerEventMatrixElements = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for matrix elements
   typedef HostBuffer<fptype, sizePerEventMatrixElements, HostBufferALIGNED> HostBufferMatrixElements;
 #else
@@ -385,7 +385,7 @@ namespace mg5amcCpu
   // A base class encapsulating a memory buffer for the helicity mask
   typedef BufferBase<bool> BufferHelicityMask;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for the helicity mask
   typedef HostBufferBase<bool, HostBufferALIGNED> HostBufferHelicityMask;
 #else
@@ -403,7 +403,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for wavefunctions
   constexpr size_t sizePerEventWavefunctions = MemoryBuffers::nw6 * MemoryBuffers::nx2;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for wavefunctions
   typedef HostBuffer<fptype, sizePerEventWavefunctions, HostBufferALIGNED> HostBufferWavefunctions;
 #else
@@ -421,7 +421,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for helicity random numbers
   constexpr size_t sizePerEventRndNumHelicity = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for helicity random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumHelicity, HostBufferALIGNED> HostBufferRndNumHelicity;
 #else
@@ -439,7 +439,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for color random numbers
   constexpr size_t sizePerEventRndNumColor = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for color random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumColor, HostBufferALIGNED> HostBufferRndNumColor;
 #else
@@ -457,7 +457,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for helicity selection
   constexpr size_t sizePerEventSelectedHelicity = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for helicity selection
   typedef HostBuffer<int, sizePerEventSelectedHelicity, HostBufferALIGNED> HostBufferSelectedHelicity;
 #else
@@ -475,7 +475,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for color selection
   constexpr size_t sizePerEventSelectedColor = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for color selection
   typedef HostBuffer<int, sizePerEventSelectedColor, HostBufferALIGNED> HostBufferSelectedColor;
 #else
@@ -487,7 +487,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<class Tdst, class Tsrc>
   void copyDeviceFromHost( Tdst& dst, const Tsrc& src ) // keep the same order of arguments as in memcpy
   {
@@ -504,13 +504,13 @@ namespace mg5amcCpu
       throw std::runtime_error( sstr.str() );
     }
     // NB (PR #45): cudaMemcpy involves an intermediate memcpy to pinned memory if host array is a not a pinned host array
-    checkCuda( cudaMemcpy( dst.data(), src.data(), src.bytes(), cudaMemcpyHostToDevice ) );
+    gpuMemcpy( dst.data(), src.data(), src.bytes(), gpuMemcpyHostToDevice );
   }
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<class Tdst, class Tsrc>
   void copyHostFromDevice( Tdst& dst, const Tsrc& src ) // keep the same order of arguments as in memcpy
   {
@@ -527,7 +527,7 @@ namespace mg5amcCpu
       throw std::runtime_error( sstr.str() );
     }
     // NB (PR #45): cudaMemcpy involves an intermediate memcpy to pinned memory if host array is a not a pinned host array
-    checkCuda( cudaMemcpy( dst.data(), src.data(), src.bytes(), cudaMemcpyDeviceToHost ) );
+    gpuMemcpy( dst.data(), src.data(), src.bytes(), gpuMemcpyDeviceToHost );
   }
 #endif
 
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/P1_Sigma_sm_gg_ttxgg/CPPProcess.cc b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/P1_Sigma_sm_gg_ttxgg/CPPProcess.cc
index f9016eaa88..f2a85b9b75 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/P1_Sigma_sm_gg_ttxgg/CPPProcess.cc
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/P1_Sigma_sm_gg_ttxgg/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_sm.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -46,7 +45,7 @@
 // Class member functions for calculating the matrix elements for
 // Process: g g > t t~ g g WEIGHTED<=4 @1
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -80,7 +79,7 @@ namespace mg5amcCpu
   __device__ const fptype cIPD[2] = { (fptype)Parameters_sm::mdl_MT, (fptype)Parameters_sm::mdl_WT };
   __device__ const fptype* cIPC = nullptr; // unused as nicoup=0
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ fptype cIPD[2];
   __device__ __constant__ fptype* cIPC = nullptr; // unused as nicoup=0
 #else
@@ -90,7 +89,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -118,13 +117,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -151,7 +150,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -187,7 +186,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -200,8 +199,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -2474,7 +2475,7 @@ namespace mg5amcCpu
         { 62, 71, -10, 80, -1, 8, -28, 62, 62, -10, -10, -1, -1, 8, -10, -1, -64, 8, 8, -64, 80, 8, 512, -64 },
         { -28, 62, 62, -10, -10, -1, 62, 71, -10, 80, -1, 8, -10, -1, -1, 8, 8, -64, 80, 8, 8, -64, -64, 512 } }; // 2-D array[24][24]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -2531,7 +2532,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -2590,7 +2591,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -2685,8 +2686,8 @@ namespace mg5amcCpu
       { 1, 1, 1, -1, -1, 1 },
       { 1, 1, 1, -1, 1, -1 },
       { 1, 1, 1, -1, 1, 1 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -2728,9 +2729,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     const fptype tIPD[2] = { (fptype)m_pars->mdl_MT, (fptype)m_pars->mdl_WT };
     //const cxtype tIPC[0] = { ... }; // nicoup=0
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) ) );
-    //checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ) ); // nicoup=0
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) );
+    //gpuMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
 #else
     memcpy( cIPD, tIPD, 2 * sizeof( fptype ) );
     //memcpy( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
@@ -2768,7 +2769,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -2797,6 +2798,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -2810,6 +2815,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -2833,12 +2840,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -2859,7 +2866,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -2985,9 +2992,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -3011,7 +3018,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -3031,7 +3038,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 512 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -3045,9 +3052,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -3075,7 +3085,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -3285,7 +3295,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/P1_Sigma_sm_gg_ttxgg/CPPProcess.h b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/P1_Sigma_sm_gg_ttxgg/CPPProcess.h
index 04f7c62976..deb1358992 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/P1_Sigma_sm_gg_ttxgg/CPPProcess.h
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/P1_Sigma_sm_gg_ttxgg/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -107,7 +107,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -120,7 +120,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -150,7 +150,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/P1_Sigma_sm_gg_ttxgg/CudaRuntime.h b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/P1_Sigma_sm_gg_ttxgg/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/P1_Sigma_sm_gg_ttxgg/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/P1_Sigma_sm_gg_ttxgg/GpuAbstraction.h b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/P1_Sigma_sm_gg_ttxgg/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/P1_Sigma_sm_gg_ttxgg/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/P1_Sigma_sm_gg_ttxgg/GpuRuntime.h b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/P1_Sigma_sm_gg_ttxgg/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/P1_Sigma_sm_gg_ttxgg/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/P1_Sigma_sm_gg_ttxgg/check_sa.cc b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/P1_Sigma_sm_gg_ttxgg/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/P1_Sigma_sm_gg_ttxgg/check_sa.cc
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/P1_Sigma_sm_gg_ttxgg/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/RamboSamplingKernels.cc b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/RamboSamplingKernels.cc
index da68aa9255..79abbcc4f8 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/RamboSamplingKernels.cc
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/RamboSamplingKernels.cc
@@ -1,11 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "RamboSamplingKernels.h"
 
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "MemoryAccessMomenta.h"
 #include "MemoryAccessRandomNumbers.h"
 #include "MemoryAccessWeights.h"
@@ -14,7 +14,7 @@
 
 #include <sstream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -92,7 +92,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingKernelDevice::RamboSamplingKernelDevice( const fptype energy,               // input: energy
                                                         const BufferRndNumMomenta& rndmom, // input: random numbers in [0,1]
                                                         BufferMomenta& momenta,            // output: momenta
@@ -135,7 +135,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   getMomentaInitialDevice( const fptype energy,
                            fptype* momenta )
@@ -147,17 +147,17 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   void
   RamboSamplingKernelDevice::getMomentaInitial()
   {
-    getMomentaInitialDevice<<<m_gpublocks, m_gputhreads>>>( m_energy, m_momenta.data() );
+    gpuLaunchKernel( getMomentaInitialDevice, m_gpublocks, m_gputhreads, m_energy, m_momenta.data() );
   }
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   getMomentaFinalDevice( const fptype energy,
                          const fptype* rndmom,
@@ -171,11 +171,11 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   void
   RamboSamplingKernelDevice::getMomentaFinal()
   {
-    getMomentaFinalDevice<<<m_gpublocks, m_gputhreads>>>( m_energy, m_rndmom.data(), m_momenta.data(), m_weights.data() );
+    gpuLaunchKernel( getMomentaFinalDevice, m_gpublocks, m_gputhreads, m_energy, m_rndmom.data(), m_momenta.data(), m_weights.data() );
   }
 #endif
 
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/RamboSamplingKernels.h b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/RamboSamplingKernels.h
index 184089efd7..7c214cd74b 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/RamboSamplingKernels.h
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/RamboSamplingKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef RAMBOSAMPLINGKERNELS_H
 #define RAMBOSAMPLINGKERNELS_H 1
@@ -10,7 +10,7 @@
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -93,7 +93,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating RAMBO phase space sampling on a GPU device
   class RamboSamplingKernelDevice final : public SamplingKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/RandomNumberKernels.h b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/RandomNumberKernels.h
index 188a72c2c9..21d63beeac 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/RandomNumberKernels.h
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/RandomNumberKernels.h
@@ -1,14 +1,14 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef RANDOMNUMBERKERNELS_H
 #define RANDOMNUMBERKERNELS_H 1
 
 #include "mgOnGpuConfig.h"
 
-// NB This must come AFTER mgOnGpuConfig.h which contains our definition of __global__ when __CUDACC__ is not defined
+// NB This must come AFTER mgOnGpuConfig.h which contains our definition of __global__ when MGONGPUCPP_GPUIMPL is not defined
 #ifndef MGONGPU_HAS_NO_CURAND
 //#include "curand.h"
 struct curandGenerator_st; // forward definition from curand.h
@@ -16,7 +16,7 @@ struct curandGenerator_st; // forward definition from curand.h
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/cudacpp.mk b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/cudacpp.mk
index 509307506b..df74dfc284 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/cudacpp.mk
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/cudacpp.mk
@@ -1,7 +1,7 @@
 # Copyright (C) 2020-2023 CERN and UCLouvain.
 # Licensed under the GNU Lesser General Public License (version 3 or later).
 # Created by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-# Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+# Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #=== Determine the name of this makefile (https://ftp.gnu.org/old-gnu/Manuals/make-3.80/html_node/make_17.html)
 #=== NB: use ':=' to ensure that the value of CUDACPP_MAKEFILE is not modified further down after including make_opts
@@ -42,10 +42,10 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure common compiler flags for C++ and CUDA
+#=== Configure common compiler flags for C++ and CUDA/HIP
 
 INCFLAGS = -I.
-OPTFLAGS = -O3 # this ends up in CUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
+OPTFLAGS = -O3 # this ends up in GPUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
 
 # Dependency on src directory
 MG5AMC_COMMONLIB = mg5amc_common
@@ -121,24 +121,45 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure the CUDA compiler
+#=== Configure the GPU compiler (CUDA or HIP)
+
+# FIXME! (AV 24.01.2024)
+# In the current implementation (without separate builds for C++ and CUDA/HIP), we first check for cudacc and hipcc in CUDA_HOME and HIP_HOME.
+# If CUDA_HOME or HIP_HOME are not set, try to determine them from the path to cudacc and hipcc.
+# While convoluted, this is currently necessary to allow disabling CUDA/HIP builds by setting CUDA_HOME or HIP_HOME to invalid paths.
+# This will (probably?) be fixed when separate C++ and CUDA/HIP builds are implemented (PR #775).
 
 # If CXX is not a single word (example "clang++ --gcc-toolchain...") then disable CUDA builds (issue #505)
-# This is because it is impossible to pass this to "CUFLAGS += -ccbin <host-compiler>" below
+# This is because it is impossible to pass this to "GPUFLAGS += -ccbin <host-compiler>" below
 ifneq ($(words $(subst ccache ,,$(CXX))),1) # allow at most "CXX=ccache <host-compiler>" from outside
   $(warning CUDA builds are not supported for multi-word CXX "$(CXX)")
   override CUDA_HOME=disabled
 endif
 
-# If CUDA_HOME is not set, try to set it from the location of nvcc
+# If CUDA_HOME is not set, try to set it from the path to nvcc
 ifndef CUDA_HOME
-  CUDA_HOME = $(patsubst %bin/nvcc,%,$(shell which nvcc 2>/dev/null))
+  CUDA_HOME = $(patsubst %/bin/nvcc,%,$(shell which nvcc 2>/dev/null))
   $(warning CUDA_HOME was not set: using "$(CUDA_HOME)")
 endif
 
-# Set NVCC as $(CUDA_HOME)/bin/nvcc if it exists
+# If HIP_HOME is not set, try to set it from the path to hipcc
+ifndef HIP_HOME
+  HIP_HOME = $(patsubst %/bin/hipcc,%,$(shell which hipcc 2>/dev/null))
+  $(warning HIP_HOME was not set: using "$(HIP_HOME)")
+endif
+
+# FIXME! (AV 24.01.2024)
+# In the current implementation (without separate builds for C++ and CUDA/HIP),
+# builds are performed for HIP only if CUDA is not found in the path.
+# If both CUDA and HIP are installed, HIP builds can be triggered by unsetting CUDA_HOME.
+# This will be fixed when separate C++ and CUDA/HIP builds are implemented (PR #775).
+
+#--- Option 1: CUDA exists -> use CUDA
+
+# Set GPUCC as $(CUDA_HOME)/bin/nvcc if it exists
 ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
-  NVCC = $(CUDA_HOME)/bin/nvcc
+
+  GPUCC = $(CUDA_HOME)/bin/nvcc
   USE_NVTX ?=-DUSE_NVTX
   # See https://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/index.html
   # See https://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/
@@ -148,7 +169,7 @@ ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
   # Examples: use 60 for P100 (Piz Daint), 80 for A100 (Juwels Booster, NVidia raplab/Curiosity).
   MADGRAPH_CUDA_ARCHITECTURE ?= 70
   ###CUARCHFLAGS = -gencode arch=compute_$(MADGRAPH_CUDA_ARCHITECTURE),code=compute_$(MADGRAPH_CUDA_ARCHITECTURE) -gencode arch=compute_$(MADGRAPH_CUDA_ARCHITECTURE),code=sm_$(MADGRAPH_CUDA_ARCHITECTURE) # Older implementation (AV): go back to this one for multi-GPU support #533
-  ###CUARCHFLAGS = --gpu-architecture=compute_$(MADGRAPH_CUDA_ARCHITECTURE) --gpu-code=sm_$(MADGRAPH_CUDA_ARCHITECTURE),compute_$(MADGRAPH_CUDA_ARCHITECTURE) # Newer implementation (SH): cannot use this as-is for multi-GPU support #533
+  ###CUARCHFLAGS = --gpu-architecture=compute_$(MADGRAPH_CUDA_ARCHITECTURE) --gpu-code=sm_$(MADGRAPH_CUDA_ARCHITECTURE),compute_$(MADGRAPH_CUDA_ARCHITECTURE)  # Newer implementation (SH): cannot use this as-is for multi-GPU support #533
   comma:=,
   CUARCHFLAGS = $(foreach arch,$(subst $(comma), ,$(MADGRAPH_CUDA_ARCHITECTURE)),-gencode arch=compute_$(arch),code=compute_$(arch) -gencode arch=compute_$(arch),code=sm_$(arch))
   CUINC = -I$(CUDA_HOME)/include/
@@ -158,41 +179,79 @@ ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
     CURANDLIBFLAGS = -L$(CUDA_HOME)/lib64/ -lcurand # NB: -lcuda is not needed here!
   endif
   CUOPTFLAGS = -lineinfo
-  CUFLAGS = $(foreach opt, $(OPTFLAGS), -Xcompiler $(opt)) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
-  ###CUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
-  ###NVCC_VERSION = $(shell $(NVCC) --version | grep 'Cuda compilation tools' | cut -d' ' -f5 | cut -d, -f1)
-  CUFLAGS += -std=c++17 # need CUDA >= 11.2 (see #333): this is enforced in mgOnGpuConfig.h
+  ###GPUFLAGS = $(OPTFLAGS) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
+  GPUFLAGS = $(foreach opt, $(OPTFLAGS), -Xcompiler $(opt)) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
+  ###GPUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
+  ###GPUCC_VERSION = $(shell $(GPUCC) --version | grep 'Cuda compilation tools' | cut -d' ' -f5 | cut -d, -f1)
+  GPUFLAGS += -std=c++17 # need CUDA >= 11.2 (see #333): this is enforced in mgOnGpuConfig.h
   # Without -maxrregcount: baseline throughput: 6.5E8 (16384 32 12) up to 7.3E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 160 # improves throughput: 6.9E8 (16384 32 12) up to 7.7E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 128 # improves throughput: 7.3E8 (16384 32 12) up to 7.6E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 96 # degrades throughput: 4.1E8 (16384 32 12) up to 4.5E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 64 # degrades throughput: 1.7E8 (16384 32 12) flat at 1.7E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 160 # improves throughput: 6.9E8 (16384 32 12) up to 7.7E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 128 # improves throughput: 7.3E8 (16384 32 12) up to 7.6E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 96 # degrades throughput: 4.1E8 (16384 32 12) up to 4.5E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 64 # degrades throughput: 1.7E8 (16384 32 12) flat at 1.7E8 (65536 128 12)
+  CUBUILDRULEFLAGS = -Xcompiler -fPIC -c
+  CCBUILDRULEFLAGS = -Xcompiler -fPIC -c -x cu
+  CUDATESTFLAGS = -lcuda
+
+  # Set the host C++ compiler for GPUCC via "-ccbin <host-compiler>"
+  # (NB issue #505: this must be a single word, "clang++ --gcc-toolchain..." is not supported)
+  GPUFLAGS += -ccbin $(shell which $(subst ccache ,,$(CXX)))
+
+  # Allow newer (unsupported) C++ compilers with older versions of CUDA if ALLOW_UNSUPPORTED_COMPILER_IN_CUDA is set (#504)
+  ifneq ($(origin ALLOW_UNSUPPORTED_COMPILER_IN_CUDA),undefined)
+  GPUFLAGS += -allow-unsupported-compiler
+  endif
+
 else ifneq ($(origin REQUIRE_CUDA),undefined)
+
   # If REQUIRE_CUDA is set but no cuda is found, stop here (e.g. for CI tests on GPU #443)
-  $(error No cuda installation found (set CUDA_HOME or make nvcc visible in PATH))
+  $(error No cuda installation found (set CUDA_HOME or make GPUCC visible in PATH))
+
+#--- Option 2: CUDA does not exist, HIP exists -> use HIP
+
+# Set GPUCC as $(HIP_HOME)/bin/hipcc if it exists
+else ifneq ($(wildcard $(HIP_HOME)/bin/hipcc),)
+
+  GPUCC = $(HIP_HOME)/bin/hipcc
+  #USE_NVTX ?=-DUSE_NVTX # should maybe find something equivalent to this in HIP?
+  HIPARCHFLAGS = -target x86_64-linux-gnu --offload-arch=gfx90a
+  HIPINC = -I$(HIP_HOME)/include/
+  # Note: -DHIP_FAST_MATH is equivalent to -use_fast_math in HIP 
+  # (but only for single precision line 208: https://rocm-developer-tools.github.io/HIP/hcc__detail_2math__functions_8h_source.html)
+  # Note: CUOPTFLAGS should not be used for HIP, it had been added here but was then removed (#808)
+  GPUFLAGS = $(OPTFLAGS) $(INCFLAGS) $(HIPINC) $(HIPARCHFLAGS) -DHIP_FAST_MATH -DHIP_PLATFORM=amd -fPIC
+  ###GPUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
+  GPUFLAGS += -std=c++17
+  ###GPUFLAGS+= --maxrregcount 255 # (AV: is this option valid on HIP and meaningful on AMD GPUs?)
+  CUBUILDRULEFLAGS = -fPIC -c
+  CCBUILDRULEFLAGS = -fPIC -c -x hip
+
+else ifneq ($(origin REQUIRE_HIP),undefined)
+
+  # If REQUIRE_HIP is set but no HIP is found, stop here (e.g. for CI tests on GPU #443)
+  $(error No hip installation found (set HIP_HOME or make GPUCC visible in PATH))
+
+#--- Option 3: CUDA does not exist, HIP does not exist -> switch off both CUDA and HIP
+
 else
-  # No cuda. Switch cuda compilation off and go to common random numbers in C++
+
+  # No cudacc and no hipcc: switch CUDA and HIP compilation off and go to common random numbers in C++
   $(warning CUDA_HOME is not set or is invalid: export CUDA_HOME to compile with cuda)
-  override NVCC=
+  $(warning HIP_HOME is not set or is invalid: export HIP_HOME to compile with hip)
+  override GPUCC=
   override USE_NVTX=
   override CUINC=
   override CURANDLIBFLAGS=
-endif
-export NVCC
-export CUFLAGS
 
-# Set the host C++ compiler for nvcc via "-ccbin <host-compiler>"
-# (NB issue #505: this must be a single word, "clang++ --gcc-toolchain..." is not supported)
-CUFLAGS += -ccbin $(shell which $(subst ccache ,,$(CXX)))
-
-# Allow newer (unsupported) C++ compilers with older versions of CUDA if ALLOW_UNSUPPORTED_COMPILER_IN_CUDA is set (#504)
-ifneq ($(origin ALLOW_UNSUPPORTED_COMPILER_IN_CUDA),undefined)
-CUFLAGS += -allow-unsupported-compiler
 endif
 
+# Export GPUCC (so that it can also be used in cudacpp_src.mk?)
+export GPUCC
+export GPUFLAGS
+
 #-------------------------------------------------------------------------------
 
-#=== Configure ccache for C++ and CUDA builds
+#=== Configure ccache for C++ and CUDA/HIP builds
 
 # Enable ccache if USECCACHE=1
 ifeq ($(USECCACHE)$(shell echo $(CXX) | grep ccache),1)
@@ -201,15 +260,15 @@ endif
 #ifeq ($(USECCACHE)$(shell echo $(AR) | grep ccache),1)
 #  override AR:=ccache $(AR)
 #endif
-ifneq ($(NVCC),)
-  ifeq ($(USECCACHE)$(shell echo $(NVCC) | grep ccache),1)
-    override NVCC:=ccache $(NVCC)
+ifneq ($(GPUCC),)
+  ifeq ($(USECCACHE)$(shell echo $(GPUCC) | grep ccache),1)
+    override GPUCC:=ccache $(GPUCC)
   endif
 endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure PowerPC-specific compiler flags for C++ and CUDA
+#=== Configure PowerPC-specific compiler flags for C++ and CUDA/HIP
 
 # PowerPC-specific CXX compiler flags (being reviewed)
 ifeq ($(UNAME_P),ppc64le)
@@ -225,9 +284,9 @@ else
   ######CXXFLAGS+= -fno-semantic-interposition # no benefit (neither alone, nor combined with -flto)
 endif
 
-# PowerPC-specific CUDA compiler flags (to be reviewed!)
+# PowerPC-specific CUDA/HIP compiler flags (to be reviewed!)
 ifeq ($(UNAME_P),ppc64le)
-  CUFLAGS+= -Xcompiler -mno-float128
+  GPUFLAGS+= -Xcompiler -mno-float128
 endif
 
 #-------------------------------------------------------------------------------
@@ -235,9 +294,11 @@ endif
 #=== Configure defaults and check if user-defined choices exist for OMPFLAGS, AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
 
 # Set the default OMPFLAGS choice
-ifneq ($(shell $(CXX) --version | egrep '^Intel'),)
+ifneq ($(findstring hipcc,$(GPUCC)),)
+override OMPFLAGS = # disable OpenMP MT when using hipcc #802
+else ifneq ($(shell $(CXX) --version | egrep '^Intel'),)
 override OMPFLAGS = -fopenmp
-###override OMPFLAGS = # disable OpenMP MT on Intel (was ok without nvcc but not ok with nvcc before #578)
+###override OMPFLAGS = # disable OpenMP MT on Intel (was ok without GPUCC but not ok with GPUCC before #578)
 else ifneq ($(shell $(CXX) --version | egrep '^(clang)'),)
 override OMPFLAGS = -fopenmp
 ###override OMPFLAGS = # disable OpenMP MT on clang (was not ok without or with nvcc before #578)
@@ -293,7 +354,10 @@ endif
 
 # Set the default RNDGEN (random number generator) choice
 ifeq ($(RNDGEN),)
-  ifeq ($(NVCC),)
+  ifeq ($(GPUCC),)
+    override RNDGEN = hasNoCurand
+  # Edgecase for HIP compilation
+  else ifeq ($(findstring hipcc,$(GPUCC)),hipcc)
     override RNDGEN = hasNoCurand
   else ifeq ($(RNDGEN),)
     override RNDGEN = hasCurand
@@ -310,7 +374,7 @@ export OMPFLAGS
 
 #-------------------------------------------------------------------------------
 
-#=== Set the CUDA/C++ compiler flags appropriate to user-defined choices of AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
+#=== Set the CUDA/HIP/C++ compiler flags appropriate to user-defined choices of AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
 
 # Set the build flags appropriate to OMPFLAGS
 $(info OMPFLAGS=$(OMPFLAGS))
@@ -368,13 +432,13 @@ CXXFLAGS+= $(AVXFLAGS)
 $(info FPTYPE=$(FPTYPE))
 ifeq ($(FPTYPE),d)
   CXXFLAGS += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
-  CUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
+  GPUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
 else ifeq ($(FPTYPE),f)
   CXXFLAGS += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
-  CUFLAGS  += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
+  GPUFLAGS  += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
 else ifeq ($(FPTYPE),m)
   CXXFLAGS += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
-  CUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
+  GPUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
 else
   $(error Unknown FPTYPE='$(FPTYPE)': only 'd', 'f' and 'm' are supported)
 endif
@@ -383,7 +447,7 @@ endif
 $(info HELINL=$(HELINL))
 ifeq ($(HELINL),1)
   CXXFLAGS += -DMGONGPU_INLINE_HELAMPS
-  CUFLAGS  += -DMGONGPU_INLINE_HELAMPS
+  GPUFLAGS  += -DMGONGPU_INLINE_HELAMPS
 else ifneq ($(HELINL),0)
   $(error Unknown HELINL='$(HELINL)': only '0' and '1' are supported)
 endif
@@ -392,7 +456,7 @@ endif
 $(info HRDCOD=$(HRDCOD))
 ifeq ($(HRDCOD),1)
   CXXFLAGS += -DMGONGPU_HARDCODE_PARAM
-  CUFLAGS  += -DMGONGPU_HARDCODE_PARAM
+  GPUFLAGS  += -DMGONGPU_HARDCODE_PARAM
 else ifneq ($(HRDCOD),0)
   $(error Unknown HRDCOD='$(HRDCOD)': only '0' and '1' are supported)
 endif
@@ -444,11 +508,11 @@ ifeq ($(UNAME_S),Darwin)
   override CULIBFLAGSRPATH2 =
 else
   # RPATH to cuda/cpp libs when linking executables
-  override CXXLIBFLAGSRPATH = -Wl,-rpath,$(LIBDIRRPATH)
-  override CULIBFLAGSRPATH = -Xlinker -rpath,$(LIBDIRRPATH)
+  override CXXLIBFLAGSRPATH = -Wl,-rpath=$(LIBDIRRPATH)
+  override CULIBFLAGSRPATH = -Xlinker -rpath=$(LIBDIRRPATH)
   # RPATH to common lib when linking cuda/cpp libs
-  override CXXLIBFLAGSRPATH2 = -Wl,-rpath,'$$ORIGIN'
-  override CULIBFLAGSRPATH2 = -Xlinker -rpath,'$$ORIGIN'
+  override CXXLIBFLAGSRPATH2 = -Wl,-rpath='$$ORIGIN'
+  override CULIBFLAGSRPATH2 = -Xlinker -rpath='$$ORIGIN'
 endif
 
 # Setting LD_LIBRARY_PATH or DYLD_LIBRARY_PATH in the RUNTIME is no longer necessary (neither on Linux nor on Mac)
@@ -461,7 +525,7 @@ override RUNTIME =
 cxx_main=$(BUILDDIR)/check.exe
 fcxx_main=$(BUILDDIR)/fcheck.exe
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 cu_main=$(BUILDDIR)/gcheck.exe
 fcu_main=$(BUILDDIR)/fgcheck.exe
 else
@@ -492,14 +556,15 @@ $(BUILDDIR)/.build.$(TAG):
 	@touch $(BUILDDIR)/.build.$(TAG)
 
 # Generic target and build rules: objects from CUDA compilation
-ifneq ($(NVCC),)
+# NB: CCBUILDRULEFLAGS includes "-x cu" for nvcc and "-x hip" for hipcc (#810)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/%.o : %.cu *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $(CUBUILDRULEFLAGS) $< -o $@
 
 $(BUILDDIR)/%_cu.o : %.cc *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c -x cu $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $(CCBUILDRULEFLAGS) $< -o $@
 endif
 
 # Generic target and build rules: objects from C++ compilation
@@ -509,11 +574,14 @@ $(BUILDDIR)/%.o : %.cc *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	$(CXX) $(CPPFLAGS) $(CXXFLAGS) -fPIC -c $< -o $@
 
 # Apply special build flags only to CrossSectionKernel.cc and gCrossSectionKernel.cu (no fast math, see #117 and #516)
+# Added edgecase for HIP compilation
 ifeq ($(shell $(CXX) --version | grep ^nvc++),)
 $(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS := $(filter-out -ffast-math,$(CXXFLAGS))
 $(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS += -fno-fast-math
-ifneq ($(NVCC),)
-$(BUILDDIR)/gCrossSectionKernels.o: CUFLAGS += -Xcompiler -fno-fast-math
+ifeq ($(findstring nvcc,$(GPUCC)),nvcc)
+  $(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -Xcompiler -fno-fast-math
+else
+  $(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -fno-fast-math
 endif
 endif
 
@@ -530,10 +598,10 @@ ifeq ($(RNDGEN),hasCurand)
 $(BUILDDIR)/CurandRandomNumberKernel.o: CXXFLAGS += $(CUINC)
 endif
 
-# Avoid "warning: builtin __has_trivial_... is deprecated; use __is_trivially_... instead" in nvcc with icx2023 (#592)
+# Avoid "warning: builtin __has_trivial_... is deprecated; use __is_trivially_... instead" in GPUCC with icx2023 (#592)
 ifneq ($(shell $(CXX) --version | egrep '^(Intel)'),)
-ifneq ($(NVCC),)
-CUFLAGS += -Xcompiler -Wno-deprecated-builtins
+ifneq ($(GPUCC),)
+GPUFLAGS += -Wno-deprecated-builtins
 endif
 endif
 
@@ -541,8 +609,8 @@ endif
 # This patch does remove the warning, but I prefer to keep it disabled for the moment...
 ###ifneq ($(shell $(CXX) --version | egrep '^(clang|Apple clang|Intel)'),)
 ###$(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS += -Wno-overriding-t-option
-###ifneq ($(NVCC),)
-###$(BUILDDIR)/gCrossSectionKernels.o: CUFLAGS += -Xcompiler -Wno-overriding-t-option
+###ifneq ($(GPUCC),)
+###$(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -Xcompiler -Wno-overriding-t-option
 ###endif
 ###endif
 
@@ -569,7 +637,7 @@ MG5AMC_CXXLIB = mg5amc_$(processid_short)_cpp
 cxx_objects_lib=$(BUILDDIR)/CPPProcess.o $(BUILDDIR)/MatrixElementKernels.o $(BUILDDIR)/BridgeKernels.o $(BUILDDIR)/CrossSectionKernels.o
 cxx_objects_exe=$(BUILDDIR)/CommonRandomNumberKernel.o $(BUILDDIR)/RamboSamplingKernels.o
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 MG5AMC_CULIB = mg5amc_$(processid_short)_cuda
 cu_objects_lib=$(BUILDDIR)/gCPPProcess.o $(BUILDDIR)/gMatrixElementKernels.o $(BUILDDIR)/gBridgeKernels.o $(BUILDDIR)/gCrossSectionKernels.o
 cu_objects_exe=$(BUILDDIR)/gCommonRandomNumberKernel.o $(BUILDDIR)/gRamboSamplingKernels.o
@@ -581,11 +649,17 @@ $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: cxx_objects_lib += $(BUILDDIR)/fbridge.o
 $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib)
 	$(CXX) -shared -o $@ $(cxx_objects_lib) $(CXXLIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: $(BUILDDIR)/fbridge_cu.o
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: cu_objects_lib += $(BUILDDIR)/fbridge_cu.o
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cu_objects_lib)
-	$(NVCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+# Bypass std::filesystem completely to ease portability on LUMI #803
+#ifneq ($(findstring hipcc,$(GPUCC)),)
+#	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB) -lstdc++fs
+#else
+#	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+#endif
 endif
 
 #-------------------------------------------------------------------------------
@@ -602,16 +676,16 @@ $(cxx_main): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PAT
 $(cxx_main): $(BUILDDIR)/check_sa.o $(LIBDIR)/lib$(MG5AMC_CXXLIB).so $(cxx_objects_exe) $(BUILDDIR)/CurandRandomNumberKernel.o
 	$(CXX) -o $@ $(BUILDDIR)/check_sa.o $(OMPFLAGS) -ldl -pthread $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe) $(BUILDDIR)/CurandRandomNumberKernel.o $(CURANDLIBFLAGS)
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(cu_main): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(cu_main): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(cu_main): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(cu_main): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 else ifneq ($(shell $(CXX) --version | grep ^nvc++),) # support nvc++ #531
 $(cu_main): LIBFLAGS += -L$(patsubst %bin/nvc++,%lib,$(subst ccache ,,$(CXX))) -lnvhpcatm -lnvcpumath -lnvc
 endif
 $(cu_main): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(cu_main): $(BUILDDIR)/gcheck_sa.o $(LIBDIR)/lib$(MG5AMC_CULIB).so $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o
-	$(NVCC) -o $@ $(BUILDDIR)/gcheck_sa.o $(CUARCHFLAGS) $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o $(CURANDLIBFLAGS)
+	$(GPUCC) -o $@ $(BUILDDIR)/gcheck_sa.o $(CUARCHFLAGS) $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o $(CURANDLIBFLAGS)
 endif
 
 #-------------------------------------------------------------------------------
@@ -635,19 +709,27 @@ $(fcxx_main): LIBFLAGS += -L$(shell dirname $(shell $(FC) --print-file-name libg
 endif
 $(fcxx_main): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(fcxx_main): $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler.o $(LIBDIR)/lib$(MG5AMC_CXXLIB).so $(cxx_objects_exe)
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(BUILDDIR)/fcheck_sa.o $(OMPFLAGS) $(BUILDDIR)/fsampler.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe) -lstdc++
+else
 	$(CXX) -o $@ $(BUILDDIR)/fcheck_sa.o $(OMPFLAGS) $(BUILDDIR)/fsampler.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe)
+endif
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(fcu_main): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(fcu_main): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(fcu_main): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(fcu_main): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 endif
 ifeq ($(UNAME_S),Darwin)
 $(fcu_main): LIBFLAGS += -L$(shell dirname $(shell $(FC) --print-file-name libgfortran.dylib)) # add path to libgfortran on Mac #375
 endif
 $(fcu_main): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(fcu_main): $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBDIR)/lib$(MG5AMC_CULIB).so $(cu_objects_exe)
-	$(NVCC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe)
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) -lstdc++ -L$(shell dirname $(shell $(GPUCC) -print-prog-name=clang))/../../lib -lamdhip64
+else
+	$(GPUCC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe)
+endif
 endif
 
 #-------------------------------------------------------------------------------
@@ -659,7 +741,7 @@ $(BUILDDIR)/testxxx.o: testxxx_cc_ref.txt
 $(testmain): $(BUILDDIR)/testxxx.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/testxxx.o # Comment out this line to skip the C++ test of xxx functions
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/testxxx_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/testxxx_cu.o: INCFLAGS += $(GTESTINC)
 $(BUILDDIR)/testxxx_cu.o: testxxx_cc_ref.txt
@@ -672,7 +754,7 @@ $(BUILDDIR)/testmisc.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/testmisc.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/testmisc.o # Comment out this line to skip the C++ miscellaneous tests
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/testmisc_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/testmisc_cu.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/testmisc_cu.o
@@ -684,12 +766,12 @@ $(BUILDDIR)/runTest.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/runTest.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/runTest.o
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/runTest_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/runTest_cu.o: INCFLAGS += $(GTESTINC)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(testmain): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(testmain): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(testmain): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(testmain): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 else ifneq ($(shell $(CXX) --version | grep ^nvc++),) # support nvc++ #531
 $(testmain): LIBFLAGS += -L$(patsubst %bin/nvc++,%lib,$(subst ccache ,,$(CXX))) -lnvhpcatm -lnvcpumath -lnvc
 endif
@@ -713,14 +795,23 @@ $(testmain): LIBFLAGS += -lgomp
 endif
 endif
 
-ifeq ($(NVCC),) # link only runTest.o
+# Bypass std::filesystem completely to ease portability on LUMI #803
+#ifneq ($(findstring hipcc,$(GPUCC)),)
+#$(testmain): LIBFLAGS += -lstdc++fs
+#endif
+
+ifeq ($(GPUCC),) # link only runTest.o
 $(testmain): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(testmain): $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib) $(cxx_objects_exe) $(GTESTLIBS)
 	$(CXX) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) -ldl -pthread $(LIBFLAGS)
 else # link both runTest.o and runTest_cu.o
 $(testmain): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(testmain): $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) $(GTESTLIBS)
-	$(NVCC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) -lcuda
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) $(CUDATESTFLAGS) -lstdc++ -lpthread  -L$(shell dirname $(shell $(GPUCC) -print-prog-name=clang))/../../lib -lamdhip64
+else
+	$(GPUCC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) $(CUDATESTFLAGS)
+endif
 endif
 
 # Use target gtestlibs to build only googletest
@@ -829,9 +920,9 @@ ifeq ($(USECCACHE),1)
 	ccache --version | head -1
 endif
 	@echo ""
-	@echo NVCC=$(NVCC)
-ifneq ($(NVCC),)
-	$(NVCC) --version
+	@echo GPUCC=$(GPUCC)
+ifneq ($(GPUCC),)
+	$(GPUCC) --version
 endif
 	@echo ""
 	@echo CXX=$(CXX)
@@ -850,7 +941,7 @@ endif
 
 # Target: check (run the C++ test executable)
 # [NB THIS IS WHAT IS USED IN THE GITHUB CI!]
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 check: runTest cmpFcheck cmpFGcheck
 else
 check: runTest cmpFcheck
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/fbridge.cc b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/fbridge.cc
index 2d2b36d560..22ce3f5115 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/fbridge.cc
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/fbridge.cc
@@ -1,11 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Roiser (Oct 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "Bridge.h"
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 
 extern "C"
 {
@@ -22,7 +22,7 @@ extern "C"
    * Using the same Fortran MadEvent code, linking to the hetrerogeneous library would allow access to both CPU and GPU implementations.
    * The specific heterogeneous configuration (how many GPUs, how many threads on each CPU, etc) could be loaded in CUDA/C++ from a data file.
    */
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -46,8 +46,8 @@ extern "C"
    */
   void fbridgecreate_( CppObjectInFortran** ppbridge, const int* pnevtF, const int* pnparF, const int* pnp4F )
   {
-#ifdef __CUDACC__
-    CudaRuntime::setUp();
+#ifdef MGONGPUCPP_GPUIMPL
+    GpuRuntime::setUp();
 #endif
     // (NB: CPPProcess::initProc no longer needs to be executed here because it is called in the Bridge constructor)
     // FIXME: disable OMP in Bridge when called from Fortran
@@ -65,8 +65,8 @@ extern "C"
     Bridge<FORTRANFPTYPE>* pbridge = dynamic_cast<Bridge<FORTRANFPTYPE>*>( *ppbridge );
     if( pbridge == 0 ) throw std::runtime_error( "fbridgedelete_: invalid Bridge address" );
     delete pbridge;
-#ifdef __CUDACC__
-    CudaRuntime::tearDown();
+#ifdef MGONGPUCPP_GPUIMPL
+    GpuRuntime::tearDown();
 #endif
   }
 
@@ -96,7 +96,7 @@ extern "C"
   {
     Bridge<FORTRANFPTYPE>* pbridge = dynamic_cast<Bridge<FORTRANFPTYPE>*>( *ppbridge );
     if( pbridge == 0 ) throw std::runtime_error( "fbridgesequence_: invalid Bridge address" );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Use the device/GPU implementation in the CUDA library
     // (there is also a host implementation in this library)
     pbridge->gpu_sequence( momenta, gs, rndhel, rndcol, *pchannelId, mes, selhel, selcol );
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/fsampler.cc b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/fsampler.cc
index 2fb445372d..3743934f41 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/fsampler.cc
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/fsampler.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Feb 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "mgOnGpuConfig.h"
 
@@ -13,7 +13,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -40,7 +40,7 @@ namespace mg5amcCpu
   private:
     const int m_nevt; // The number of events in each iteration
     int m_iiter;      // The iteration counter (for random number seeding)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     HostBufferRndNumMomenta m_hstRndmom; // Memory buffers for random numbers
     HostBufferMomenta m_hstMomenta;      // Memory buffers for momenta
     HostBufferWeights m_hstWeights;      // Memory buffers for sampling weights
@@ -105,7 +105,7 @@ namespace mg5amcCpu
 
 extern "C"
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/runTest.cc b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/runTest.cc
index d4a760a71b..de327f2321 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/runTest.cc
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/runTest.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Hageboeck (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*xxx to run only testxxx.cc tests
 //----------------------------------------------------------------------------
@@ -18,7 +18,7 @@
 #include "RandomNumberKernels.h"
 #include "epoch_process_id.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using namespace mg5amcGpu;
 #else
 using namespace mg5amcCpu;
@@ -35,7 +35,7 @@ struct CUDA_CPU_TestBase : public TestDriverBase
     : TestDriverBase( npar, refFileName ) {}
 };
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 struct CPUTest : public CUDA_CPU_TestBase
 {
   // Struct data members (process, and memory structures for random numbers, momenta, matrix elements and weights on host and device)
@@ -119,7 +119,7 @@ struct CPUTest : public CUDA_CPU_TestBase
 };
 #endif
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 struct CUDATest : public CUDA_CPU_TestBase
 {
   // Reset the device when our test goes out of scope. Note that this should happen after
@@ -128,7 +128,7 @@ struct CUDATest : public CUDA_CPU_TestBase
   {
     ~DeviceReset()
     {
-      checkCuda( cudaDeviceReset() ); // this is needed by cuda-memcheck --leak-check full
+      checkGpu( gpuDeviceReset() ); // this is needed by cuda-memcheck --leak-check full
     }
   } deviceResetter;
 
@@ -256,7 +256,7 @@ INSTANTIATE_TEST_SUITE_P( prefix, \
                           test_suite_name, \
                           testing::Values( new CUDATest( MG_EPOCH_REFERENCE_FILE_NAME ) ) );
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 MG_INSTANTIATE_TEST_SUITE_GPU( XTESTID_GPU( MG_EPOCH_PROCESS_ID ), MadgraphTest );
 #else
 MG_INSTANTIATE_TEST_SUITE_CPU( XTESTID_CPU( MG_EPOCH_PROCESS_ID ), MadgraphTest );
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/testmisc.cc b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/testmisc.cc
index 895d6eeb56..ba9e59a8a3 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/testmisc.cc
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/testmisc.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*misc to run only testmisc.cc tests
 //----------------------------------------------------------------------------
@@ -17,7 +17,7 @@
 #include <sstream>
 #include <typeinfo>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #define TESTID( s ) s##_GPU_MISC
 #else
 #define TESTID( s ) s##_CPU_MISC
@@ -26,7 +26,7 @@
 #define XTESTID( s ) TESTID( s )
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -59,7 +59,7 @@ namespace mg5amcCpu
 
 TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testmisc )
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/testxxx.cc b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/testxxx.cc
index 3361fe5aa9..e5167de00c 100644
--- a/epochX/cudacpp/gg_ttgg.sa/SubProcesses/testxxx.cc
+++ b/epochX/cudacpp/gg_ttgg.sa/SubProcesses/testxxx.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Apr 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*xxx to run only testxxx.cc tests
 //----------------------------------------------------------------------------
@@ -24,7 +24,7 @@
 #include <iomanip>
 #include <iostream>
 #include <vector>
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #define TESTID( s ) s##_GPU_XXX
 #else
 #define TESTID( s ) s##_CPU_XXX
@@ -32,7 +32,7 @@
 
 #define XTESTID( s ) TESTID( s )
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -42,7 +42,7 @@ namespace mg5amcCpu
   int FPEhandlerIevt = -1;
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU): '" << FPEhandlerMessage << "' ievt=" << FPEhandlerIevt << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU neppV=" << neppV << "): '" << FPEhandlerMessage << "' ievt=" << FPEhandlerIevt << std::endl;
@@ -53,7 +53,7 @@ namespace mg5amcCpu
 
 TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -77,7 +77,7 @@ TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
   assert( nevt % neppM == 0 ); // nevt must be a multiple of neppM
   assert( nevt % neppV == 0 ); // nevt must be a multiple of neppV
   // Fill in the input momenta
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   mg5amcGpu::PinnedHostBufferMomenta hstMomenta( nevt ); // AOSOA[npagM][npar=4][np4=4][neppM]
 #else
   mg5amcCpu::HostBufferMomenta hstMomenta( nevt ); // AOSOA[npagM][npar=4][np4=4][neppM]
@@ -322,7 +322,7 @@ TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
   {
     for( int ievt = 0; ievt < nevt; ievt++ )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       using namespace mg5amcGpu;
 #else
       using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_ttgg.sa/src/HelAmps_sm.h b/epochX/cudacpp/gg_ttgg.sa/src/HelAmps_sm.h
index 8df465ad6d..8b4ad719be 100644
--- a/epochX/cudacpp/gg_ttgg.sa/src/HelAmps_sm.h
+++ b/epochX/cudacpp/gg_ttgg.sa/src/HelAmps_sm.h
@@ -5,7 +5,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -28,7 +28,7 @@
 //#include <iomanip>
 //#include <iostream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttgg.sa/src/Parameters_sm.cc b/epochX/cudacpp/gg_ttgg.sa/src/Parameters_sm.cc
index 64fc3fea62..067445b198 100644
--- a/epochX/cudacpp/gg_ttgg.sa/src/Parameters_sm.cc
+++ b/epochX/cudacpp/gg_ttgg.sa/src/Parameters_sm.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -17,7 +17,7 @@
 #include <iomanip>
 #include <iostream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using namespace mg5amcGpu;
 #else
 using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_ttgg.sa/src/Parameters_sm.h b/epochX/cudacpp/gg_ttgg.sa/src/Parameters_sm.h
index b6568d3761..9581d66e0e 100644
--- a/epochX/cudacpp/gg_ttgg.sa/src/Parameters_sm.h
+++ b/epochX/cudacpp/gg_ttgg.sa/src/Parameters_sm.h
@@ -27,7 +27,7 @@
 #include "read_slha.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -93,7 +93,7 @@ namespace mg5amcCpu
 #include <limits>
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -218,7 +218,7 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -239,7 +239,7 @@ namespace mg5amcCpu
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wunused-variable"  // e.g. <<warning: unused variable ‘mdl_G__exp__2’ [-Wunused-variable]>>
 #pragma GCC diagnostic ignored "-Wunused-parameter" // e.g. <<warning: unused parameter ‘G’ [-Wunused-parameter]>>
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #pragma nv_diagnostic push
 #pragma nv_diag_suppress 177 // e.g. <<warning #177-D: variable "mdl_G__exp__2" was declared but never referenced>>
 #endif
@@ -267,7 +267,7 @@ namespace mg5amcCpu
       // End SM implementation - no special handling of vectors of floats as in EFT (#439)
       return out;
     }
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #pragma GCC diagnostic pop
 #pragma nv_diagnostic pop
 #endif
diff --git a/epochX/cudacpp/gg_ttgg.sa/src/cudacpp_src.mk b/epochX/cudacpp/gg_ttgg.sa/src/cudacpp_src.mk
index d4cc628aec..b2b9da5288 100644
--- a/epochX/cudacpp/gg_ttgg.sa/src/cudacpp_src.mk
+++ b/epochX/cudacpp/gg_ttgg.sa/src/cudacpp_src.mk
@@ -19,7 +19,7 @@ SHELL := /bin/bash
 #=== Configure common compiler flags for CUDA and C++
 
 INCFLAGS = -I.
-OPTFLAGS = -O3 # this ends up in CUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
+OPTFLAGS = -O3 # this ends up in GPUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
 
 #-------------------------------------------------------------------------------
 
@@ -45,13 +45,13 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure the CUDA compiler (note: NVCC is already exported including ccache)
+#=== Configure the CUDA compiler (note: GPUCC is already exported including ccache)
 
-###$(info NVCC=$(NVCC))
+###$(info GPUCC=$(GPUCC))
 
 #-------------------------------------------------------------------------------
 
-#=== Configure ccache for C++ builds (note: NVCC is already exported including ccache)
+#=== Configure ccache for C++ builds (note: GPUCC is already exported including ccache)
 
 # Enable ccache if USECCACHE=1
 ifeq ($(USECCACHE)$(shell echo $(CXX) | grep ccache),1)
@@ -92,6 +92,13 @@ endif
 ###$(info OMPFLAGS=$(OMPFLAGS))
 CXXFLAGS += $(OMPFLAGS)
 
+# Add correct flags for nvcc (-x cu) and hipcc (-x hip) for GPU code (see #810)
+ifeq ($(findstring nvcc,$(GPUCC)),nvcc)
+  GPUFLAGS += -Xcompiler -fPIC -c -x cu
+else ifeq ($(findstring hipcc,$(GPUCC)),hipcc)
+  GPUFLAGS += -fPIC -c -x hip
+endif
+
 # Set the build flags appropriate to each AVX choice (example: "make AVX=none")
 # [NB MGONGPU_PVW512 is needed because "-mprefer-vector-width=256" is not exposed in a macro]
 # [See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96476]
@@ -253,20 +260,20 @@ $(BUILDDIR)/%.o : %.cc *.h $(BUILDDIR)/.build.$(TAG)
 # Generic target and build rules: objects from CUDA compilation
 $(BUILDDIR)/%_cu.o : %.cc *.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c -x cu $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $< -o $@
 
 #-------------------------------------------------------------------------------
 
 cxx_objects=$(addprefix $(BUILDDIR)/, Parameters_sm.o read_slha.o)
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 cu_objects=$(addprefix $(BUILDDIR)/, Parameters_sm_cu.o)
 endif
 
 # Target (and build rules): common (src) library
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so : $(cxx_objects) $(cu_objects)
 	@if [ ! -d $(LIBDIR) ]; then echo "mkdir -p $(LIBDIR)"; mkdir -p $(LIBDIR); fi
-	$(NVCC) -shared -o $@ $(cxx_objects) $(cu_objects) $(LDFLAGS)
+	$(GPUCC) -shared -o $@ $(cxx_objects) $(cu_objects) $(LDFLAGS)
 else
 $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so : $(cxx_objects)
 	@if [ ! -d $(LIBDIR) ]; then echo "mkdir -p $(LIBDIR)"; mkdir -p $(LIBDIR); fi
diff --git a/epochX/cudacpp/gg_ttgg.sa/src/mgOnGpuConfig.h b/epochX/cudacpp/gg_ttgg.sa/src/mgOnGpuConfig.h
index b247654dcf..475749ca7c 100644
--- a/epochX/cudacpp/gg_ttgg.sa/src/mgOnGpuConfig.h
+++ b/epochX/cudacpp/gg_ttgg.sa/src/mgOnGpuConfig.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jul 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUCONFIG_H
 #define MGONGPUCONFIG_H 1
@@ -10,12 +10,26 @@
 // There are two different code bases for standalone_cudacpp (without multichannel) and madevent+cudacpp (with multichannel)
 #undef MGONGPU_SUPPORTS_MULTICHANNEL
 
+// Is this a GPU (CUDA, HIP) or CPU implementation?
+#ifdef __CUDACC__
+#define MGONGPUCPP_GPUIMPL cuda
+#elif defined __HIPCC__
+#define MGONGPUCPP_GPUIMPL hip
+#include "hip/hip_runtime.h" // needed for blockDim, blockIdx, threadIdx: better in mgOnGpuConfig.h than in GpuAbstraction.h
+#else
+#undef MGONGPUCPP_GPUIMPL
+#endif
+
 // ** NB1 Throughputs (e.g. 6.8E8) are events/sec for "./gcheck.exe -p 65536 128 12"
 // ** NB2 Baseline on b7g47n0004 fluctuates (probably depends on load on other VMs)
 
 // Choose if curand is supported for generating random numbers
+// For HIP, by default, do not use curand (common random numbers will be used instead)
 // For both CUDA and C++, by default, do not inline, but allow this macro to be set from outside with e.g. -DMGONGPU_HAS_NO_CURAND
-// (there exist CUDA installations, e.g. using the HPC package, which do not include curand - see PR #784)
+// (there exist CUDA installations, e.g. using the HPC package, which do not include curand - see PR #784 and #785)
+#if defined __HIPCC__
+#define MGONGPU_HAS_NO_CURAND 1
+#else
 //#ifdef __CUDACC__
 //#undef MGONGPU_HAS_NO_CURAND // default
 ////#define MGONGPU_HAS_NO_CURAND 1
@@ -23,6 +37,7 @@
 //#undef MGONGPU_HAS_NO_CURAND // default
 ////#define MGONGPU_HAS_NO_CURAND 1
 //#endif
+#endif
 
 // Choose floating point precision (for everything but color algebra #537)
 // If one of these macros has been set from outside with e.g. -DMGONGPU_FPTYPE_FLOAT, nothing happens (issue #167)
@@ -54,23 +69,31 @@
 //#undef MGONGPU_HARDCODE_PARAM // default
 ////#define MGONGPU_HARDCODE_PARAM 1
 
-// Complex type in c++: std::complex or cxsmpl (CHOOSE ONLY ONE)
-#ifndef __CUDACC__
-//#define MGONGPU_CPPCXTYPE_STDCOMPLEX 1 // ~8 percent slower on float, same on double (5.1E6/double, 9.4E6/float)
-#define MGONGPU_CPPCXTYPE_CXSMPL 1 // new default (5.1E6/double, 10.2E6/float)
-#endif
-
-// Complex type in cuda: thrust or cucomplex or cxsmpl (CHOOSE ONLY ONE)
+// Complex type in CUDA: thrust or cucomplex or cxsmpl (CHOOSE ONLY ONE)
+// (NB THIS IS MGONGPU_*CU*CXTYPE_xxx)
 #ifdef __CUDACC__
 #define MGONGPU_CUCXTYPE_THRUST 1 // default (~1.15E9/double, ~3.2E9/float)
 //#define MGONGPU_CUCXTYPE_CUCOMPLEX 1 // ~10 percent slower (1.03E9/double, ~2.8E9/float)
 //#define MGONGPU_CUCXTYPE_CXSMPL 1 // ~10 percent slower (1.00E9/double, ~2.9E9/float)
+
+// Complex type in HIP: cxsmpl (ONLY ONE OPTION POSSIBLE? #810)
+// (NB THIS IS MGONGPU_*HIP*CXTYPE_xxx)
+#elif defined __HIPCC__
+#define MGONGPU_HIPCXTYPE_CXSMPL 1 // default for HIP
+
+// Complex type in C++: std::complex or cxsmpl (CHOOSE ONLY ONE)
+// (NB THIS IS MGONGPU_*CPP*CXTYPE_xxx)
+#else
+//#define MGONGPU_CPPCXTYPE_STDCOMPLEX 1 // ~8 percent slower on float, same on double (5.1E6/double, 9.4E6/float)
+#define MGONGPU_CPPCXTYPE_CXSMPL 1 // new default (5.1E6/double, 10.2E6/float)
 #endif
 
-// Cuda nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
+// CUDA nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
 #ifdef __CUDACC__
-#undef MGONGPU_NSIGHT_DEBUG // default
+#undef MGONGPU_NSIGHT_DEBUG // default in CUDA
 //#define MGONGPU_NSIGHT_DEBUG 1
+#else
+#undef MGONGPU_NSIGHT_DEBUG // only option in HIP or C++
 #endif
 
 // SANITY CHECKS (floating point precision for everything but color algebra #537)
@@ -86,17 +109,21 @@
 #error You cannot use double precision for color algebra and single precision elsewhere
 #endif
 
-// SANITY CHECKS (c++ complex number implementation)
-#ifndef __CUDACC__
-#if defined MGONGPU_CPPCXTYPE_STDCOMPLEX and defined MGONGPU_CPPCXTYPE_CXSMPL
-#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CPPCXTYPE_STDCOMPLEX or MGONGPU_CPPCXTYPE_CXSMPL
+// SANITY CHECKS (CUDA complex number implementation)
+#ifdef __CUDACC__
+#if defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CUCOMPLEX
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX for CUDA
+#elif defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CXSMPL for CUDA
+#elif defined MGONGPU_CUCXTYPE_CUCOMPLEX and defined MGONGPU_CUCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE OF MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL for CUDA
 #endif
 #endif
 
-// SANITY CHECKS (cuda complex number implementation)
-#ifdef __CUDACC__
-#if defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CUCOMPLEX and defined MGONGPU_CUCXTYPE_CXSMPL
-#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL
+// SANITY CHECKS (C++ complex number implementation)
+#ifndef MGONGPUCPP_GPUIMPL
+#if defined MGONGPU_CPPCXTYPE_STDCOMPLEX and defined MGONGPU_CPPCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CPPCXTYPE_STDCOMPLEX or MGONGPU_CPPCXTYPE_CXSMPL for C++
 #endif
 #endif
 
@@ -134,7 +161,7 @@ namespace mgOnGpu
   // Alignment requirement for using reinterpret_cast with SIMD vectorized code
   // (using reinterpret_cast with non aligned memory may lead to segmentation faults!)
   // Only needed for C++ code but can be enforced also in NVCC builds of C++ code using CUDA>=11.2 and C++17 (#318, #319, #333)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   constexpr int cppAlign = 64; // alignment requirement for SIMD vectorization (64-byte i.e. 512-bit)
 #endif
 
@@ -145,7 +172,7 @@ using mgOnGpu::fptype;
 using mgOnGpu::fptype2;
 
 // C++ SIMD vectorization width (this will be used to set neppV)
-#ifdef __CUDACC__ // CUDA implementation has no SIMD
+#ifdef MGONGPUCPP_GPUIMPL // CUDA and HIP implementations have no SIMD
 #undef MGONGPU_CPPSIMD
 #elif defined __AVX512VL__ && defined MGONGPU_PVW512 // C++ "512z" AVX512 with 512 width (512-bit ie 64-byte): 8 (DOUBLE) or 16 (FLOAT)
 #ifdef MGONGPU_FPTYPE_DOUBLE
@@ -175,9 +202,9 @@ using mgOnGpu::fptype2;
 #undef MGONGPU_CPPSIMD
 #endif
 
-// Cuda nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
+// CUDA nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
 // Arguments (not used so far): text is __FUNCTION__, code is 0 (start) or 1 (end)
-#if defined __CUDACC__ && defined MGONGPU_NSIGHT_DEBUG /* clang-format off */
+#if defined __CUDA__ && defined MGONGPU_NSIGHT_DEBUG /* clang-format off */
 #define mgDebugDeclare() __shared__ float mgDebugCounter[mgOnGpu::ntpbMAX];
 #define mgDebugInitialise() { mgDebugCounter[threadIdx.x] = 0; }
 #define mgDebug( code, text ) { mgDebugCounter[threadIdx.x] += 1; }
@@ -189,8 +216,8 @@ using mgOnGpu::fptype2;
 #define mgDebugFinalise() { /*noop*/ }
 #endif /* clang-format on */
 
-// Define empty CUDA declaration specifiers for C++
-#ifndef __CUDACC__
+// Define empty CUDA/HIP declaration specifiers for C++
+#ifndef MGONGPUCPP_GPUIMPL
 #define __global__
 #define __host__
 #define __device__
diff --git a/epochX/cudacpp/gg_ttgg.sa/src/mgOnGpuCxtypes.h b/epochX/cudacpp/gg_ttgg.sa/src/mgOnGpuCxtypes.h
index ca9a9f00c0..7ede1dbfae 100644
--- a/epochX/cudacpp/gg_ttgg.sa/src/mgOnGpuCxtypes.h
+++ b/epochX/cudacpp/gg_ttgg.sa/src/mgOnGpuCxtypes.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022, based on earlier work by D. Smith) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUCXTYPES_H
 #define MGONGPUCXTYPES_H 1
@@ -19,7 +19,7 @@
 #include <complex>
 
 // Complex type in cuda: thrust or cucomplex or cxsmpl
-#ifdef __CUDACC__
+#ifdef __CUDACC__ // this must be __CUDAC__ (not MGONGPUCPP_GPUIMPL)
 #if defined MGONGPU_CUCXTYPE_THRUST
 #pragma clang diagnostic push
 #pragma clang diagnostic ignored "-Wtautological-compare" // for icpx2021/clang13 (https://stackoverflow.com/a/15864661)
@@ -30,8 +30,13 @@
 #elif not defined MGONGPU_CUCXTYPE_CXSMPL
 #error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL
 #endif
+// Complex type in HIP: cxsmpl
+#elif defined __HIPCC__
+#if not defined MGONGPU_HIPCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_HIPCXTYPE_CXSMPL
+#endif
 #else
-// Complex type in c++: std::complex or cxsmpl
+// Complex type in c++ or HIP: std::complex or cxsmpl
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
 #include <cmath>
 #elif not defined MGONGPU_CPPCXTYPE_CXSMPL
@@ -82,7 +87,7 @@ namespace mgOnGpu /* clang-format off */
 using mgOnGpu::cxsmpl;
 
 // Printout to stream for user defined types
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -92,7 +97,7 @@ namespace mg5amcCpu
   inline __host__ std::ostream&
   operator<<( std::ostream& out, const cxsmpl<FP>& c )
   {
-    out << std::complex( c.real(), c.imag() );
+    out << std::complex<FP>( c.real(), c.imag() );
     return out;
   }
 
@@ -215,14 +220,14 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
   // --- Type definitions (complex type: cxtype)
-#ifdef __CUDACC__ // cuda
+#ifdef __CUDACC__ // this must be __CUDAC__ (not MGONGPUCPP_GPUIMPL)
 #if defined MGONGPU_CUCXTYPE_THRUST
   typedef thrust::complex<fptype> cxtype;
 #elif defined MGONGPU_CUCXTYPE_CUCOMPLEX
@@ -255,13 +260,13 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
-#if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+#if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
 
   //------------------------------
   // CUDA or C++ - using cxsmpl
@@ -303,11 +308,11 @@ namespace mg5amcCpu
     return cxmake( c.real(), c.imag() );
   }
 
-#endif // #if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+#endif // #if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
 
   //==========================================================================
 
-#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_THRUST // cuda + thrust
+#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_THRUST // cuda + thrust (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // CUDA - using thrust::complex
@@ -347,7 +352,7 @@ namespace mg5amcCpu
 
   //==========================================================================
 
-#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_CUCOMPLEX // cuda + cucomplex
+#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_CUCOMPLEX // cuda + cucomplex (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // CUDA - using cuComplex
@@ -566,7 +571,7 @@ namespace mg5amcCpu
 
   //==========================================================================
 
-#if not defined __CUDACC__ and defined MGONGPU_CPPCXTYPE_STDCOMPLEX // c++ + stdcomplex
+#if not defined __CUDACC__ and defined MGONGPU_CPPCXTYPE_STDCOMPLEX // c++/hip + stdcomplex (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // C++ - using std::complex
@@ -633,7 +638,7 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttgg.sa/src/mgOnGpuFptypes.h b/epochX/cudacpp/gg_ttgg.sa/src/mgOnGpuFptypes.h
index 905c97d700..fa3a02664b 100644
--- a/epochX/cudacpp/gg_ttgg.sa/src/mgOnGpuFptypes.h
+++ b/epochX/cudacpp/gg_ttgg.sa/src/mgOnGpuFptypes.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUFPTYPES_H
 #define MGONGPUFPTYPES_H 1
@@ -12,7 +12,7 @@
 #include <cmath>
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL // cuda
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -20,7 +20,7 @@ namespace mg5amcCpu
 {
   //==========================================================================
 
-#ifdef __CUDACC__ // cuda
+#ifdef MGONGPUCPP_GPUIMPL // cuda
 
   //------------------------------
   // Floating point types - Cuda
@@ -64,11 +64,11 @@ namespace mg5amcCpu
 #endif
   }
 
-#endif // #ifdef __CUDACC__
+#endif // #ifdef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
   //------------------------------
   // Floating point types - C++
@@ -92,7 +92,7 @@ namespace mg5amcCpu
     return std::sqrt( f );
   }
 
-#endif // #ifndef __CUDACC__
+#endif // #ifndef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
diff --git a/epochX/cudacpp/gg_ttgg.sa/src/mgOnGpuVectors.h b/epochX/cudacpp/gg_ttgg.sa/src/mgOnGpuVectors.h
index e1299ba81e..cdae04326b 100644
--- a/epochX/cudacpp/gg_ttgg.sa/src/mgOnGpuVectors.h
+++ b/epochX/cudacpp/gg_ttgg.sa/src/mgOnGpuVectors.h
@@ -32,7 +32,7 @@
 #endif
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -131,7 +131,7 @@ namespace mg5amcCpu
 #endif
 #endif
 
-#else // i.e #ifndef MGONGPU_CPPSIMD (this includes #ifdef __CUDACC__)
+#else // i.e #ifndef MGONGPU_CPPSIMD (this includes #ifdef MGONGPUCPP_GPUIMPL)
 
   const int neppV = 1;
 
@@ -153,13 +153,13 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
   // Printout to stream for user defined types
 
@@ -805,11 +805,11 @@ namespace mg5amcCpu
 
 #endif // #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
 
-#endif // #ifndef __CUDACC__
+#endif // #ifndef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
   //------------------------------
   // Vector types - CUDA
@@ -853,12 +853,12 @@ namespace mg5amcCpu
     return mask;
   }
 
-#endif // #ifdef __CUDACC__
+#endif // #ifdef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
   // Scalar-or-vector types: scalar in CUDA, vector or scalar in C++
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   typedef bool bool_sv;
   typedef fptype fptype_sv;
   typedef fptype2 fptype2_sv;
@@ -879,7 +879,7 @@ namespace mg5amcCpu
 #endif
 
   // Scalar-or-vector zeros: scalar in CUDA, vector or scalar in C++
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   inline __host__ __device__ cxtype cxzero_sv(){ return cxtype( 0, 0 ); }
 #elif defined MGONGPU_CPPSIMD
   inline cxtype_v cxzero_sv() { return cxtype_v(); } // RRRR=0000 IIII=0000
diff --git a/epochX/cudacpp/gg_ttgg.sa/src/rambo.h b/epochX/cudacpp/gg_ttgg.sa/src/rambo.h
index e02ea52496..cd7e1008ea 100644
--- a/epochX/cudacpp/gg_ttgg.sa/src/rambo.h
+++ b/epochX/cudacpp/gg_ttgg.sa/src/rambo.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -18,7 +18,7 @@
 #include <iostream>
 
 // Simplified rambo version for 2 to N (with N>=2) processes with massless particles
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +83,7 @@ namespace mg5amcCpu
       static bool first = true;
       if( first )
       {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
         if constexpr( M_ACCESS::isOnDevice() ) // avoid
         {
           const int ievt0 = 0;
@@ -166,7 +166,7 @@ namespace mg5amcCpu
     wt = po2log;
     if( nparf != 2 ) wt = ( 2. * nparf - 4. ) * log( energy ) + z[nparf - 1];
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     // issue warnings if weight is too small or too large
     static int iwarn[5] = { 0, 0, 0, 0, 0 };
     if( wt < -180. )
diff --git a/epochX/cudacpp/gg_ttgg.sa/src/read_slha.cc b/epochX/cudacpp/gg_ttgg.sa/src/read_slha.cc
index 055b19a779..f8e46f2e66 100644
--- a/epochX/cudacpp/gg_ttgg.sa/src/read_slha.cc
+++ b/epochX/cudacpp/gg_ttgg.sa/src/read_slha.cc
@@ -11,7 +11,11 @@
 
 #include <algorithm>
 #include <cstdlib>
-#include <filesystem>
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
 #include <fstream>
 #include <iostream>
 
@@ -60,7 +64,14 @@ SLHAReader::read_slha_file( std::string file_name, bool verbose )
     {
       std::cout << "WARNING! Card file '" << file_name << "' does not exist:"
                 << " look for the file in directory $" << envpath << "='" << getenv( envpath ) << "'" << std::endl;
+      /*
+#ifdef __HIPCC__
+      const std::string file_name2 = std::experimental::filesystem::path( getenv( envpath ) ) / std::experimental::filesystem::path( file_name ).filename();
+#else
       const std::string file_name2 = std::filesystem::path( getenv( envpath ) ) / std::filesystem::path( file_name ).filename();
+#endif
+      */
+      const std::string file_name2 = std::string( getenv( envpath ) ) + "/" + file_name; // bypass std::filesystem #803
       param_card.open( file_name2.c_str(), std::ifstream::in );
       if( param_card.good() )
       {
diff --git a/epochX/cudacpp/gg_ttggg.mad/CODEGEN_mad_gg_ttggg_log.txt b/epochX/cudacpp/gg_ttggg.mad/CODEGEN_mad_gg_ttggg_log.txt
index 2480a22f8d..37ad313b62 100644
--- a/epochX/cudacpp/gg_ttggg.mad/CODEGEN_mad_gg_ttggg_log.txt
+++ b/epochX/cudacpp/gg_ttggg.mad/CODEGEN_mad_gg_ttggg_log.txt
@@ -52,7 +52,7 @@ Note that you can still compile and run aMC@NLO with the built-in PDFs
 
 Using default text editor "vi". Set another one in ./input/mg5_configuration.txt
 Using default eps viewer "evince". Set another one in ./input/mg5_configuration.txt
-No valid web browser found. Please set in ./input/mg5_configuration.txt
+Using default web browser "firefox". Set another one in ./input/mg5_configuration.txt
 import /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_mad_gg_ttggg.mg
 The import format was not given, so we guess it as command
 set stdout_level DEBUG
@@ -62,7 +62,7 @@ generate g g > t t~ g g g
 No model currently active, so we import the Standard Model
 INFO: load particles 
 INFO: load vertices 
-DEBUG: model prefixing  takes 0.005784511566162109 
+DEBUG: model prefixing  takes 0.005320072174072266 
 INFO: Restrict model sm with file models/sm/restrict_default.dat . 
 DEBUG: Simplifying conditional expressions 
 DEBUG: remove interactions: u s w+ at order: QED=1 
@@ -155,7 +155,7 @@ INFO: Please specify coupling orders to bypass this step.
 INFO: Trying coupling order WEIGHTED<=5: WEIGTHED IS QCD+2*QED 
 INFO: Trying process: g g > t t~ g g g WEIGHTED<=5 @1  
 INFO: Process has 1240 diagrams 
-1 processes with 1240 diagrams generated in 1.929 s
+1 processes with 1240 diagrams generated in 1.852 s
 Total: 1 processes with 1240 diagrams
 output madevent ../TMPOUT/CODEGEN_mad_gg_ttggg --hel_recycling=False --vector_size=32 --me_exporter=standalone_cudacpp
 Load PLUGIN.CUDACPP_OUTPUT
@@ -177,7 +177,7 @@ INFO: Creating files in directory P1_gg_ttxggg
 INFO: Computing Color-Flow optimization [15120 term] 
 INFO: Color-Flow passed to 1630 term in 8s. Introduce 3030 contraction 
 DEBUG:  kwargs[prefix] = 0 [model_handling.py at line 1058] 
-DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7fc088fe88e0> [export_v4.py at line 6262] 
+DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7f1803d2a8e0> [export_v4.py at line 6262] 
 INFO: Creating files in directory . 
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.h
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.cc
@@ -192,15 +192,15 @@ INFO: Created files CPPProcess.h and CPPProcess.cc in directory ./.
 DEBUG:  vector, subproc_group,self.opt['vector_size'] =  32 True 32 [export_v4.py at line 1872] 
 INFO: Generating Feynman diagrams for Process: g g > t t~ g g g WEIGHTED<=5 @1 
 INFO: Finding symmetric diagrams for subprocess group gg_ttxggg 
-Generated helas calls for 1 subprocesses (1240 diagrams) in 6.718 s
-Wrote files for 2281 helas calls in 18.893 s
+Generated helas calls for 1 subprocesses (1240 diagrams) in 6.532 s
+Wrote files for 2281 helas calls in 18.428 s
 ALOHA: aloha starts to compute helicity amplitudes
 ALOHA: aloha creates VVV1 routines
 ALOHA: aloha creates FFV1 routines
 ALOHA: aloha creates VVVV1 routines
 ALOHA: aloha creates VVVV3 routines
 ALOHA: aloha creates VVVV4 routines
-ALOHA: aloha creates 5 routines in  0.322 s
+ALOHA: aloha creates 5 routines in  0.343 s
 DEBUG:  Entering PLUGIN_ProcessExporter.convert_model (create the model) [output.py at line 202] 
 ALOHA: aloha starts to compute helicity amplitudes
 ALOHA: aloha creates VVV1 routines
@@ -208,7 +208,7 @@ ALOHA: aloha creates FFV1 routines
 ALOHA: aloha creates VVVV1 routines
 ALOHA: aloha creates VVVV3 routines
 ALOHA: aloha creates VVVV4 routines
-ALOHA: aloha creates 10 routines in  0.319 s
+ALOHA: aloha creates 10 routines in  0.327 s
 <class 'aloha.create_aloha.AbstractRoutine'> VVV1
 <class 'aloha.create_aloha.AbstractRoutine'> VVV1
 <class 'aloha.create_aloha.AbstractRoutine'> FFV1
@@ -257,10 +257,10 @@ Type "launch" to generate events from this process, or see
 Run "open index.html" to see more information about this process.
 quit
 
-real	0m29.815s
-user	0m29.332s
-sys	0m0.380s
-Code generation completed in 30 seconds
+real	0m29.153s
+user	0m28.584s
+sys	0m0.440s
+Code generation completed in 29 seconds
 ************************************************************
 *                                                          *
 *                      W E L C O M E to                    *
@@ -286,7 +286,7 @@ INFO: load configuration from /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/mg5amc
 INFO: load configuration from /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_mad_gg_ttggg/Cards/me5_configuration.txt  
 Using default text editor "vi". Set another one in ./input/mg5_configuration.txt
 Using default eps viewer "evince". Set another one in ./input/mg5_configuration.txt
-No valid web browser found. Please set in ./input/mg5_configuration.txt
+Using default web browser "firefox". Set another one in ./input/mg5_configuration.txt
 treatcards run
 quit
 INFO:  
@@ -316,7 +316,7 @@ INFO: load configuration from /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/mg5amc
 INFO: load configuration from /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_mad_gg_ttggg/Cards/me5_configuration.txt  
 Using default text editor "vi". Set another one in ./input/mg5_configuration.txt
 Using default eps viewer "evince". Set another one in ./input/mg5_configuration.txt
-No valid web browser found. Please set in ./input/mg5_configuration.txt
+Using default web browser "firefox". Set another one in ./input/mg5_configuration.txt
 treatcards param
 quit
 INFO:  
diff --git a/epochX/cudacpp/gg_ttggg.mad/COPYRIGHT b/epochX/cudacpp/gg_ttggg.mad/COPYRIGHT
index a134b5fef9..9036d9260a 100644
--- a/epochX/cudacpp/gg_ttggg.mad/COPYRIGHT
+++ b/epochX/cudacpp/gg_ttggg.mad/COPYRIGHT
@@ -15,6 +15,7 @@ The full development team currently includes the following authors :
   Stephan Hageboeck (CERN)
   Olivier Mattelaer (Universite Catholique de Louvain, original author)
   Stefan Roiser (CERN, original author)
+  Jorgen Teig (CERN)
   Andrea Valassi (CERN, original author)
   Zenny Wettersten (CERN)
 See https://github.com/madgraph5/madgraph4gpu for more details. For the full
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/Bridge.h b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/Bridge.h
index bf8b5e024d..f9ed70dfde 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/Bridge.h
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/Bridge.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Roiser (Nov 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef BRIDGE_H
 #define BRIDGE_H 1
@@ -14,16 +14,23 @@
 #include "MemoryAccessMomenta.h"  // for MemoryAccessMomenta::neppM
 #include "MemoryBuffers.h"        // for HostBufferMomenta, DeviceBufferMomenta etc
 
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
+
+#include <sys/stat.h> // bypass std::filesystem #803
+
 #include <algorithm>
 #include <cassert>
 #include <cmath>
 #include <cstring>
-#include <filesystem>
 #include <iostream>
 #include <memory>
 #include <type_traits>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +90,7 @@ namespace mg5amcCpu
     Bridge& operator=( const Bridge& ) = delete;
     Bridge& operator=( Bridge&& ) = delete;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     /**
      * Set the gpublocks and gputhreads for the gpusequence - throws if evnt != gpublocks*gputhreads
      * (this is needed for BridgeKernel tests rather than for actual production use in Fortran)
@@ -150,7 +157,7 @@ namespace mg5amcCpu
     unsigned int m_nevt; // number of events
     int m_nGoodHel;      // the number of good helicities (-1 initially when they have not yet been calculated)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     int m_gputhreads; // number of gpu threads (default set from number of events, can be modified)
     int m_gpublocks;  // number of gpu blocks (default set from number of events, can be modified)
     DeviceBuffer<FORTRANFPTYPE, sizePerEventMomenta> m_devMomentaF;
@@ -187,12 +194,12 @@ namespace mg5amcCpu
   // Forward declare transposition methods
   //
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
   template<typename Tin, typename Tout>
   __global__ void dev_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt );
 
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
 
   template<typename Tin, typename Tout>
   void hst_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt );
@@ -209,7 +216,7 @@ namespace mg5amcCpu
   Bridge<FORTRANFPTYPE>::Bridge( unsigned int nevtF, unsigned int nparF, unsigned int np4F )
     : m_nevt( nevtF )
     , m_nGoodHel( -1 )
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     , m_gputhreads( 256 )                  // default number of gpu threads
     , m_gpublocks( m_nevt / m_gputhreads ) // this ensures m_nevt <= m_gpublocks*m_gputhreads
     , m_devMomentaF( m_nevt )
@@ -233,7 +240,7 @@ namespace mg5amcCpu
   {
     if( nparF != CPPProcess::npar ) throw std::runtime_error( "Bridge constructor: npar mismatch" );
     if( np4F != CPPProcess::np4 ) throw std::runtime_error( "Bridge constructor: np4 mismatch" );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( ( m_nevt < s_gputhreadsmin ) || ( m_nevt % s_gputhreadsmin != 0 ) )
       throw std::runtime_error( "Bridge constructor: nevt should be a multiple of " + std::to_string( s_gputhreadsmin ) );
     while( m_nevt != m_gpublocks * m_gputhreads )
@@ -249,20 +256,28 @@ namespace mg5amcCpu
 #else
     std::cout << "WARNING! Instantiate host Bridge (nevt=" << m_nevt << ")" << std::endl;
     m_pmek.reset( new MatrixElementKernelHost( m_hstMomentaC, m_hstGs, m_hstRndHel, m_hstRndCol, m_hstMEs, m_hstSelHel, m_hstSelCol, m_nevt ) );
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
     // Create a process object, read param card and set parameters
     // FIXME: the process instance can happily go out of scope because it is only needed to read parameters?
     // FIXME: the CPPProcess should really be a singleton? what if fbridgecreate is called from several Fortran threads?
     CPPProcess process( /*verbose=*/false );
     std::string paramCard = "../../Cards/param_card.dat";
-    if( !std::filesystem::exists( paramCard ) )
-    {
-      paramCard = "../" + paramCard;
-    }
+    /*
+#ifdef __HIPCC__
+    if( !std::experimental::filesystem::exists( paramCard ) ) paramCard = "../" + paramCard;
+#else
+    if( !std::filesystem::exists( paramCard ) ) paramCard = "../" + paramCard;
+#endif
+    */
+    //struct stat dummybuffer; // bypass std::filesystem #803
+    //if( !( stat( paramCard.c_str(), &dummyBuffer ) == 0 ) ) paramCard = "../" + paramCard; //
+    auto fileExists = []( std::string& fileName )
+    { struct stat buffer; return stat( fileName.c_str(), &buffer ) == 0; };
+    if( !fileExists( paramCard ) ) paramCard = "../" + paramCard; // bypass std::filesystem #803
     process.initProc( paramCard );
   }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::set_gpugrid( const int gpublocks, const int gputhreads )
   {
@@ -276,7 +291,7 @@ namespace mg5amcCpu
   }
 #endif
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::gpu_sequence( const FORTRANFPTYPE* momenta,
                                             const FORTRANFPTYPE* gs,
@@ -291,14 +306,14 @@ namespace mg5amcCpu
     constexpr int neppM = MemoryAccessMomenta::neppM;
     if constexpr( neppM == 1 && std::is_same_v<FORTRANFPTYPE, fptype> )
     {
-      checkCuda( cudaMemcpy( m_devMomentaC.data(), momenta, m_devMomentaC.bytes(), cudaMemcpyHostToDevice ) );
+      gpuMemcpy( m_devMomentaC.data(), momenta, m_devMomentaC.bytes(), gpuMemcpyHostToDevice );
     }
     else
     {
-      checkCuda( cudaMemcpy( m_devMomentaF.data(), momenta, m_devMomentaF.bytes(), cudaMemcpyHostToDevice ) );
+      gpuMemcpy( m_devMomentaF.data(), momenta, m_devMomentaF.bytes(), gpuMemcpyHostToDevice );
       const int thrPerEvt = CPPProcess::npar * CPPProcess::np4; // AV: transpose alg does 1 element per thread (NOT 1 event per thread)
       //const int thrPerEvt = 1; // AV: try new alg with 1 event per thread... this seems slower
-      dev_transposeMomentaF2C<<<m_gpublocks * thrPerEvt, m_gputhreads>>>( m_devMomentaF.data(), m_devMomentaC.data(), m_nevt );
+      gpuLaunchKernel( dev_transposeMomentaF2C, m_gpublocks * thrPerEvt, m_gputhreads, m_devMomentaF.data(), m_devMomentaC.data(), m_nevt );
     }
     if constexpr( std::is_same_v<FORTRANFPTYPE, fptype> )
     {
@@ -341,7 +356,7 @@ namespace mg5amcCpu
   }
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::cpu_sequence( const FORTRANFPTYPE* momenta,
                                             const FORTRANFPTYPE* gs,
@@ -396,7 +411,7 @@ namespace mg5amcCpu
   // - C++ array: momenta[npagM][npar][np4][neppM] with nevt=npagM*neppM (AOSOA)
   //
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename Tin, typename Tout>
   __global__ void dev_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt )
   {
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/BridgeKernels.cc b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/BridgeKernels.cc
index d58066c9c1..eaf4037a24 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/BridgeKernels.cc
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/BridgeKernels.cc
@@ -1,17 +1,18 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "BridgeKernels.h"
 
+#include "GpuAbstraction.h"
 #include "MemoryAccessMomenta.h"
 
 #include <sstream>
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -45,7 +46,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu
 {
 
@@ -96,7 +97,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/BridgeKernels.h b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/BridgeKernels.h
index 15eb4bff4d..3efef8ce97 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/BridgeKernels.h
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/BridgeKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef BRIDGEKERNELS_H
 #define BRIDGEKERNELS_H 1
@@ -12,7 +12,7 @@
 #include "MatrixElementKernels.h"
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -49,7 +49,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A Bridge wrapper class encapsulating matrix element calculations on a CPU host
   class BridgeKernelHost final : public BridgeKernelBase
   {
@@ -89,7 +89,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A Bridge wrapper class encapsulating matrix element calculations on a GPU device
   class BridgeKernelDevice : public BridgeKernelBase
   {
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/CommonRandomNumberKernel.cc b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/CommonRandomNumberKernel.cc
index 985b39f576..010bc4cbd0 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/CommonRandomNumberKernel.cc
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/CommonRandomNumberKernel.cc
@@ -1,15 +1,16 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "CommonRandomNumbers.h"
+#include "GpuAbstraction.h"
 #include "MemoryBuffers.h"
 #include "RandomNumberKernels.h"
 
 #include <cassert>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/CrossSectionKernels.cc b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/CrossSectionKernels.cc
index 0b355a3c8d..c15b39844d 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/CrossSectionKernels.cc
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/CrossSectionKernels.cc
@@ -1,10 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "CrossSectionKernels.h"
 
+#include "GpuAbstraction.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessWeights.h"
 #include "MemoryBuffers.h"
@@ -77,7 +78,7 @@ debug_me_is_abnormal( const fptype& me, size_t ievtALL )
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -185,7 +186,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/CrossSectionKernels.h b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/CrossSectionKernels.h
index 7933ca4bbf..4d9659e04e 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/CrossSectionKernels.h
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/CrossSectionKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef CROSSSECTIONKERNELS_H
 #define CROSSSECTIONKERNELS_H 1
@@ -13,7 +13,7 @@
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -96,7 +96,7 @@ namespace mg5amcCpu
   //--------------------------------------------------------------------------
 
   /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating the calculation of event statistics on a GPU device
   class CrossSectionKernelDevice : public CrossSectionKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/CudaRuntime.h b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/CudaRuntime.h
deleted file mode 100644
index 64ce52f4b3..0000000000
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/CudaRuntime.h
+++ /dev/null
@@ -1,85 +0,0 @@
-// Copyright (C) 2020-2023 CERN and UCLouvain.
-// Licensed under the GNU Lesser General Public License (version 3 or later).
-// Created by: S. Roiser (Jul 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
-
-#ifndef MG5AMC_CUDARUNTIME_H
-#define MG5AMC_CUDARUNTIME_H 1
-
-// MG5AMC on GPU uses the CUDA runtime API, not the lower level CUDA driver API
-// See https://docs.nvidia.com/cuda/cuda-runtime-api/driver-vs-runtime-api.html#driver-vs-runtime-api
-
-#include <cassert>
-#include <iostream>
-
-//--------------------------------------------------------------------------
-
-// See https://stackoverflow.com/a/14038590
-#ifdef __CUDACC__ /* clang-format off */
-#define checkCuda( code ) { assertCuda( code, __FILE__, __LINE__ ); }
-inline void assertCuda( cudaError_t code, const char* file, int line, bool abort = true )
-{
-  if( code != cudaSuccess )
-  {
-    printf( "ERROR! assertCuda: '%s' (%d) in %s:%d\n", cudaGetErrorString( code ), code, file, line );
-    if( abort ) assert( code == cudaSuccess );
-  }
-}
-#endif /* clang-format on */
-
-//--------------------------------------------------------------------------
-
-#ifdef __CUDACC__
-namespace mg5amcGpu
-{
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  // *** FIXME! This will all need to be designed differently when going to multi-GPU nodes! ***
-  struct CudaRuntime final
-  {
-    CudaRuntime( const bool debug = true )
-      : m_debug( debug ) { setUp( m_debug ); }
-    ~CudaRuntime() { tearDown( m_debug ); }
-    CudaRuntime( const CudaRuntime& ) = delete;
-    CudaRuntime( CudaRuntime&& ) = delete;
-    CudaRuntime& operator=( const CudaRuntime& ) = delete;
-    CudaRuntime& operator=( CudaRuntime&& ) = delete;
-    bool m_debug;
-
-    // Set up CUDA application
-    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
-    // Calling cudaSetDevice on startup is useful to properly book-keep the time spent in CUDA initialization
-    static void setUp( const bool debug = true )
-    {
-      // ** NB: it is useful to call cudaSetDevice, or cudaFree, to properly book-keep the time spent in CUDA initialization
-      // ** NB: otherwise, the first CUDA operation (eg a cudaMemcpyToSymbol in CPPProcess ctor) appears to take much longer!
-      /*
-      // [We initially added cudaFree(0) to "ease profile analysis" only because it shows up as a big recognizable block!]
-      // No explicit initialization is needed: https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#initialization
-      // It is not clear what cudaFree(0) does at all: https://stackoverflow.com/questions/69967813/
-      if ( debug ) std::cout << "__CudaRuntime: calling cudaFree(0)" << std::endl;
-      checkCuda( cudaFree( 0 ) ); // SLOW!
-      */
-      // Replace cudaFree(0) by cudaSetDevice(0), even if it is not really needed either
-      // (but see https://developer.nvidia.com/blog/cuda-pro-tip-always-set-current-device-avoid-multithreading-bugs)
-      if( debug ) std::cout << "__CudaRuntime: calling cudaSetDevice(0)" << std::endl;
-      checkCuda( cudaSetDevice( 0 ) ); // SLOW!
-    }
-
-    // Tear down CUDA application (call cudaDeviceReset)
-    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
-    // Calling cudaDeviceReset on shutdown is only needed for checking memory leaks in cuda-memcheck
-    // See https://docs.nvidia.com/cuda/cuda-memcheck/index.html#leak-checking
-    static void tearDown( const bool debug = true )
-    {
-      if( debug ) std::cout << "__CudaRuntime: calling cudaDeviceReset()" << std::endl;
-      checkCuda( cudaDeviceReset() );
-    }
-  };
-
-}
-#endif
-
-//--------------------------------------------------------------------------
-
-#endif // MG5AMC_CUDARUNTIME_H
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/CurandRandomNumberKernel.cc b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/CurandRandomNumberKernel.cc
index eb56333b03..08a16f6f2c 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/CurandRandomNumberKernel.cc
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/CurandRandomNumberKernel.cc
@@ -1,9 +1,9 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "MemoryBuffers.h"
 #include "RandomNumberKernels.h"
 
@@ -22,7 +22,7 @@ inline void assertCurand( curandStatus_t code, const char *file, int line, bool
 }
 #endif /* clang-format on */
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -36,7 +36,7 @@ namespace mg5amcCpu
   {
     if( m_isOnDevice )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if( !m_rnarray.isOnDevice() )
         throw std::runtime_error( "CurandRandomNumberKernel on device with a host random number array" );
 #else
@@ -114,7 +114,7 @@ namespace mg5amcCpu
     /*
     printf( "\nCurandRandomNumberKernel::generateRnarray size = %d\n", (int)m_rnarray.size() );
     fptype* data = m_rnarray.data();
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( m_rnarray.isOnDevice() )
     {
       data = new fptype[m_rnarray.size()]();
@@ -123,7 +123,7 @@ namespace mg5amcCpu
 #endif
     for( int i = 0; i < ( (int)m_rnarray.size() / 4 ); i++ )
       printf( "[%4d] %f %f %f %f\n", i * 4, data[i * 4], data[i * 4 + 2], data[i * 4 + 2], data[i * 4 + 3] );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( m_rnarray.isOnDevice() ) delete[] data;
 #endif
     */
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/EventStatistics.h b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/EventStatistics.h
index 48b51e0a49..b425a5bade 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/EventStatistics.h
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/EventStatistics.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef EventStatistics_H
 #define EventStatistics_H 1
@@ -16,7 +16,7 @@
 #include <limits>
 #include <string>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/GpuAbstraction.h b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/GpuAbstraction.h
new file mode 100644
index 0000000000..9c467b1e04
--- /dev/null
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/GpuAbstraction.h
@@ -0,0 +1,69 @@
+// Copyright (C) 2020-2023 CERN and UCLouvain.
+// Licensed under the GNU Lesser General Public License (version 3 or later).
+// Created by: J. Teig (Jul 2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+
+#ifndef MG5AMC_GPUABSTRACTION_H
+#define MG5AMC_GPUABSTRACTION_H 1
+
+#include <cassert>
+
+//--------------------------------------------------------------------------
+
+#ifdef __CUDACC__
+
+#define gpuError_t cudaError_t
+#define gpuPeekAtLastError cudaPeekAtLastError
+#define gpuGetErrorString cudaGetErrorString
+#define gpuSuccess cudaSuccess
+
+#define gpuMallocHost( ptr, size ) checkGpu( cudaMallocHost( ptr, size ) )
+#define gpuMalloc( ptr, size ) checkGpu( cudaMalloc( ptr, size ) )
+
+#define gpuMemcpy( dstData, srcData, srcBytes, func ) checkGpu( cudaMemcpy( dstData, srcData, srcBytes, func ) )
+#define gpuMemcpyHostToDevice cudaMemcpyHostToDevice
+#define gpuMemcpyDeviceToHost cudaMemcpyDeviceToHost
+#define gpuMemcpyToSymbol( type1, type2, size ) checkGpu( cudaMemcpyToSymbol( type1, type2, size ) )
+
+#define gpuFree( ptr ) checkGpu( cudaFree( ptr ) )
+#define gpuFreeHost( ptr ) checkGpu( cudaFreeHost( ptr ) )
+
+#define gpuSetDevice cudaSetDevice
+#define gpuDeviceSynchronize cudaDeviceSynchronize
+#define gpuDeviceReset cudaDeviceReset
+
+#define gpuLaunchKernel( kernel, blocks, threads, ... ) kernel<<<blocks, threads>>>( __VA_ARGS__ )
+#define gpuLaunchKernelSharedMem( kernel, blocks, threads, sharedMem, ... ) kernel<<<blocks, threads, sharedMem>>>( __VA_ARGS__ )
+
+//--------------------------------------------------------------------------
+
+#elif defined __HIPCC__
+
+#define gpuError_t hipError_t
+#define gpuPeekAtLastError hipPeekAtLastError
+#define gpuGetErrorString hipGetErrorString
+#define gpuSuccess hipSuccess
+
+#define gpuMallocHost( ptr, size ) checkGpu( hipHostMalloc( ptr, size ) ) // HostMalloc better
+#define gpuMalloc( ptr, size ) checkGpu( hipMalloc( ptr, size ) )
+
+#define gpuMemcpy( dstData, srcData, srcBytes, func ) checkGpu( hipMemcpy( dstData, srcData, srcBytes, func ) )
+#define gpuMemcpyHostToDevice hipMemcpyHostToDevice
+#define gpuMemcpyDeviceToHost hipMemcpyDeviceToHost
+#define gpuMemcpyToSymbol( type1, type2, size ) checkGpu( hipMemcpyToSymbol( type1, type2, size ) )
+
+#define gpuFree( ptr ) checkGpu( hipFree( ptr ) )
+#define gpuFreeHost( ptr ) checkGpu( hipHostFree( ptr ) )
+
+#define gpuSetDevice hipSetDevice
+#define gpuDeviceSynchronize hipDeviceSynchronize
+#define gpuDeviceReset hipDeviceReset
+
+#define gpuLaunchKernel( kernel, blocks, threads, ... ) kernel<<<blocks, threads>>>( __VA_ARGS__ )
+#define gpuLaunchKernelSharedMem( kernel, blocks, threads, sharedMem, ... ) kernel<<<blocks, threads, sharedMem>>>( __VA_ARGS__ )
+
+//--------------------------------------------------------------------------
+
+#endif
+
+#endif // MG5AMC_GPUABSTRACTION_H
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/GpuRuntime.h b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/GpuRuntime.h
new file mode 100644
index 0000000000..93579ef08b
--- /dev/null
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/GpuRuntime.h
@@ -0,0 +1,85 @@
+// Copyright (C) 2020-2023 CERN and UCLouvain.
+// Licensed under the GNU Lesser General Public License (version 3 or later).
+// Created by: J. Teig (Jun 2023, based on earlier work by S. Roiser) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+
+#ifndef MG5AMC_GPURUNTIME_H
+#define MG5AMC_GPURUNTIME_H 1
+
+// MG5AMC on GPU uses the CUDA runtime API, not the lower level CUDA driver API
+// See https://docs.nvidia.com/cuda/cuda-runtime-api/driver-vs-runtime-api.html#driver-vs-runtime-api
+
+#include "GpuAbstraction.h"
+
+#include <iostream>
+
+//--------------------------------------------------------------------------
+
+// See https://stackoverflow.com/a/14038590
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
+#define checkGpu( code ) { assertGpu( code, __FILE__, __LINE__ ); }
+inline void assertGpu( gpuError_t code, const char* file, int line, bool abort = true )
+{
+  if( code != gpuSuccess )
+  {
+    printf( "ERROR! assertGpu: '%s' (%d) in %s:%d\n", gpuGetErrorString( code ), code, file, line );
+    if( abort ) assert( code == gpuSuccess );
+  }
+}
+#endif /* clang-format on */
+
+//--------------------------------------------------------------------------
+
+#ifdef MGONGPUCPP_GPUIMPL
+namespace mg5amcGpu
+{
+  // Instantiate a GpuRuntime at the beginnining of the application's main to
+  // invoke gpuSetDevice(0) in the constructor and book a gpuDeviceReset() call in the destructor
+  // *** FIXME! This will all need to be designed differently when going to multi-GPU nodes! ***
+  struct GpuRuntime final
+  {
+    GpuRuntime( const bool debug = true )
+      : m_debug( debug ) { setUp( m_debug ); }
+    ~GpuRuntime() { tearDown( m_debug ); }
+    GpuRuntime( const GpuRuntime& ) = delete;
+    GpuRuntime( GpuRuntime&& ) = delete;
+    GpuRuntime& operator=( const GpuRuntime& ) = delete;
+    GpuRuntime& operator=( GpuRuntime&& ) = delete;
+    bool m_debug;
+
+    // Set up CUDA application
+    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
+    // Calling cudaSetDevice on startup is useful to properly book-keep the time spent in CUDA initialization
+    static void setUp( const bool debug = true )
+    {
+      // ** NB: it is useful to call cudaSetDevice, or cudaFree, to properly book-keep the time spent in CUDA initialization
+      // ** NB: otherwise, the first CUDA operation (eg a cudaMemcpyToSymbol in CPPProcess ctor) appears to take much longer!
+      /*
+      // [We initially added cudaFree(0) to "ease profile analysis" only because it shows up as a big recognizable block!]
+      // No explicit initialization is needed: https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#initialization
+      // It is not clear what cudaFree(0) does at all: https://stackoverflow.com/questions/69967813/
+      if ( debug ) std::cout << "__CudaRuntime: calling cudaFree(0)" << std::endl;
+      checkCuda( cudaFree( 0 ) ); // SLOW!
+      */
+      // Replace cudaFree(0) by cudaSetDevice(0), even if it is not really needed either
+      // (but see https://developer.nvidia.com/blog/cuda-pro-tip-always-set-current-device-avoid-multithreading-bugs)
+      if( debug ) std::cout << "__GpuRuntime: calling GpuSetDevice(0)" << std::endl;
+      checkGpu( gpuSetDevice( 0 ) ); // SLOW!
+    }
+
+    // Tear down CUDA application (call cudaDeviceReset)
+    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
+    // Calling cudaDeviceReset on shutdown is only needed for checking memory leaks in cuda-memcheck
+    // See https://docs.nvidia.com/cuda/cuda-memcheck/index.html#leak-checking
+    static void tearDown( const bool debug = true )
+    {
+      if( debug ) std::cout << "__GpuRuntime: calling GpuDeviceReset()" << std::endl;
+      checkGpu( gpuDeviceReset() );
+    }
+  };
+}
+#endif
+
+//--------------------------------------------------------------------------
+
+#endif // MG5AMC_GPURUNTIME_H
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MadgraphTest.h b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MadgraphTest.h
index ef40624c88..6054185300 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MadgraphTest.h
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MadgraphTest.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Hageboeck (Dec 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MADGRAPHTEST_H_
 #define MADGRAPHTEST_H_ 1
@@ -14,7 +14,11 @@
 
 #include <array>
 #include <cmath>
-#include <filesystem>
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
 #include <fstream>
 #include <iomanip>
 #include <memory>
@@ -22,7 +26,7 @@
 #include <string>
 #include <vector>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using mg5amcGpu::CPPProcess;
 #else
 using mg5amcCpu::CPPProcess;
@@ -201,7 +205,7 @@ class MadgraphTest : public testing::TestWithParam<TestDriverBase*>
 
 // Since we link both the CPU-only and GPU tests into the same executable, we prevent
 // a multiply defined symbol by only compiling this in the non-CUDA phase:
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
 /// Compare momenta and matrix elements.
 /// This uses an implementation of TestDriverBase to run a madgraph workflow,
@@ -219,7 +223,14 @@ TEST_P( MadgraphTest, CompareMomentaAndME )
   const char* dumpEventsC = getenv( "CUDACPP_RUNTEST_DUMPEVENTS" );
   const bool dumpEvents = ( dumpEventsC != 0 ) && ( std::string( dumpEventsC ) != "" );
   const std::string refFileName = testDriver->getRefFileName();
+  /*
+#ifdef __HIPCC__
+  const std::string dumpFileName = std::experimental::filesystem::path( refFileName ).filename();
+#else
   const std::string dumpFileName = std::filesystem::path( refFileName ).filename();
+#endif
+  */
+  const std::string dumpFileName = refFileName; // bypass std::filesystem #803
   std::ofstream dumpFile;
   if( dumpEvents )
   {
@@ -307,6 +318,6 @@ TEST_P( MadgraphTest, CompareMomentaAndME )
   }
 }
 
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
 
 #endif /* MADGRAPHTEST_H_ */
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MatrixElementKernels.cc b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MatrixElementKernels.cc
index 74b5239ebf..81699dfea9 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MatrixElementKernels.cc
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MatrixElementKernels.cc
@@ -1,12 +1,12 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "MatrixElementKernels.h"
 
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h" // Includes the abstraction for Nvidia/AMD compilation
 #include "MemoryAccessMomenta.h"
 #include "MemoryBuffers.h"
 
@@ -14,7 +14,7 @@
 
 //============================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu
 {
 
@@ -150,7 +150,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
@@ -209,13 +209,13 @@ namespace mg5amcGpu
     PinnedHostBufferHelicityMask hstIsGoodHel( ncomb );
     DeviceBufferHelicityMask devIsGoodHel( ncomb );
     // ... 0d1. Compute good helicity mask on the device
-    computeDependentCouplings<<<m_gpublocks, m_gputhreads>>>( m_gs.data(), m_couplings.data() );
+    gpuLaunchKernel( computeDependentCouplings, m_gpublocks, m_gputhreads, m_gs.data(), m_couplings.data() );
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-    sigmaKin_getGoodHel<<<m_gpublocks, m_gputhreads>>>( m_momenta.data(), m_couplings.data(), m_matrixElements.data(), m_numerators.data(), m_denominators.data(), devIsGoodHel.data() );
+    gpuLaunchKernel( sigmaKin_getGoodHel, m_gpublocks, m_gputhreads, m_momenta.data(), m_couplings.data(), m_matrixElements.data(), m_numerators.data(), m_denominators.data(), devIsGoodHel.data() );
 #else
-    sigmaKin_getGoodHel<<<m_gpublocks, m_gputhreads>>>( m_momenta.data(), m_couplings.data(), m_matrixElements.data(), devIsGoodHel.data() );
+    gpuLaunchKernel( sigmaKin_getGoodHel, m_gpublocks, m_gputhreads, m_momenta.data(), m_couplings.data(), m_matrixElements.data(), devIsGoodHel.data() );
 #endif
-    checkCuda( cudaPeekAtLastError() );
+    checkGpu( gpuPeekAtLastError() );
     // ... 0d2. Copy back good helicity mask to the host
     copyHostFromDevice( hstIsGoodHel, devIsGoodHel );
     // ... 0d3. Copy back good helicity list to constant memory on the device
@@ -226,19 +226,19 @@ namespace mg5amcGpu
 
   void MatrixElementKernelDevice::computeMatrixElements( const unsigned int channelId )
   {
-    computeDependentCouplings<<<m_gpublocks, m_gputhreads>>>( m_gs.data(), m_couplings.data() );
+    gpuLaunchKernel( computeDependentCouplings, m_gpublocks, m_gputhreads, m_gs.data(), m_couplings.data() );
 #ifndef MGONGPU_NSIGHT_DEBUG
     constexpr unsigned int sharedMemSize = 0;
 #else
     constexpr unsigned int sharedMemSize = ntpbMAX * sizeof( float );
 #endif
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-    sigmaKin<<<m_gpublocks, m_gputhreads, sharedMemSize>>>( m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), channelId, m_numerators.data(), m_denominators.data(), m_selhel.data(), m_selcol.data() );
+    gpuLaunchKernelSharedMem( sigmaKin, m_gpublocks, m_gputhreads, sharedMemSize, m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), channelId, m_numerators.data(), m_denominators.data(), m_selhel.data(), m_selcol.data() );
 #else
-    sigmaKin<<<m_gpublocks, m_gputhreads, sharedMemSize>>>( m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), m_selhel.data(), m_selcol.data() );
+    gpuLaunchKernelSharedMem( sigmaKin, m_gpublocks, m_gputhreads, sharedMemSize, m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), m_selhel.data(), m_selcol.data() );
 #endif
-    checkCuda( cudaPeekAtLastError() );
-    checkCuda( cudaDeviceSynchronize() );
+    checkGpu( gpuPeekAtLastError() );
+    checkGpu( gpuDeviceSynchronize() );
   }
 
   //--------------------------------------------------------------------------
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MatrixElementKernels.h b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MatrixElementKernels.h
index 23e84757a2..72bd8f195b 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MatrixElementKernels.h
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MatrixElementKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MATRIXELEMENTKERNELS_H
 #define MATRIXELEMENTKERNELS_H 1
@@ -10,7 +10,7 @@
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -81,7 +81,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating matrix element calculations on a CPU host
   class MatrixElementKernelHost final : public MatrixElementKernelBase, public NumberOfEvents
   {
@@ -130,7 +130,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating matrix element calculations on a GPU device
   class MatrixElementKernelDevice : public MatrixElementKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessAmplitudes.h b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessAmplitudes.h
index 573b3bbbc9..ffb76e93de 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessAmplitudes.h
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessAmplitudes.h
@@ -15,7 +15,7 @@
 #define MGONGPU_TRIVIAL_AMPLITUDES 1
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessCouplings.h b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessCouplings.h
index 35a3af42e0..3afdf3e554 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessCouplings.h
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessCouplings.h
@@ -15,7 +15,7 @@
 #include "MemoryBuffers.h"       // for HostBufferCouplings::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessCouplingsFixed.h b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessCouplingsFixed.h
index dc0d93afff..ffcdf4dbef 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessCouplingsFixed.h
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessCouplingsFixed.h
@@ -14,7 +14,7 @@
 //#include "MemoryAccessHelpers.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessDenominators.h b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessDenominators.h
index 3bce635718..66f2d32a6b 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessDenominators.h
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessDenominators.h
@@ -10,7 +10,7 @@
 #include "MemoryAccessGs.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessGs.h b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessGs.h
index 31311aa375..4c726b30f3 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessGs.h
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessGs.h
@@ -13,7 +13,7 @@
 #include "MemoryBuffers.h" // for HostBufferMatrixElements::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessHelpers.h b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessHelpers.h
index c82a6c7635..db73e4e064 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessHelpers.h
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessHelpers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessHelpers_H
 #define MemoryAccessHelpers_H 1
@@ -105,7 +105,7 @@ class KernelAccessHelper : public MemoryAccessHelper<T>
     }
     else
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
       //printf( "kernelAccessRecord: ievt=%d threadId=%d\n", ievt, threadIdx.x );
       return T::ieventAccessRecord( buffer, ievt ); // NB fptype and fptype_sv coincide for CUDA
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessMatrixElements.h b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessMatrixElements.h
index f32e6fea5b..3741011971 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessMatrixElements.h
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessMatrixElements.h
@@ -13,7 +13,7 @@
 #include "MemoryBuffers.h" // for HostBufferMatrixElements::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessMomenta.h b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessMomenta.h
index 29266de32c..3be229d392 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessMomenta.h
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessMomenta.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessMomenta_H
 #define MemoryAccessMomenta_H 1
@@ -13,7 +13,7 @@
 #include "MemoryAccessVectors.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -30,7 +30,7 @@ namespace mg5amcCpu
 
     // Number of Events Per Page in the momenta AOSOA memory buffer layout
     // (these are all best kept as a compile-time constants: see issue #23)
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
     // -----------------------------------------------------------------------------------------------
     // --- GPUs: neppM is best set to a power of 2 times the number of fptype's in a 32-byte cacheline
     // --- This is relevant to ensure coalesced access to momenta in global memory
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessNumerators.h b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessNumerators.h
index b152183b28..18991f4fa6 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessNumerators.h
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessNumerators.h
@@ -10,7 +10,7 @@
 #include "MemoryAccessGs.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessRandomNumbers.h b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessRandomNumbers.h
index e2988d39f3..40cb089135 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessRandomNumbers.h
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessRandomNumbers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessRandomNumbers_H
 #define MemoryAccessRandomNumbers_H 1
@@ -11,7 +11,7 @@
 #include "CPPProcess.h"
 #include "MemoryAccessHelpers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using mg5amcGpu::CPPProcess;
 #else
 using mg5amcCpu::CPPProcess;
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessVectors.h b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessVectors.h
index e9b197368e..08faccff0f 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessVectors.h
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessVectors.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessVectors_H
 #define MemoryAccessVectors_H 1
@@ -10,7 +10,7 @@
 
 #include "mgOnGpuVectors.h"
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu // this is only needed for CPU SIMD vectorization
 {
 
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessWavefunctions.h b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessWavefunctions.h
index 5428aaf933..33bef4559e 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessWavefunctions.h
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryAccessWavefunctions.h
@@ -15,7 +15,7 @@
 #define MGONGPU_TRIVIAL_WAVEFUNCTIONS 1
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryBuffers.h b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryBuffers.h
index 3093e6ed18..7756a71621 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryBuffers.h
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/MemoryBuffers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021, based on earlier work by S. Hageboeck) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryBuffers_H
 #define MemoryBuffers_H 1
@@ -11,12 +11,12 @@
 #include "mgOnGpuCxtypes.h"
 
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "Parameters_sm.h"
 
 #include <sstream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -87,7 +87,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   constexpr bool HostBufferALIGNED = false;   // ismisaligned=false
   constexpr bool HostBufferMISALIGNED = true; // ismisaligned=true
 
@@ -119,7 +119,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA pinned host buffer
   template<typename T>
   class PinnedHostBufferBase : public BufferBase<T>
@@ -128,18 +128,18 @@ namespace mg5amcCpu
     PinnedHostBufferBase( const size_t size )
       : BufferBase<T>( size, false )
     {
-      checkCuda( cudaMallocHost( &( this->m_data ), this->bytes() ) );
+      gpuMallocHost( &( this->m_data ), this->bytes() );
     }
     virtual ~PinnedHostBufferBase()
     {
-      checkCuda( cudaFreeHost( this->m_data ) );
+      gpuFreeHost( this->m_data );
     }
   };
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA device buffer
   template<typename T>
   class DeviceBufferBase : public BufferBase<T>
@@ -148,18 +148,18 @@ namespace mg5amcCpu
     DeviceBufferBase( const size_t size )
       : BufferBase<T>( size, true )
     {
-      checkCuda( cudaMalloc( &( this->m_data ), this->bytes() ) );
+      gpuMalloc( &( this->m_data ), this->bytes() );
     }
     virtual ~DeviceBufferBase()
     {
-      checkCuda( cudaFree( this->m_data ) );
+      gpuFree( this->m_data );
     }
   };
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for a given number of events
   template<typename T, size_t sizePerEvent, bool ismisaligned>
   class HostBuffer : public HostBufferBase<T, ismisaligned>, virtual private NumberOfEvents
@@ -175,7 +175,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA pinned host buffer for a given number of events
   template<typename T, size_t sizePerEvent>
   class PinnedHostBuffer : public PinnedHostBufferBase<T>, virtual private NumberOfEvents
@@ -191,7 +191,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA device buffer for a given number of events
   template<typename T, size_t sizePerEvent>
   class DeviceBuffer : public DeviceBufferBase<T>, virtual private NumberOfEvents
@@ -213,7 +213,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for momenta random numbers
   constexpr size_t sizePerEventRndNumMomenta = MemoryBuffers::np4 * MemoryBuffers::nparf;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for momenta random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumMomenta, HostBufferALIGNED> HostBufferRndNumMomenta;
 #else
@@ -232,7 +232,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer with ONE fptype per event
   constexpr size_t sizePerEventOneFp = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer with ONE fptype per event
   typedef HostBuffer<fptype, sizePerEventOneFp, HostBufferALIGNED> HostBufferOneFp;
 #else
@@ -257,7 +257,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for Gs
   constexpr size_t sizePerEventGs = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventGs, HostBufferALIGNED> HostBufferGs;
 #else
@@ -276,7 +276,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for numerators
   constexpr size_t sizePerEventNumerators = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventNumerators, HostBufferALIGNED> HostBufferNumerators;
 #else
@@ -296,7 +296,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for denominators
   constexpr size_t sizePerEventDenominators = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventDenominators, HostBufferALIGNED> HostBufferDenominators;
 #else
@@ -315,7 +315,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for random numbers
   constexpr size_t sizePerEventCouplings = MemoryBuffers::ndcoup * MemoryBuffers::nx2;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventCouplings, HostBufferALIGNED> HostBufferCouplings;
 #else
@@ -333,7 +333,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for momenta
   constexpr size_t sizePerEventMomenta = MemoryBuffers::np4 * MemoryBuffers::npar;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for momenta
   typedef HostBuffer<fptype, sizePerEventMomenta, HostBufferALIGNED> HostBufferMomenta;
   //typedef HostBuffer<fptype, sizePerEventMomenta, HostBufferMISALIGNED> HostBufferMomenta; // TEST MISALIGNMENT!
@@ -352,7 +352,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for sampling weights
   constexpr size_t sizePerEventWeights = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for sampling weights
   typedef HostBuffer<fptype, sizePerEventWeights, HostBufferALIGNED> HostBufferWeights;
 #else
@@ -370,7 +370,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for matrix elements
   constexpr size_t sizePerEventMatrixElements = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for matrix elements
   typedef HostBuffer<fptype, sizePerEventMatrixElements, HostBufferALIGNED> HostBufferMatrixElements;
 #else
@@ -385,7 +385,7 @@ namespace mg5amcCpu
   // A base class encapsulating a memory buffer for the helicity mask
   typedef BufferBase<bool> BufferHelicityMask;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for the helicity mask
   typedef HostBufferBase<bool, HostBufferALIGNED> HostBufferHelicityMask;
 #else
@@ -403,7 +403,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for wavefunctions
   constexpr size_t sizePerEventWavefunctions = MemoryBuffers::nw6 * MemoryBuffers::nx2;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for wavefunctions
   typedef HostBuffer<fptype, sizePerEventWavefunctions, HostBufferALIGNED> HostBufferWavefunctions;
 #else
@@ -421,7 +421,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for helicity random numbers
   constexpr size_t sizePerEventRndNumHelicity = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for helicity random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumHelicity, HostBufferALIGNED> HostBufferRndNumHelicity;
 #else
@@ -439,7 +439,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for color random numbers
   constexpr size_t sizePerEventRndNumColor = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for color random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumColor, HostBufferALIGNED> HostBufferRndNumColor;
 #else
@@ -457,7 +457,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for helicity selection
   constexpr size_t sizePerEventSelectedHelicity = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for helicity selection
   typedef HostBuffer<int, sizePerEventSelectedHelicity, HostBufferALIGNED> HostBufferSelectedHelicity;
 #else
@@ -475,7 +475,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for color selection
   constexpr size_t sizePerEventSelectedColor = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for color selection
   typedef HostBuffer<int, sizePerEventSelectedColor, HostBufferALIGNED> HostBufferSelectedColor;
 #else
@@ -487,7 +487,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<class Tdst, class Tsrc>
   void copyDeviceFromHost( Tdst& dst, const Tsrc& src ) // keep the same order of arguments as in memcpy
   {
@@ -504,13 +504,13 @@ namespace mg5amcCpu
       throw std::runtime_error( sstr.str() );
     }
     // NB (PR #45): cudaMemcpy involves an intermediate memcpy to pinned memory if host array is a not a pinned host array
-    checkCuda( cudaMemcpy( dst.data(), src.data(), src.bytes(), cudaMemcpyHostToDevice ) );
+    gpuMemcpy( dst.data(), src.data(), src.bytes(), gpuMemcpyHostToDevice );
   }
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<class Tdst, class Tsrc>
   void copyHostFromDevice( Tdst& dst, const Tsrc& src ) // keep the same order of arguments as in memcpy
   {
@@ -527,7 +527,7 @@ namespace mg5amcCpu
       throw std::runtime_error( sstr.str() );
     }
     // NB (PR #45): cudaMemcpy involves an intermediate memcpy to pinned memory if host array is a not a pinned host array
-    checkCuda( cudaMemcpy( dst.data(), src.data(), src.bytes(), cudaMemcpyDeviceToHost ) );
+    gpuMemcpy( dst.data(), src.data(), src.bytes(), gpuMemcpyDeviceToHost );
   }
 #endif
 
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/CPPProcess.cc b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/CPPProcess.cc
index 19e6cd201c..a041636caf 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/CPPProcess.cc
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_sm.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -46,7 +45,7 @@
 // Class member functions for calculating the matrix elements for
 // Process: g g > t t~ g g g WEIGHTED<=5 @1
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -80,7 +79,7 @@ namespace mg5amcCpu
   __device__ const fptype cIPD[2] = { (fptype)Parameters_sm::mdl_MT, (fptype)Parameters_sm::mdl_WT };
   __device__ const fptype* cIPC = nullptr; // unused as nicoup=0
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ fptype cIPD[2];
   __device__ __constant__ fptype* cIPC = nullptr; // unused as nicoup=0
 #else
@@ -90,7 +89,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -118,13 +117,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -151,7 +150,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -187,7 +186,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -200,8 +199,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -30018,7 +30019,7 @@ namespace mg5amcCpu
         { -116, 442, 442, -134, -134, 505, -44, -134, 28, -224, -62, 496, -53, 19, -62, 496, 10, -80, -62, -71, 10, -80, 1, -8, 136, -116, -116, -44, -44, 514, -116, 442, -44, 28, -53, -62, -44, -53, 514, -62, 100, 10, 28, -62, -62, 10, 10, 1, 514, 505, -62, 496, -71, 568, -44, -134, -53, -62, 19, -71, 10, -80, 100, 10, 640, -80, 1, -8, 10, 1, 64, -8, -62, -71, 10, -80, 1, -8, 28, -62, -62, 10, 10, 1, 1, -8, 10, 1, 64, -8, -8, 64, -80, -8, -512, 64, 496, 568, -80, 640, -8, 64, -224, 496, 496, -80, -80, -8, -8, 64, -80, -8, -512, 64, 64, -512, 640, 64, 4096, -512 },
         { 136, -116, -116, -44, -44, 514, -116, 442, -44, 28, -53, -62, -44, -53, 514, -62, 100, 10, 28, -62, -62, 10, 10, 1, -116, 442, 442, -134, -134, 505, -44, -134, 28, -224, -62, 496, -53, 19, -62, 496, 10, -80, -62, -71, 10, -80, 1, -8, -44, -134, -53, -62, 19, -71, 514, 505, -62, 496, -71, 568, 100, 10, 10, -80, -80, 640, 10, 1, 1, -8, -8, 64, 28, -62, -62, 10, 10, 1, -62, -71, 10, -80, 1, -8, 10, 1, 1, -8, -8, 64, -80, -8, -8, 64, 64, -512, -224, 496, 496, -80, -80, -8, 496, 568, -80, 640, -8, 64, -80, -8, -8, 64, 64, -512, 640, 64, 64, -512, -512, 4096 } }; // 2-D array[120][120]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -30075,7 +30076,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -30134,7 +30135,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -30293,8 +30294,8 @@ namespace mg5amcCpu
       { 1, 1, 1, -1, 1, -1, 1 },
       { 1, 1, 1, -1, 1, 1, -1 },
       { 1, 1, 1, -1, 1, 1, 1 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -30337,9 +30338,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     const fptype tIPD[2] = { (fptype)m_pars->mdl_MT, (fptype)m_pars->mdl_WT };
     //const cxtype tIPC[0] = { ... }; // nicoup=0
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) ) );
-    //checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ) ); // nicoup=0
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) );
+    //gpuMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
 #else
     memcpy( cIPD, tIPD, 2 * sizeof( fptype ) );
     //memcpy( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
@@ -30378,7 +30379,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -30407,6 +30408,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -30420,6 +30425,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -30443,12 +30450,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -30469,7 +30476,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -30595,9 +30602,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -30621,7 +30628,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -30641,7 +30648,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 1536 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -30655,9 +30662,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -30685,7 +30695,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -30895,7 +30905,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/CPPProcess.h b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/CPPProcess.h
index 2565923dde..fff95b66e2 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/CPPProcess.h
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -107,7 +107,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -120,7 +120,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -150,7 +150,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/CudaRuntime.h b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/GpuAbstraction.h b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/GpuRuntime.h b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/check_sa.cc b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/check_sa.cc
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/RamboSamplingKernels.cc b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/RamboSamplingKernels.cc
index da68aa9255..79abbcc4f8 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/RamboSamplingKernels.cc
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/RamboSamplingKernels.cc
@@ -1,11 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "RamboSamplingKernels.h"
 
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "MemoryAccessMomenta.h"
 #include "MemoryAccessRandomNumbers.h"
 #include "MemoryAccessWeights.h"
@@ -14,7 +14,7 @@
 
 #include <sstream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -92,7 +92,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingKernelDevice::RamboSamplingKernelDevice( const fptype energy,               // input: energy
                                                         const BufferRndNumMomenta& rndmom, // input: random numbers in [0,1]
                                                         BufferMomenta& momenta,            // output: momenta
@@ -135,7 +135,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   getMomentaInitialDevice( const fptype energy,
                            fptype* momenta )
@@ -147,17 +147,17 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   void
   RamboSamplingKernelDevice::getMomentaInitial()
   {
-    getMomentaInitialDevice<<<m_gpublocks, m_gputhreads>>>( m_energy, m_momenta.data() );
+    gpuLaunchKernel( getMomentaInitialDevice, m_gpublocks, m_gputhreads, m_energy, m_momenta.data() );
   }
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   getMomentaFinalDevice( const fptype energy,
                          const fptype* rndmom,
@@ -171,11 +171,11 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   void
   RamboSamplingKernelDevice::getMomentaFinal()
   {
-    getMomentaFinalDevice<<<m_gpublocks, m_gputhreads>>>( m_energy, m_rndmom.data(), m_momenta.data(), m_weights.data() );
+    gpuLaunchKernel( getMomentaFinalDevice, m_gpublocks, m_gputhreads, m_energy, m_rndmom.data(), m_momenta.data(), m_weights.data() );
   }
 #endif
 
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/RamboSamplingKernels.h b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/RamboSamplingKernels.h
index 184089efd7..7c214cd74b 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/RamboSamplingKernels.h
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/RamboSamplingKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef RAMBOSAMPLINGKERNELS_H
 #define RAMBOSAMPLINGKERNELS_H 1
@@ -10,7 +10,7 @@
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -93,7 +93,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating RAMBO phase space sampling on a GPU device
   class RamboSamplingKernelDevice final : public SamplingKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/RandomNumberKernels.h b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/RandomNumberKernels.h
index 188a72c2c9..21d63beeac 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/RandomNumberKernels.h
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/RandomNumberKernels.h
@@ -1,14 +1,14 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef RANDOMNUMBERKERNELS_H
 #define RANDOMNUMBERKERNELS_H 1
 
 #include "mgOnGpuConfig.h"
 
-// NB This must come AFTER mgOnGpuConfig.h which contains our definition of __global__ when __CUDACC__ is not defined
+// NB This must come AFTER mgOnGpuConfig.h which contains our definition of __global__ when MGONGPUCPP_GPUIMPL is not defined
 #ifndef MGONGPU_HAS_NO_CURAND
 //#include "curand.h"
 struct curandGenerator_st; // forward definition from curand.h
@@ -16,7 +16,7 @@ struct curandGenerator_st; // forward definition from curand.h
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/cudacpp.mk b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/cudacpp.mk
index 509307506b..df74dfc284 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/cudacpp.mk
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/cudacpp.mk
@@ -1,7 +1,7 @@
 # Copyright (C) 2020-2023 CERN and UCLouvain.
 # Licensed under the GNU Lesser General Public License (version 3 or later).
 # Created by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-# Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+# Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #=== Determine the name of this makefile (https://ftp.gnu.org/old-gnu/Manuals/make-3.80/html_node/make_17.html)
 #=== NB: use ':=' to ensure that the value of CUDACPP_MAKEFILE is not modified further down after including make_opts
@@ -42,10 +42,10 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure common compiler flags for C++ and CUDA
+#=== Configure common compiler flags for C++ and CUDA/HIP
 
 INCFLAGS = -I.
-OPTFLAGS = -O3 # this ends up in CUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
+OPTFLAGS = -O3 # this ends up in GPUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
 
 # Dependency on src directory
 MG5AMC_COMMONLIB = mg5amc_common
@@ -121,24 +121,45 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure the CUDA compiler
+#=== Configure the GPU compiler (CUDA or HIP)
+
+# FIXME! (AV 24.01.2024)
+# In the current implementation (without separate builds for C++ and CUDA/HIP), we first check for cudacc and hipcc in CUDA_HOME and HIP_HOME.
+# If CUDA_HOME or HIP_HOME are not set, try to determine them from the path to cudacc and hipcc.
+# While convoluted, this is currently necessary to allow disabling CUDA/HIP builds by setting CUDA_HOME or HIP_HOME to invalid paths.
+# This will (probably?) be fixed when separate C++ and CUDA/HIP builds are implemented (PR #775).
 
 # If CXX is not a single word (example "clang++ --gcc-toolchain...") then disable CUDA builds (issue #505)
-# This is because it is impossible to pass this to "CUFLAGS += -ccbin <host-compiler>" below
+# This is because it is impossible to pass this to "GPUFLAGS += -ccbin <host-compiler>" below
 ifneq ($(words $(subst ccache ,,$(CXX))),1) # allow at most "CXX=ccache <host-compiler>" from outside
   $(warning CUDA builds are not supported for multi-word CXX "$(CXX)")
   override CUDA_HOME=disabled
 endif
 
-# If CUDA_HOME is not set, try to set it from the location of nvcc
+# If CUDA_HOME is not set, try to set it from the path to nvcc
 ifndef CUDA_HOME
-  CUDA_HOME = $(patsubst %bin/nvcc,%,$(shell which nvcc 2>/dev/null))
+  CUDA_HOME = $(patsubst %/bin/nvcc,%,$(shell which nvcc 2>/dev/null))
   $(warning CUDA_HOME was not set: using "$(CUDA_HOME)")
 endif
 
-# Set NVCC as $(CUDA_HOME)/bin/nvcc if it exists
+# If HIP_HOME is not set, try to set it from the path to hipcc
+ifndef HIP_HOME
+  HIP_HOME = $(patsubst %/bin/hipcc,%,$(shell which hipcc 2>/dev/null))
+  $(warning HIP_HOME was not set: using "$(HIP_HOME)")
+endif
+
+# FIXME! (AV 24.01.2024)
+# In the current implementation (without separate builds for C++ and CUDA/HIP),
+# builds are performed for HIP only if CUDA is not found in the path.
+# If both CUDA and HIP are installed, HIP builds can be triggered by unsetting CUDA_HOME.
+# This will be fixed when separate C++ and CUDA/HIP builds are implemented (PR #775).
+
+#--- Option 1: CUDA exists -> use CUDA
+
+# Set GPUCC as $(CUDA_HOME)/bin/nvcc if it exists
 ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
-  NVCC = $(CUDA_HOME)/bin/nvcc
+
+  GPUCC = $(CUDA_HOME)/bin/nvcc
   USE_NVTX ?=-DUSE_NVTX
   # See https://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/index.html
   # See https://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/
@@ -148,7 +169,7 @@ ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
   # Examples: use 60 for P100 (Piz Daint), 80 for A100 (Juwels Booster, NVidia raplab/Curiosity).
   MADGRAPH_CUDA_ARCHITECTURE ?= 70
   ###CUARCHFLAGS = -gencode arch=compute_$(MADGRAPH_CUDA_ARCHITECTURE),code=compute_$(MADGRAPH_CUDA_ARCHITECTURE) -gencode arch=compute_$(MADGRAPH_CUDA_ARCHITECTURE),code=sm_$(MADGRAPH_CUDA_ARCHITECTURE) # Older implementation (AV): go back to this one for multi-GPU support #533
-  ###CUARCHFLAGS = --gpu-architecture=compute_$(MADGRAPH_CUDA_ARCHITECTURE) --gpu-code=sm_$(MADGRAPH_CUDA_ARCHITECTURE),compute_$(MADGRAPH_CUDA_ARCHITECTURE) # Newer implementation (SH): cannot use this as-is for multi-GPU support #533
+  ###CUARCHFLAGS = --gpu-architecture=compute_$(MADGRAPH_CUDA_ARCHITECTURE) --gpu-code=sm_$(MADGRAPH_CUDA_ARCHITECTURE),compute_$(MADGRAPH_CUDA_ARCHITECTURE)  # Newer implementation (SH): cannot use this as-is for multi-GPU support #533
   comma:=,
   CUARCHFLAGS = $(foreach arch,$(subst $(comma), ,$(MADGRAPH_CUDA_ARCHITECTURE)),-gencode arch=compute_$(arch),code=compute_$(arch) -gencode arch=compute_$(arch),code=sm_$(arch))
   CUINC = -I$(CUDA_HOME)/include/
@@ -158,41 +179,79 @@ ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
     CURANDLIBFLAGS = -L$(CUDA_HOME)/lib64/ -lcurand # NB: -lcuda is not needed here!
   endif
   CUOPTFLAGS = -lineinfo
-  CUFLAGS = $(foreach opt, $(OPTFLAGS), -Xcompiler $(opt)) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
-  ###CUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
-  ###NVCC_VERSION = $(shell $(NVCC) --version | grep 'Cuda compilation tools' | cut -d' ' -f5 | cut -d, -f1)
-  CUFLAGS += -std=c++17 # need CUDA >= 11.2 (see #333): this is enforced in mgOnGpuConfig.h
+  ###GPUFLAGS = $(OPTFLAGS) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
+  GPUFLAGS = $(foreach opt, $(OPTFLAGS), -Xcompiler $(opt)) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
+  ###GPUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
+  ###GPUCC_VERSION = $(shell $(GPUCC) --version | grep 'Cuda compilation tools' | cut -d' ' -f5 | cut -d, -f1)
+  GPUFLAGS += -std=c++17 # need CUDA >= 11.2 (see #333): this is enforced in mgOnGpuConfig.h
   # Without -maxrregcount: baseline throughput: 6.5E8 (16384 32 12) up to 7.3E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 160 # improves throughput: 6.9E8 (16384 32 12) up to 7.7E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 128 # improves throughput: 7.3E8 (16384 32 12) up to 7.6E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 96 # degrades throughput: 4.1E8 (16384 32 12) up to 4.5E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 64 # degrades throughput: 1.7E8 (16384 32 12) flat at 1.7E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 160 # improves throughput: 6.9E8 (16384 32 12) up to 7.7E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 128 # improves throughput: 7.3E8 (16384 32 12) up to 7.6E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 96 # degrades throughput: 4.1E8 (16384 32 12) up to 4.5E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 64 # degrades throughput: 1.7E8 (16384 32 12) flat at 1.7E8 (65536 128 12)
+  CUBUILDRULEFLAGS = -Xcompiler -fPIC -c
+  CCBUILDRULEFLAGS = -Xcompiler -fPIC -c -x cu
+  CUDATESTFLAGS = -lcuda
+
+  # Set the host C++ compiler for GPUCC via "-ccbin <host-compiler>"
+  # (NB issue #505: this must be a single word, "clang++ --gcc-toolchain..." is not supported)
+  GPUFLAGS += -ccbin $(shell which $(subst ccache ,,$(CXX)))
+
+  # Allow newer (unsupported) C++ compilers with older versions of CUDA if ALLOW_UNSUPPORTED_COMPILER_IN_CUDA is set (#504)
+  ifneq ($(origin ALLOW_UNSUPPORTED_COMPILER_IN_CUDA),undefined)
+  GPUFLAGS += -allow-unsupported-compiler
+  endif
+
 else ifneq ($(origin REQUIRE_CUDA),undefined)
+
   # If REQUIRE_CUDA is set but no cuda is found, stop here (e.g. for CI tests on GPU #443)
-  $(error No cuda installation found (set CUDA_HOME or make nvcc visible in PATH))
+  $(error No cuda installation found (set CUDA_HOME or make GPUCC visible in PATH))
+
+#--- Option 2: CUDA does not exist, HIP exists -> use HIP
+
+# Set GPUCC as $(HIP_HOME)/bin/hipcc if it exists
+else ifneq ($(wildcard $(HIP_HOME)/bin/hipcc),)
+
+  GPUCC = $(HIP_HOME)/bin/hipcc
+  #USE_NVTX ?=-DUSE_NVTX # should maybe find something equivalent to this in HIP?
+  HIPARCHFLAGS = -target x86_64-linux-gnu --offload-arch=gfx90a
+  HIPINC = -I$(HIP_HOME)/include/
+  # Note: -DHIP_FAST_MATH is equivalent to -use_fast_math in HIP 
+  # (but only for single precision line 208: https://rocm-developer-tools.github.io/HIP/hcc__detail_2math__functions_8h_source.html)
+  # Note: CUOPTFLAGS should not be used for HIP, it had been added here but was then removed (#808)
+  GPUFLAGS = $(OPTFLAGS) $(INCFLAGS) $(HIPINC) $(HIPARCHFLAGS) -DHIP_FAST_MATH -DHIP_PLATFORM=amd -fPIC
+  ###GPUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
+  GPUFLAGS += -std=c++17
+  ###GPUFLAGS+= --maxrregcount 255 # (AV: is this option valid on HIP and meaningful on AMD GPUs?)
+  CUBUILDRULEFLAGS = -fPIC -c
+  CCBUILDRULEFLAGS = -fPIC -c -x hip
+
+else ifneq ($(origin REQUIRE_HIP),undefined)
+
+  # If REQUIRE_HIP is set but no HIP is found, stop here (e.g. for CI tests on GPU #443)
+  $(error No hip installation found (set HIP_HOME or make GPUCC visible in PATH))
+
+#--- Option 3: CUDA does not exist, HIP does not exist -> switch off both CUDA and HIP
+
 else
-  # No cuda. Switch cuda compilation off and go to common random numbers in C++
+
+  # No cudacc and no hipcc: switch CUDA and HIP compilation off and go to common random numbers in C++
   $(warning CUDA_HOME is not set or is invalid: export CUDA_HOME to compile with cuda)
-  override NVCC=
+  $(warning HIP_HOME is not set or is invalid: export HIP_HOME to compile with hip)
+  override GPUCC=
   override USE_NVTX=
   override CUINC=
   override CURANDLIBFLAGS=
-endif
-export NVCC
-export CUFLAGS
 
-# Set the host C++ compiler for nvcc via "-ccbin <host-compiler>"
-# (NB issue #505: this must be a single word, "clang++ --gcc-toolchain..." is not supported)
-CUFLAGS += -ccbin $(shell which $(subst ccache ,,$(CXX)))
-
-# Allow newer (unsupported) C++ compilers with older versions of CUDA if ALLOW_UNSUPPORTED_COMPILER_IN_CUDA is set (#504)
-ifneq ($(origin ALLOW_UNSUPPORTED_COMPILER_IN_CUDA),undefined)
-CUFLAGS += -allow-unsupported-compiler
 endif
 
+# Export GPUCC (so that it can also be used in cudacpp_src.mk?)
+export GPUCC
+export GPUFLAGS
+
 #-------------------------------------------------------------------------------
 
-#=== Configure ccache for C++ and CUDA builds
+#=== Configure ccache for C++ and CUDA/HIP builds
 
 # Enable ccache if USECCACHE=1
 ifeq ($(USECCACHE)$(shell echo $(CXX) | grep ccache),1)
@@ -201,15 +260,15 @@ endif
 #ifeq ($(USECCACHE)$(shell echo $(AR) | grep ccache),1)
 #  override AR:=ccache $(AR)
 #endif
-ifneq ($(NVCC),)
-  ifeq ($(USECCACHE)$(shell echo $(NVCC) | grep ccache),1)
-    override NVCC:=ccache $(NVCC)
+ifneq ($(GPUCC),)
+  ifeq ($(USECCACHE)$(shell echo $(GPUCC) | grep ccache),1)
+    override GPUCC:=ccache $(GPUCC)
   endif
 endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure PowerPC-specific compiler flags for C++ and CUDA
+#=== Configure PowerPC-specific compiler flags for C++ and CUDA/HIP
 
 # PowerPC-specific CXX compiler flags (being reviewed)
 ifeq ($(UNAME_P),ppc64le)
@@ -225,9 +284,9 @@ else
   ######CXXFLAGS+= -fno-semantic-interposition # no benefit (neither alone, nor combined with -flto)
 endif
 
-# PowerPC-specific CUDA compiler flags (to be reviewed!)
+# PowerPC-specific CUDA/HIP compiler flags (to be reviewed!)
 ifeq ($(UNAME_P),ppc64le)
-  CUFLAGS+= -Xcompiler -mno-float128
+  GPUFLAGS+= -Xcompiler -mno-float128
 endif
 
 #-------------------------------------------------------------------------------
@@ -235,9 +294,11 @@ endif
 #=== Configure defaults and check if user-defined choices exist for OMPFLAGS, AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
 
 # Set the default OMPFLAGS choice
-ifneq ($(shell $(CXX) --version | egrep '^Intel'),)
+ifneq ($(findstring hipcc,$(GPUCC)),)
+override OMPFLAGS = # disable OpenMP MT when using hipcc #802
+else ifneq ($(shell $(CXX) --version | egrep '^Intel'),)
 override OMPFLAGS = -fopenmp
-###override OMPFLAGS = # disable OpenMP MT on Intel (was ok without nvcc but not ok with nvcc before #578)
+###override OMPFLAGS = # disable OpenMP MT on Intel (was ok without GPUCC but not ok with GPUCC before #578)
 else ifneq ($(shell $(CXX) --version | egrep '^(clang)'),)
 override OMPFLAGS = -fopenmp
 ###override OMPFLAGS = # disable OpenMP MT on clang (was not ok without or with nvcc before #578)
@@ -293,7 +354,10 @@ endif
 
 # Set the default RNDGEN (random number generator) choice
 ifeq ($(RNDGEN),)
-  ifeq ($(NVCC),)
+  ifeq ($(GPUCC),)
+    override RNDGEN = hasNoCurand
+  # Edgecase for HIP compilation
+  else ifeq ($(findstring hipcc,$(GPUCC)),hipcc)
     override RNDGEN = hasNoCurand
   else ifeq ($(RNDGEN),)
     override RNDGEN = hasCurand
@@ -310,7 +374,7 @@ export OMPFLAGS
 
 #-------------------------------------------------------------------------------
 
-#=== Set the CUDA/C++ compiler flags appropriate to user-defined choices of AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
+#=== Set the CUDA/HIP/C++ compiler flags appropriate to user-defined choices of AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
 
 # Set the build flags appropriate to OMPFLAGS
 $(info OMPFLAGS=$(OMPFLAGS))
@@ -368,13 +432,13 @@ CXXFLAGS+= $(AVXFLAGS)
 $(info FPTYPE=$(FPTYPE))
 ifeq ($(FPTYPE),d)
   CXXFLAGS += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
-  CUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
+  GPUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
 else ifeq ($(FPTYPE),f)
   CXXFLAGS += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
-  CUFLAGS  += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
+  GPUFLAGS  += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
 else ifeq ($(FPTYPE),m)
   CXXFLAGS += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
-  CUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
+  GPUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
 else
   $(error Unknown FPTYPE='$(FPTYPE)': only 'd', 'f' and 'm' are supported)
 endif
@@ -383,7 +447,7 @@ endif
 $(info HELINL=$(HELINL))
 ifeq ($(HELINL),1)
   CXXFLAGS += -DMGONGPU_INLINE_HELAMPS
-  CUFLAGS  += -DMGONGPU_INLINE_HELAMPS
+  GPUFLAGS  += -DMGONGPU_INLINE_HELAMPS
 else ifneq ($(HELINL),0)
   $(error Unknown HELINL='$(HELINL)': only '0' and '1' are supported)
 endif
@@ -392,7 +456,7 @@ endif
 $(info HRDCOD=$(HRDCOD))
 ifeq ($(HRDCOD),1)
   CXXFLAGS += -DMGONGPU_HARDCODE_PARAM
-  CUFLAGS  += -DMGONGPU_HARDCODE_PARAM
+  GPUFLAGS  += -DMGONGPU_HARDCODE_PARAM
 else ifneq ($(HRDCOD),0)
   $(error Unknown HRDCOD='$(HRDCOD)': only '0' and '1' are supported)
 endif
@@ -444,11 +508,11 @@ ifeq ($(UNAME_S),Darwin)
   override CULIBFLAGSRPATH2 =
 else
   # RPATH to cuda/cpp libs when linking executables
-  override CXXLIBFLAGSRPATH = -Wl,-rpath,$(LIBDIRRPATH)
-  override CULIBFLAGSRPATH = -Xlinker -rpath,$(LIBDIRRPATH)
+  override CXXLIBFLAGSRPATH = -Wl,-rpath=$(LIBDIRRPATH)
+  override CULIBFLAGSRPATH = -Xlinker -rpath=$(LIBDIRRPATH)
   # RPATH to common lib when linking cuda/cpp libs
-  override CXXLIBFLAGSRPATH2 = -Wl,-rpath,'$$ORIGIN'
-  override CULIBFLAGSRPATH2 = -Xlinker -rpath,'$$ORIGIN'
+  override CXXLIBFLAGSRPATH2 = -Wl,-rpath='$$ORIGIN'
+  override CULIBFLAGSRPATH2 = -Xlinker -rpath='$$ORIGIN'
 endif
 
 # Setting LD_LIBRARY_PATH or DYLD_LIBRARY_PATH in the RUNTIME is no longer necessary (neither on Linux nor on Mac)
@@ -461,7 +525,7 @@ override RUNTIME =
 cxx_main=$(BUILDDIR)/check.exe
 fcxx_main=$(BUILDDIR)/fcheck.exe
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 cu_main=$(BUILDDIR)/gcheck.exe
 fcu_main=$(BUILDDIR)/fgcheck.exe
 else
@@ -492,14 +556,15 @@ $(BUILDDIR)/.build.$(TAG):
 	@touch $(BUILDDIR)/.build.$(TAG)
 
 # Generic target and build rules: objects from CUDA compilation
-ifneq ($(NVCC),)
+# NB: CCBUILDRULEFLAGS includes "-x cu" for nvcc and "-x hip" for hipcc (#810)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/%.o : %.cu *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $(CUBUILDRULEFLAGS) $< -o $@
 
 $(BUILDDIR)/%_cu.o : %.cc *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c -x cu $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $(CCBUILDRULEFLAGS) $< -o $@
 endif
 
 # Generic target and build rules: objects from C++ compilation
@@ -509,11 +574,14 @@ $(BUILDDIR)/%.o : %.cc *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	$(CXX) $(CPPFLAGS) $(CXXFLAGS) -fPIC -c $< -o $@
 
 # Apply special build flags only to CrossSectionKernel.cc and gCrossSectionKernel.cu (no fast math, see #117 and #516)
+# Added edgecase for HIP compilation
 ifeq ($(shell $(CXX) --version | grep ^nvc++),)
 $(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS := $(filter-out -ffast-math,$(CXXFLAGS))
 $(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS += -fno-fast-math
-ifneq ($(NVCC),)
-$(BUILDDIR)/gCrossSectionKernels.o: CUFLAGS += -Xcompiler -fno-fast-math
+ifeq ($(findstring nvcc,$(GPUCC)),nvcc)
+  $(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -Xcompiler -fno-fast-math
+else
+  $(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -fno-fast-math
 endif
 endif
 
@@ -530,10 +598,10 @@ ifeq ($(RNDGEN),hasCurand)
 $(BUILDDIR)/CurandRandomNumberKernel.o: CXXFLAGS += $(CUINC)
 endif
 
-# Avoid "warning: builtin __has_trivial_... is deprecated; use __is_trivially_... instead" in nvcc with icx2023 (#592)
+# Avoid "warning: builtin __has_trivial_... is deprecated; use __is_trivially_... instead" in GPUCC with icx2023 (#592)
 ifneq ($(shell $(CXX) --version | egrep '^(Intel)'),)
-ifneq ($(NVCC),)
-CUFLAGS += -Xcompiler -Wno-deprecated-builtins
+ifneq ($(GPUCC),)
+GPUFLAGS += -Wno-deprecated-builtins
 endif
 endif
 
@@ -541,8 +609,8 @@ endif
 # This patch does remove the warning, but I prefer to keep it disabled for the moment...
 ###ifneq ($(shell $(CXX) --version | egrep '^(clang|Apple clang|Intel)'),)
 ###$(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS += -Wno-overriding-t-option
-###ifneq ($(NVCC),)
-###$(BUILDDIR)/gCrossSectionKernels.o: CUFLAGS += -Xcompiler -Wno-overriding-t-option
+###ifneq ($(GPUCC),)
+###$(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -Xcompiler -Wno-overriding-t-option
 ###endif
 ###endif
 
@@ -569,7 +637,7 @@ MG5AMC_CXXLIB = mg5amc_$(processid_short)_cpp
 cxx_objects_lib=$(BUILDDIR)/CPPProcess.o $(BUILDDIR)/MatrixElementKernels.o $(BUILDDIR)/BridgeKernels.o $(BUILDDIR)/CrossSectionKernels.o
 cxx_objects_exe=$(BUILDDIR)/CommonRandomNumberKernel.o $(BUILDDIR)/RamboSamplingKernels.o
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 MG5AMC_CULIB = mg5amc_$(processid_short)_cuda
 cu_objects_lib=$(BUILDDIR)/gCPPProcess.o $(BUILDDIR)/gMatrixElementKernels.o $(BUILDDIR)/gBridgeKernels.o $(BUILDDIR)/gCrossSectionKernels.o
 cu_objects_exe=$(BUILDDIR)/gCommonRandomNumberKernel.o $(BUILDDIR)/gRamboSamplingKernels.o
@@ -581,11 +649,17 @@ $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: cxx_objects_lib += $(BUILDDIR)/fbridge.o
 $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib)
 	$(CXX) -shared -o $@ $(cxx_objects_lib) $(CXXLIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: $(BUILDDIR)/fbridge_cu.o
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: cu_objects_lib += $(BUILDDIR)/fbridge_cu.o
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cu_objects_lib)
-	$(NVCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+# Bypass std::filesystem completely to ease portability on LUMI #803
+#ifneq ($(findstring hipcc,$(GPUCC)),)
+#	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB) -lstdc++fs
+#else
+#	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+#endif
 endif
 
 #-------------------------------------------------------------------------------
@@ -602,16 +676,16 @@ $(cxx_main): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PAT
 $(cxx_main): $(BUILDDIR)/check_sa.o $(LIBDIR)/lib$(MG5AMC_CXXLIB).so $(cxx_objects_exe) $(BUILDDIR)/CurandRandomNumberKernel.o
 	$(CXX) -o $@ $(BUILDDIR)/check_sa.o $(OMPFLAGS) -ldl -pthread $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe) $(BUILDDIR)/CurandRandomNumberKernel.o $(CURANDLIBFLAGS)
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(cu_main): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(cu_main): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(cu_main): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(cu_main): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 else ifneq ($(shell $(CXX) --version | grep ^nvc++),) # support nvc++ #531
 $(cu_main): LIBFLAGS += -L$(patsubst %bin/nvc++,%lib,$(subst ccache ,,$(CXX))) -lnvhpcatm -lnvcpumath -lnvc
 endif
 $(cu_main): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(cu_main): $(BUILDDIR)/gcheck_sa.o $(LIBDIR)/lib$(MG5AMC_CULIB).so $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o
-	$(NVCC) -o $@ $(BUILDDIR)/gcheck_sa.o $(CUARCHFLAGS) $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o $(CURANDLIBFLAGS)
+	$(GPUCC) -o $@ $(BUILDDIR)/gcheck_sa.o $(CUARCHFLAGS) $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o $(CURANDLIBFLAGS)
 endif
 
 #-------------------------------------------------------------------------------
@@ -635,19 +709,27 @@ $(fcxx_main): LIBFLAGS += -L$(shell dirname $(shell $(FC) --print-file-name libg
 endif
 $(fcxx_main): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(fcxx_main): $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler.o $(LIBDIR)/lib$(MG5AMC_CXXLIB).so $(cxx_objects_exe)
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(BUILDDIR)/fcheck_sa.o $(OMPFLAGS) $(BUILDDIR)/fsampler.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe) -lstdc++
+else
 	$(CXX) -o $@ $(BUILDDIR)/fcheck_sa.o $(OMPFLAGS) $(BUILDDIR)/fsampler.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe)
+endif
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(fcu_main): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(fcu_main): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(fcu_main): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(fcu_main): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 endif
 ifeq ($(UNAME_S),Darwin)
 $(fcu_main): LIBFLAGS += -L$(shell dirname $(shell $(FC) --print-file-name libgfortran.dylib)) # add path to libgfortran on Mac #375
 endif
 $(fcu_main): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(fcu_main): $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBDIR)/lib$(MG5AMC_CULIB).so $(cu_objects_exe)
-	$(NVCC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe)
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) -lstdc++ -L$(shell dirname $(shell $(GPUCC) -print-prog-name=clang))/../../lib -lamdhip64
+else
+	$(GPUCC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe)
+endif
 endif
 
 #-------------------------------------------------------------------------------
@@ -659,7 +741,7 @@ $(BUILDDIR)/testxxx.o: testxxx_cc_ref.txt
 $(testmain): $(BUILDDIR)/testxxx.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/testxxx.o # Comment out this line to skip the C++ test of xxx functions
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/testxxx_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/testxxx_cu.o: INCFLAGS += $(GTESTINC)
 $(BUILDDIR)/testxxx_cu.o: testxxx_cc_ref.txt
@@ -672,7 +754,7 @@ $(BUILDDIR)/testmisc.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/testmisc.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/testmisc.o # Comment out this line to skip the C++ miscellaneous tests
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/testmisc_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/testmisc_cu.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/testmisc_cu.o
@@ -684,12 +766,12 @@ $(BUILDDIR)/runTest.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/runTest.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/runTest.o
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/runTest_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/runTest_cu.o: INCFLAGS += $(GTESTINC)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(testmain): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(testmain): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(testmain): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(testmain): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 else ifneq ($(shell $(CXX) --version | grep ^nvc++),) # support nvc++ #531
 $(testmain): LIBFLAGS += -L$(patsubst %bin/nvc++,%lib,$(subst ccache ,,$(CXX))) -lnvhpcatm -lnvcpumath -lnvc
 endif
@@ -713,14 +795,23 @@ $(testmain): LIBFLAGS += -lgomp
 endif
 endif
 
-ifeq ($(NVCC),) # link only runTest.o
+# Bypass std::filesystem completely to ease portability on LUMI #803
+#ifneq ($(findstring hipcc,$(GPUCC)),)
+#$(testmain): LIBFLAGS += -lstdc++fs
+#endif
+
+ifeq ($(GPUCC),) # link only runTest.o
 $(testmain): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(testmain): $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib) $(cxx_objects_exe) $(GTESTLIBS)
 	$(CXX) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) -ldl -pthread $(LIBFLAGS)
 else # link both runTest.o and runTest_cu.o
 $(testmain): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(testmain): $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) $(GTESTLIBS)
-	$(NVCC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) -lcuda
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) $(CUDATESTFLAGS) -lstdc++ -lpthread  -L$(shell dirname $(shell $(GPUCC) -print-prog-name=clang))/../../lib -lamdhip64
+else
+	$(GPUCC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) $(CUDATESTFLAGS)
+endif
 endif
 
 # Use target gtestlibs to build only googletest
@@ -829,9 +920,9 @@ ifeq ($(USECCACHE),1)
 	ccache --version | head -1
 endif
 	@echo ""
-	@echo NVCC=$(NVCC)
-ifneq ($(NVCC),)
-	$(NVCC) --version
+	@echo GPUCC=$(GPUCC)
+ifneq ($(GPUCC),)
+	$(GPUCC) --version
 endif
 	@echo ""
 	@echo CXX=$(CXX)
@@ -850,7 +941,7 @@ endif
 
 # Target: check (run the C++ test executable)
 # [NB THIS IS WHAT IS USED IN THE GITHUB CI!]
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 check: runTest cmpFcheck cmpFGcheck
 else
 check: runTest cmpFcheck
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/fbridge.cc b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/fbridge.cc
index 2d2b36d560..22ce3f5115 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/fbridge.cc
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/fbridge.cc
@@ -1,11 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Roiser (Oct 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "Bridge.h"
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 
 extern "C"
 {
@@ -22,7 +22,7 @@ extern "C"
    * Using the same Fortran MadEvent code, linking to the hetrerogeneous library would allow access to both CPU and GPU implementations.
    * The specific heterogeneous configuration (how many GPUs, how many threads on each CPU, etc) could be loaded in CUDA/C++ from a data file.
    */
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -46,8 +46,8 @@ extern "C"
    */
   void fbridgecreate_( CppObjectInFortran** ppbridge, const int* pnevtF, const int* pnparF, const int* pnp4F )
   {
-#ifdef __CUDACC__
-    CudaRuntime::setUp();
+#ifdef MGONGPUCPP_GPUIMPL
+    GpuRuntime::setUp();
 #endif
     // (NB: CPPProcess::initProc no longer needs to be executed here because it is called in the Bridge constructor)
     // FIXME: disable OMP in Bridge when called from Fortran
@@ -65,8 +65,8 @@ extern "C"
     Bridge<FORTRANFPTYPE>* pbridge = dynamic_cast<Bridge<FORTRANFPTYPE>*>( *ppbridge );
     if( pbridge == 0 ) throw std::runtime_error( "fbridgedelete_: invalid Bridge address" );
     delete pbridge;
-#ifdef __CUDACC__
-    CudaRuntime::tearDown();
+#ifdef MGONGPUCPP_GPUIMPL
+    GpuRuntime::tearDown();
 #endif
   }
 
@@ -96,7 +96,7 @@ extern "C"
   {
     Bridge<FORTRANFPTYPE>* pbridge = dynamic_cast<Bridge<FORTRANFPTYPE>*>( *ppbridge );
     if( pbridge == 0 ) throw std::runtime_error( "fbridgesequence_: invalid Bridge address" );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Use the device/GPU implementation in the CUDA library
     // (there is also a host implementation in this library)
     pbridge->gpu_sequence( momenta, gs, rndhel, rndcol, *pchannelId, mes, selhel, selcol );
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/fsampler.cc b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/fsampler.cc
index 2fb445372d..3743934f41 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/fsampler.cc
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/fsampler.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Feb 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "mgOnGpuConfig.h"
 
@@ -13,7 +13,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -40,7 +40,7 @@ namespace mg5amcCpu
   private:
     const int m_nevt; // The number of events in each iteration
     int m_iiter;      // The iteration counter (for random number seeding)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     HostBufferRndNumMomenta m_hstRndmom; // Memory buffers for random numbers
     HostBufferMomenta m_hstMomenta;      // Memory buffers for momenta
     HostBufferWeights m_hstWeights;      // Memory buffers for sampling weights
@@ -105,7 +105,7 @@ namespace mg5amcCpu
 
 extern "C"
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/runTest.cc b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/runTest.cc
index d4a760a71b..de327f2321 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/runTest.cc
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/runTest.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Hageboeck (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*xxx to run only testxxx.cc tests
 //----------------------------------------------------------------------------
@@ -18,7 +18,7 @@
 #include "RandomNumberKernels.h"
 #include "epoch_process_id.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using namespace mg5amcGpu;
 #else
 using namespace mg5amcCpu;
@@ -35,7 +35,7 @@ struct CUDA_CPU_TestBase : public TestDriverBase
     : TestDriverBase( npar, refFileName ) {}
 };
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 struct CPUTest : public CUDA_CPU_TestBase
 {
   // Struct data members (process, and memory structures for random numbers, momenta, matrix elements and weights on host and device)
@@ -119,7 +119,7 @@ struct CPUTest : public CUDA_CPU_TestBase
 };
 #endif
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 struct CUDATest : public CUDA_CPU_TestBase
 {
   // Reset the device when our test goes out of scope. Note that this should happen after
@@ -128,7 +128,7 @@ struct CUDATest : public CUDA_CPU_TestBase
   {
     ~DeviceReset()
     {
-      checkCuda( cudaDeviceReset() ); // this is needed by cuda-memcheck --leak-check full
+      checkGpu( gpuDeviceReset() ); // this is needed by cuda-memcheck --leak-check full
     }
   } deviceResetter;
 
@@ -256,7 +256,7 @@ INSTANTIATE_TEST_SUITE_P( prefix, \
                           test_suite_name, \
                           testing::Values( new CUDATest( MG_EPOCH_REFERENCE_FILE_NAME ) ) );
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 MG_INSTANTIATE_TEST_SUITE_GPU( XTESTID_GPU( MG_EPOCH_PROCESS_ID ), MadgraphTest );
 #else
 MG_INSTANTIATE_TEST_SUITE_CPU( XTESTID_CPU( MG_EPOCH_PROCESS_ID ), MadgraphTest );
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/testmisc.cc b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/testmisc.cc
index 895d6eeb56..ba9e59a8a3 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/testmisc.cc
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/testmisc.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*misc to run only testmisc.cc tests
 //----------------------------------------------------------------------------
@@ -17,7 +17,7 @@
 #include <sstream>
 #include <typeinfo>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #define TESTID( s ) s##_GPU_MISC
 #else
 #define TESTID( s ) s##_CPU_MISC
@@ -26,7 +26,7 @@
 #define XTESTID( s ) TESTID( s )
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -59,7 +59,7 @@ namespace mg5amcCpu
 
 TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testmisc )
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/testxxx.cc b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/testxxx.cc
index 3361fe5aa9..e5167de00c 100644
--- a/epochX/cudacpp/gg_ttggg.mad/SubProcesses/testxxx.cc
+++ b/epochX/cudacpp/gg_ttggg.mad/SubProcesses/testxxx.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Apr 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*xxx to run only testxxx.cc tests
 //----------------------------------------------------------------------------
@@ -24,7 +24,7 @@
 #include <iomanip>
 #include <iostream>
 #include <vector>
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #define TESTID( s ) s##_GPU_XXX
 #else
 #define TESTID( s ) s##_CPU_XXX
@@ -32,7 +32,7 @@
 
 #define XTESTID( s ) TESTID( s )
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -42,7 +42,7 @@ namespace mg5amcCpu
   int FPEhandlerIevt = -1;
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU): '" << FPEhandlerMessage << "' ievt=" << FPEhandlerIevt << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU neppV=" << neppV << "): '" << FPEhandlerMessage << "' ievt=" << FPEhandlerIevt << std::endl;
@@ -53,7 +53,7 @@ namespace mg5amcCpu
 
 TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -77,7 +77,7 @@ TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
   assert( nevt % neppM == 0 ); // nevt must be a multiple of neppM
   assert( nevt % neppV == 0 ); // nevt must be a multiple of neppV
   // Fill in the input momenta
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   mg5amcGpu::PinnedHostBufferMomenta hstMomenta( nevt ); // AOSOA[npagM][npar=4][np4=4][neppM]
 #else
   mg5amcCpu::HostBufferMomenta hstMomenta( nevt ); // AOSOA[npagM][npar=4][np4=4][neppM]
@@ -322,7 +322,7 @@ TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
   {
     for( int ievt = 0; ievt < nevt; ievt++ )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       using namespace mg5amcGpu;
 #else
       using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_ttggg.mad/bin/internal/launch_plugin.py b/epochX/cudacpp/gg_ttggg.mad/bin/internal/launch_plugin.py
index 3b09713e12..90af6b7053 100644
--- a/epochX/cudacpp/gg_ttggg.mad/bin/internal/launch_plugin.py
+++ b/epochX/cudacpp/gg_ttggg.mad/bin/internal/launch_plugin.py
@@ -103,7 +103,7 @@ def default_setup(self):
                        fct_mod=(self.reset_makeopts,(),{}),
                        allowed=['auto', 'none', 'sse4', 'avx2','512y','512z'])
         self.add_param('cudacpp_backend', 'CPP', include=False, hidden=False,
-                       allowed=['Fortan', 'CPP', 'CUDA'])
+                       allowed=['Fortran', 'CPP', 'CUDA'])
         self['vector_size'] = 16 # already setup in default class (just change value)
         self['aloha_flag'] = '--fast-math'
         self['matrix_flag'] = '-O3'
diff --git a/epochX/cudacpp/gg_ttggg.mad/src/HelAmps_sm.h b/epochX/cudacpp/gg_ttggg.mad/src/HelAmps_sm.h
index 8df465ad6d..8b4ad719be 100644
--- a/epochX/cudacpp/gg_ttggg.mad/src/HelAmps_sm.h
+++ b/epochX/cudacpp/gg_ttggg.mad/src/HelAmps_sm.h
@@ -5,7 +5,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -28,7 +28,7 @@
 //#include <iomanip>
 //#include <iostream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttggg.mad/src/Parameters_sm.cc b/epochX/cudacpp/gg_ttggg.mad/src/Parameters_sm.cc
index 64fc3fea62..067445b198 100644
--- a/epochX/cudacpp/gg_ttggg.mad/src/Parameters_sm.cc
+++ b/epochX/cudacpp/gg_ttggg.mad/src/Parameters_sm.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -17,7 +17,7 @@
 #include <iomanip>
 #include <iostream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using namespace mg5amcGpu;
 #else
 using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_ttggg.mad/src/Parameters_sm.h b/epochX/cudacpp/gg_ttggg.mad/src/Parameters_sm.h
index b6568d3761..9581d66e0e 100644
--- a/epochX/cudacpp/gg_ttggg.mad/src/Parameters_sm.h
+++ b/epochX/cudacpp/gg_ttggg.mad/src/Parameters_sm.h
@@ -27,7 +27,7 @@
 #include "read_slha.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -93,7 +93,7 @@ namespace mg5amcCpu
 #include <limits>
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -218,7 +218,7 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -239,7 +239,7 @@ namespace mg5amcCpu
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wunused-variable"  // e.g. <<warning: unused variable ‘mdl_G__exp__2’ [-Wunused-variable]>>
 #pragma GCC diagnostic ignored "-Wunused-parameter" // e.g. <<warning: unused parameter ‘G’ [-Wunused-parameter]>>
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #pragma nv_diagnostic push
 #pragma nv_diag_suppress 177 // e.g. <<warning #177-D: variable "mdl_G__exp__2" was declared but never referenced>>
 #endif
@@ -267,7 +267,7 @@ namespace mg5amcCpu
       // End SM implementation - no special handling of vectors of floats as in EFT (#439)
       return out;
     }
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #pragma GCC diagnostic pop
 #pragma nv_diagnostic pop
 #endif
diff --git a/epochX/cudacpp/gg_ttggg.mad/src/cudacpp_src.mk b/epochX/cudacpp/gg_ttggg.mad/src/cudacpp_src.mk
index d4cc628aec..b2b9da5288 100644
--- a/epochX/cudacpp/gg_ttggg.mad/src/cudacpp_src.mk
+++ b/epochX/cudacpp/gg_ttggg.mad/src/cudacpp_src.mk
@@ -19,7 +19,7 @@ SHELL := /bin/bash
 #=== Configure common compiler flags for CUDA and C++
 
 INCFLAGS = -I.
-OPTFLAGS = -O3 # this ends up in CUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
+OPTFLAGS = -O3 # this ends up in GPUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
 
 #-------------------------------------------------------------------------------
 
@@ -45,13 +45,13 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure the CUDA compiler (note: NVCC is already exported including ccache)
+#=== Configure the CUDA compiler (note: GPUCC is already exported including ccache)
 
-###$(info NVCC=$(NVCC))
+###$(info GPUCC=$(GPUCC))
 
 #-------------------------------------------------------------------------------
 
-#=== Configure ccache for C++ builds (note: NVCC is already exported including ccache)
+#=== Configure ccache for C++ builds (note: GPUCC is already exported including ccache)
 
 # Enable ccache if USECCACHE=1
 ifeq ($(USECCACHE)$(shell echo $(CXX) | grep ccache),1)
@@ -92,6 +92,13 @@ endif
 ###$(info OMPFLAGS=$(OMPFLAGS))
 CXXFLAGS += $(OMPFLAGS)
 
+# Add correct flags for nvcc (-x cu) and hipcc (-x hip) for GPU code (see #810)
+ifeq ($(findstring nvcc,$(GPUCC)),nvcc)
+  GPUFLAGS += -Xcompiler -fPIC -c -x cu
+else ifeq ($(findstring hipcc,$(GPUCC)),hipcc)
+  GPUFLAGS += -fPIC -c -x hip
+endif
+
 # Set the build flags appropriate to each AVX choice (example: "make AVX=none")
 # [NB MGONGPU_PVW512 is needed because "-mprefer-vector-width=256" is not exposed in a macro]
 # [See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96476]
@@ -253,20 +260,20 @@ $(BUILDDIR)/%.o : %.cc *.h $(BUILDDIR)/.build.$(TAG)
 # Generic target and build rules: objects from CUDA compilation
 $(BUILDDIR)/%_cu.o : %.cc *.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c -x cu $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $< -o $@
 
 #-------------------------------------------------------------------------------
 
 cxx_objects=$(addprefix $(BUILDDIR)/, Parameters_sm.o read_slha.o)
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 cu_objects=$(addprefix $(BUILDDIR)/, Parameters_sm_cu.o)
 endif
 
 # Target (and build rules): common (src) library
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so : $(cxx_objects) $(cu_objects)
 	@if [ ! -d $(LIBDIR) ]; then echo "mkdir -p $(LIBDIR)"; mkdir -p $(LIBDIR); fi
-	$(NVCC) -shared -o $@ $(cxx_objects) $(cu_objects) $(LDFLAGS)
+	$(GPUCC) -shared -o $@ $(cxx_objects) $(cu_objects) $(LDFLAGS)
 else
 $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so : $(cxx_objects)
 	@if [ ! -d $(LIBDIR) ]; then echo "mkdir -p $(LIBDIR)"; mkdir -p $(LIBDIR); fi
diff --git a/epochX/cudacpp/gg_ttggg.mad/src/mgOnGpuConfig.h b/epochX/cudacpp/gg_ttggg.mad/src/mgOnGpuConfig.h
index 80032e528b..6bde4466d0 100644
--- a/epochX/cudacpp/gg_ttggg.mad/src/mgOnGpuConfig.h
+++ b/epochX/cudacpp/gg_ttggg.mad/src/mgOnGpuConfig.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jul 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUCONFIG_H
 #define MGONGPUCONFIG_H 1
@@ -10,12 +10,26 @@
 // There are two different code bases for standalone_cudacpp (without multichannel) and madevent+cudacpp (with multichannel)
 #define MGONGPU_SUPPORTS_MULTICHANNEL 1
 
+// Is this a GPU (CUDA, HIP) or CPU implementation?
+#ifdef __CUDACC__
+#define MGONGPUCPP_GPUIMPL cuda
+#elif defined __HIPCC__
+#define MGONGPUCPP_GPUIMPL hip
+#include "hip/hip_runtime.h" // needed for blockDim, blockIdx, threadIdx: better in mgOnGpuConfig.h than in GpuAbstraction.h
+#else
+#undef MGONGPUCPP_GPUIMPL
+#endif
+
 // ** NB1 Throughputs (e.g. 6.8E8) are events/sec for "./gcheck.exe -p 65536 128 12"
 // ** NB2 Baseline on b7g47n0004 fluctuates (probably depends on load on other VMs)
 
 // Choose if curand is supported for generating random numbers
+// For HIP, by default, do not use curand (common random numbers will be used instead)
 // For both CUDA and C++, by default, do not inline, but allow this macro to be set from outside with e.g. -DMGONGPU_HAS_NO_CURAND
-// (there exist CUDA installations, e.g. using the HPC package, which do not include curand - see PR #784)
+// (there exist CUDA installations, e.g. using the HPC package, which do not include curand - see PR #784 and #785)
+#if defined __HIPCC__
+#define MGONGPU_HAS_NO_CURAND 1
+#else
 //#ifdef __CUDACC__
 //#undef MGONGPU_HAS_NO_CURAND // default
 ////#define MGONGPU_HAS_NO_CURAND 1
@@ -23,6 +37,7 @@
 //#undef MGONGPU_HAS_NO_CURAND // default
 ////#define MGONGPU_HAS_NO_CURAND 1
 //#endif
+#endif
 
 // Choose floating point precision (for everything but color algebra #537)
 // If one of these macros has been set from outside with e.g. -DMGONGPU_FPTYPE_FLOAT, nothing happens (issue #167)
@@ -54,23 +69,31 @@
 //#undef MGONGPU_HARDCODE_PARAM // default
 ////#define MGONGPU_HARDCODE_PARAM 1
 
-// Complex type in c++: std::complex or cxsmpl (CHOOSE ONLY ONE)
-#ifndef __CUDACC__
-//#define MGONGPU_CPPCXTYPE_STDCOMPLEX 1 // ~8 percent slower on float, same on double (5.1E6/double, 9.4E6/float)
-#define MGONGPU_CPPCXTYPE_CXSMPL 1 // new default (5.1E6/double, 10.2E6/float)
-#endif
-
-// Complex type in cuda: thrust or cucomplex or cxsmpl (CHOOSE ONLY ONE)
+// Complex type in CUDA: thrust or cucomplex or cxsmpl (CHOOSE ONLY ONE)
+// (NB THIS IS MGONGPU_*CU*CXTYPE_xxx)
 #ifdef __CUDACC__
 #define MGONGPU_CUCXTYPE_THRUST 1 // default (~1.15E9/double, ~3.2E9/float)
 //#define MGONGPU_CUCXTYPE_CUCOMPLEX 1 // ~10 percent slower (1.03E9/double, ~2.8E9/float)
 //#define MGONGPU_CUCXTYPE_CXSMPL 1 // ~10 percent slower (1.00E9/double, ~2.9E9/float)
+
+// Complex type in HIP: cxsmpl (ONLY ONE OPTION POSSIBLE? #810)
+// (NB THIS IS MGONGPU_*HIP*CXTYPE_xxx)
+#elif defined __HIPCC__
+#define MGONGPU_HIPCXTYPE_CXSMPL 1 // default for HIP
+
+// Complex type in C++: std::complex or cxsmpl (CHOOSE ONLY ONE)
+// (NB THIS IS MGONGPU_*CPP*CXTYPE_xxx)
+#else
+//#define MGONGPU_CPPCXTYPE_STDCOMPLEX 1 // ~8 percent slower on float, same on double (5.1E6/double, 9.4E6/float)
+#define MGONGPU_CPPCXTYPE_CXSMPL 1 // new default (5.1E6/double, 10.2E6/float)
 #endif
 
-// Cuda nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
+// CUDA nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
 #ifdef __CUDACC__
-#undef MGONGPU_NSIGHT_DEBUG // default
+#undef MGONGPU_NSIGHT_DEBUG // default in CUDA
 //#define MGONGPU_NSIGHT_DEBUG 1
+#else
+#undef MGONGPU_NSIGHT_DEBUG // only option in HIP or C++
 #endif
 
 // SANITY CHECKS (floating point precision for everything but color algebra #537)
@@ -86,17 +109,21 @@
 #error You cannot use double precision for color algebra and single precision elsewhere
 #endif
 
-// SANITY CHECKS (c++ complex number implementation)
-#ifndef __CUDACC__
-#if defined MGONGPU_CPPCXTYPE_STDCOMPLEX and defined MGONGPU_CPPCXTYPE_CXSMPL
-#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CPPCXTYPE_STDCOMPLEX or MGONGPU_CPPCXTYPE_CXSMPL
+// SANITY CHECKS (CUDA complex number implementation)
+#ifdef __CUDACC__
+#if defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CUCOMPLEX
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX for CUDA
+#elif defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CXSMPL for CUDA
+#elif defined MGONGPU_CUCXTYPE_CUCOMPLEX and defined MGONGPU_CUCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE OF MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL for CUDA
 #endif
 #endif
 
-// SANITY CHECKS (cuda complex number implementation)
-#ifdef __CUDACC__
-#if defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CUCOMPLEX and defined MGONGPU_CUCXTYPE_CXSMPL
-#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL
+// SANITY CHECKS (C++ complex number implementation)
+#ifndef MGONGPUCPP_GPUIMPL
+#if defined MGONGPU_CPPCXTYPE_STDCOMPLEX and defined MGONGPU_CPPCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CPPCXTYPE_STDCOMPLEX or MGONGPU_CPPCXTYPE_CXSMPL for C++
 #endif
 #endif
 
@@ -134,7 +161,7 @@ namespace mgOnGpu
   // Alignment requirement for using reinterpret_cast with SIMD vectorized code
   // (using reinterpret_cast with non aligned memory may lead to segmentation faults!)
   // Only needed for C++ code but can be enforced also in NVCC builds of C++ code using CUDA>=11.2 and C++17 (#318, #319, #333)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   constexpr int cppAlign = 64; // alignment requirement for SIMD vectorization (64-byte i.e. 512-bit)
 #endif
 
@@ -145,7 +172,7 @@ using mgOnGpu::fptype;
 using mgOnGpu::fptype2;
 
 // C++ SIMD vectorization width (this will be used to set neppV)
-#ifdef __CUDACC__ // CUDA implementation has no SIMD
+#ifdef MGONGPUCPP_GPUIMPL // CUDA and HIP implementations have no SIMD
 #undef MGONGPU_CPPSIMD
 #elif defined __AVX512VL__ && defined MGONGPU_PVW512 // C++ "512z" AVX512 with 512 width (512-bit ie 64-byte): 8 (DOUBLE) or 16 (FLOAT)
 #ifdef MGONGPU_FPTYPE_DOUBLE
@@ -175,9 +202,9 @@ using mgOnGpu::fptype2;
 #undef MGONGPU_CPPSIMD
 #endif
 
-// Cuda nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
+// CUDA nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
 // Arguments (not used so far): text is __FUNCTION__, code is 0 (start) or 1 (end)
-#if defined __CUDACC__ && defined MGONGPU_NSIGHT_DEBUG /* clang-format off */
+#if defined __CUDA__ && defined MGONGPU_NSIGHT_DEBUG /* clang-format off */
 #define mgDebugDeclare() __shared__ float mgDebugCounter[mgOnGpu::ntpbMAX];
 #define mgDebugInitialise() { mgDebugCounter[threadIdx.x] = 0; }
 #define mgDebug( code, text ) { mgDebugCounter[threadIdx.x] += 1; }
@@ -189,8 +216,8 @@ using mgOnGpu::fptype2;
 #define mgDebugFinalise() { /*noop*/ }
 #endif /* clang-format on */
 
-// Define empty CUDA declaration specifiers for C++
-#ifndef __CUDACC__
+// Define empty CUDA/HIP declaration specifiers for C++
+#ifndef MGONGPUCPP_GPUIMPL
 #define __global__
 #define __host__
 #define __device__
diff --git a/epochX/cudacpp/gg_ttggg.mad/src/mgOnGpuCxtypes.h b/epochX/cudacpp/gg_ttggg.mad/src/mgOnGpuCxtypes.h
index ca9a9f00c0..7ede1dbfae 100644
--- a/epochX/cudacpp/gg_ttggg.mad/src/mgOnGpuCxtypes.h
+++ b/epochX/cudacpp/gg_ttggg.mad/src/mgOnGpuCxtypes.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022, based on earlier work by D. Smith) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUCXTYPES_H
 #define MGONGPUCXTYPES_H 1
@@ -19,7 +19,7 @@
 #include <complex>
 
 // Complex type in cuda: thrust or cucomplex or cxsmpl
-#ifdef __CUDACC__
+#ifdef __CUDACC__ // this must be __CUDAC__ (not MGONGPUCPP_GPUIMPL)
 #if defined MGONGPU_CUCXTYPE_THRUST
 #pragma clang diagnostic push
 #pragma clang diagnostic ignored "-Wtautological-compare" // for icpx2021/clang13 (https://stackoverflow.com/a/15864661)
@@ -30,8 +30,13 @@
 #elif not defined MGONGPU_CUCXTYPE_CXSMPL
 #error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL
 #endif
+// Complex type in HIP: cxsmpl
+#elif defined __HIPCC__
+#if not defined MGONGPU_HIPCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_HIPCXTYPE_CXSMPL
+#endif
 #else
-// Complex type in c++: std::complex or cxsmpl
+// Complex type in c++ or HIP: std::complex or cxsmpl
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
 #include <cmath>
 #elif not defined MGONGPU_CPPCXTYPE_CXSMPL
@@ -82,7 +87,7 @@ namespace mgOnGpu /* clang-format off */
 using mgOnGpu::cxsmpl;
 
 // Printout to stream for user defined types
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -92,7 +97,7 @@ namespace mg5amcCpu
   inline __host__ std::ostream&
   operator<<( std::ostream& out, const cxsmpl<FP>& c )
   {
-    out << std::complex( c.real(), c.imag() );
+    out << std::complex<FP>( c.real(), c.imag() );
     return out;
   }
 
@@ -215,14 +220,14 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
   // --- Type definitions (complex type: cxtype)
-#ifdef __CUDACC__ // cuda
+#ifdef __CUDACC__ // this must be __CUDAC__ (not MGONGPUCPP_GPUIMPL)
 #if defined MGONGPU_CUCXTYPE_THRUST
   typedef thrust::complex<fptype> cxtype;
 #elif defined MGONGPU_CUCXTYPE_CUCOMPLEX
@@ -255,13 +260,13 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
-#if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+#if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
 
   //------------------------------
   // CUDA or C++ - using cxsmpl
@@ -303,11 +308,11 @@ namespace mg5amcCpu
     return cxmake( c.real(), c.imag() );
   }
 
-#endif // #if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+#endif // #if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
 
   //==========================================================================
 
-#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_THRUST // cuda + thrust
+#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_THRUST // cuda + thrust (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // CUDA - using thrust::complex
@@ -347,7 +352,7 @@ namespace mg5amcCpu
 
   //==========================================================================
 
-#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_CUCOMPLEX // cuda + cucomplex
+#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_CUCOMPLEX // cuda + cucomplex (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // CUDA - using cuComplex
@@ -566,7 +571,7 @@ namespace mg5amcCpu
 
   //==========================================================================
 
-#if not defined __CUDACC__ and defined MGONGPU_CPPCXTYPE_STDCOMPLEX // c++ + stdcomplex
+#if not defined __CUDACC__ and defined MGONGPU_CPPCXTYPE_STDCOMPLEX // c++/hip + stdcomplex (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // C++ - using std::complex
@@ -633,7 +638,7 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttggg.mad/src/mgOnGpuFptypes.h b/epochX/cudacpp/gg_ttggg.mad/src/mgOnGpuFptypes.h
index 905c97d700..fa3a02664b 100644
--- a/epochX/cudacpp/gg_ttggg.mad/src/mgOnGpuFptypes.h
+++ b/epochX/cudacpp/gg_ttggg.mad/src/mgOnGpuFptypes.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUFPTYPES_H
 #define MGONGPUFPTYPES_H 1
@@ -12,7 +12,7 @@
 #include <cmath>
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL // cuda
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -20,7 +20,7 @@ namespace mg5amcCpu
 {
   //==========================================================================
 
-#ifdef __CUDACC__ // cuda
+#ifdef MGONGPUCPP_GPUIMPL // cuda
 
   //------------------------------
   // Floating point types - Cuda
@@ -64,11 +64,11 @@ namespace mg5amcCpu
 #endif
   }
 
-#endif // #ifdef __CUDACC__
+#endif // #ifdef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
   //------------------------------
   // Floating point types - C++
@@ -92,7 +92,7 @@ namespace mg5amcCpu
     return std::sqrt( f );
   }
 
-#endif // #ifndef __CUDACC__
+#endif // #ifndef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
diff --git a/epochX/cudacpp/gg_ttggg.mad/src/mgOnGpuVectors.h b/epochX/cudacpp/gg_ttggg.mad/src/mgOnGpuVectors.h
index e1299ba81e..cdae04326b 100644
--- a/epochX/cudacpp/gg_ttggg.mad/src/mgOnGpuVectors.h
+++ b/epochX/cudacpp/gg_ttggg.mad/src/mgOnGpuVectors.h
@@ -32,7 +32,7 @@
 #endif
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -131,7 +131,7 @@ namespace mg5amcCpu
 #endif
 #endif
 
-#else // i.e #ifndef MGONGPU_CPPSIMD (this includes #ifdef __CUDACC__)
+#else // i.e #ifndef MGONGPU_CPPSIMD (this includes #ifdef MGONGPUCPP_GPUIMPL)
 
   const int neppV = 1;
 
@@ -153,13 +153,13 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
   // Printout to stream for user defined types
 
@@ -805,11 +805,11 @@ namespace mg5amcCpu
 
 #endif // #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
 
-#endif // #ifndef __CUDACC__
+#endif // #ifndef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
   //------------------------------
   // Vector types - CUDA
@@ -853,12 +853,12 @@ namespace mg5amcCpu
     return mask;
   }
 
-#endif // #ifdef __CUDACC__
+#endif // #ifdef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
   // Scalar-or-vector types: scalar in CUDA, vector or scalar in C++
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   typedef bool bool_sv;
   typedef fptype fptype_sv;
   typedef fptype2 fptype2_sv;
@@ -879,7 +879,7 @@ namespace mg5amcCpu
 #endif
 
   // Scalar-or-vector zeros: scalar in CUDA, vector or scalar in C++
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   inline __host__ __device__ cxtype cxzero_sv(){ return cxtype( 0, 0 ); }
 #elif defined MGONGPU_CPPSIMD
   inline cxtype_v cxzero_sv() { return cxtype_v(); } // RRRR=0000 IIII=0000
diff --git a/epochX/cudacpp/gg_ttggg.mad/src/rambo.h b/epochX/cudacpp/gg_ttggg.mad/src/rambo.h
index e02ea52496..cd7e1008ea 100644
--- a/epochX/cudacpp/gg_ttggg.mad/src/rambo.h
+++ b/epochX/cudacpp/gg_ttggg.mad/src/rambo.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -18,7 +18,7 @@
 #include <iostream>
 
 // Simplified rambo version for 2 to N (with N>=2) processes with massless particles
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +83,7 @@ namespace mg5amcCpu
       static bool first = true;
       if( first )
       {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
         if constexpr( M_ACCESS::isOnDevice() ) // avoid
         {
           const int ievt0 = 0;
@@ -166,7 +166,7 @@ namespace mg5amcCpu
     wt = po2log;
     if( nparf != 2 ) wt = ( 2. * nparf - 4. ) * log( energy ) + z[nparf - 1];
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     // issue warnings if weight is too small or too large
     static int iwarn[5] = { 0, 0, 0, 0, 0 };
     if( wt < -180. )
diff --git a/epochX/cudacpp/gg_ttggg.mad/src/read_slha.cc b/epochX/cudacpp/gg_ttggg.mad/src/read_slha.cc
index 055b19a779..f8e46f2e66 100644
--- a/epochX/cudacpp/gg_ttggg.mad/src/read_slha.cc
+++ b/epochX/cudacpp/gg_ttggg.mad/src/read_slha.cc
@@ -11,7 +11,11 @@
 
 #include <algorithm>
 #include <cstdlib>
-#include <filesystem>
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
 #include <fstream>
 #include <iostream>
 
@@ -60,7 +64,14 @@ SLHAReader::read_slha_file( std::string file_name, bool verbose )
     {
       std::cout << "WARNING! Card file '" << file_name << "' does not exist:"
                 << " look for the file in directory $" << envpath << "='" << getenv( envpath ) << "'" << std::endl;
+      /*
+#ifdef __HIPCC__
+      const std::string file_name2 = std::experimental::filesystem::path( getenv( envpath ) ) / std::experimental::filesystem::path( file_name ).filename();
+#else
       const std::string file_name2 = std::filesystem::path( getenv( envpath ) ) / std::filesystem::path( file_name ).filename();
+#endif
+      */
+      const std::string file_name2 = std::string( getenv( envpath ) ) + "/" + file_name; // bypass std::filesystem #803
       param_card.open( file_name2.c_str(), std::ifstream::in );
       if( param_card.good() )
       {
diff --git a/epochX/cudacpp/gg_ttggg.sa/CODEGEN_cudacpp_gg_ttggg_log.txt b/epochX/cudacpp/gg_ttggg.sa/CODEGEN_cudacpp_gg_ttggg_log.txt
index 0970bf8b4c..382962d284 100644
--- a/epochX/cudacpp/gg_ttggg.sa/CODEGEN_cudacpp_gg_ttggg_log.txt
+++ b/epochX/cudacpp/gg_ttggg.sa/CODEGEN_cudacpp_gg_ttggg_log.txt
@@ -52,7 +52,7 @@ Note that you can still compile and run aMC@NLO with the built-in PDFs
 
 Using default text editor "vi". Set another one in ./input/mg5_configuration.txt
 Using default eps viewer "evince". Set another one in ./input/mg5_configuration.txt
-No valid web browser found. Please set in ./input/mg5_configuration.txt
+Using default web browser "firefox". Set another one in ./input/mg5_configuration.txt
 import /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_cudacpp_gg_ttggg.mg
 The import format was not given, so we guess it as command
 set stdout_level DEBUG
@@ -62,7 +62,7 @@ generate g g > t t~ g g g
 No model currently active, so we import the Standard Model
 INFO: load particles 
 INFO: load vertices 
-DEBUG: model prefixing  takes 0.005753755569458008 
+DEBUG: model prefixing  takes 0.005497932434082031 
 INFO: Restrict model sm with file models/sm/restrict_default.dat . 
 DEBUG: Simplifying conditional expressions 
 DEBUG: remove interactions: u s w+ at order: QED=1 
@@ -155,7 +155,7 @@ INFO: Please specify coupling orders to bypass this step.
 INFO: Trying coupling order WEIGHTED<=5: WEIGTHED IS QCD+2*QED 
 INFO: Trying process: g g > t t~ g g g WEIGHTED<=5 @1  
 INFO: Process has 1240 diagrams 
-1 processes with 1240 diagrams generated in 1.912 s
+1 processes with 1240 diagrams generated in 1.865 s
 Total: 1 processes with 1240 diagrams
 output standalone_cudacpp ../TMPOUT/CODEGEN_cudacpp_gg_ttggg
 Load PLUGIN.CUDACPP_OUTPUT
@@ -175,7 +175,7 @@ INFO: Creating files in directory /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TM
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_cudacpp_gg_ttggg/SubProcesses/P1_Sigma_sm_gg_ttxggg/./CPPProcess.h
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_cudacpp_gg_ttggg/SubProcesses/P1_Sigma_sm_gg_ttxggg/./CPPProcess.cc
 INFO: Created files CPPProcess.h and CPPProcess.cc in directory /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_cudacpp_gg_ttggg/SubProcesses/P1_Sigma_sm_gg_ttxggg/. 
-Generated helas calls for 1 subprocesses (1240 diagrams) in 6.716 s
+Generated helas calls for 1 subprocesses (1240 diagrams) in 6.712 s
 DEBUG:  Entering PLUGIN_ProcessExporter.convert_model (create the model) [output.py at line 202] 
 ALOHA: aloha starts to compute helicity amplitudes
 ALOHA: aloha creates VVV1 routines
@@ -183,7 +183,7 @@ ALOHA: aloha creates FFV1 routines
 ALOHA: aloha creates VVVV1 routines
 ALOHA: aloha creates VVVV3 routines
 ALOHA: aloha creates VVVV4 routines
-ALOHA: aloha creates 5 routines in  0.352 s
+ALOHA: aloha creates 5 routines in  0.345 s
 <class 'aloha.create_aloha.AbstractRoutine'> VVV1
 <class 'aloha.create_aloha.AbstractRoutine'> VVV1
 <class 'aloha.create_aloha.AbstractRoutine'> FFV1
@@ -206,7 +206,7 @@ INFO: Created files Parameters_sm.h and Parameters_sm.cc in directory
 INFO: /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_cudacpp_gg_ttggg/src/. and /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_cudacpp_gg_ttggg/src/. 
 quit
 
-real	0m13.290s
-user	0m13.123s
-sys	0m0.115s
+real	0m13.123s
+user	0m12.875s
+sys	0m0.149s
 Code generation completed in 13 seconds
diff --git a/epochX/cudacpp/gg_ttggg.sa/COPYRIGHT b/epochX/cudacpp/gg_ttggg.sa/COPYRIGHT
index a134b5fef9..9036d9260a 100644
--- a/epochX/cudacpp/gg_ttggg.sa/COPYRIGHT
+++ b/epochX/cudacpp/gg_ttggg.sa/COPYRIGHT
@@ -15,6 +15,7 @@ The full development team currently includes the following authors :
   Stephan Hageboeck (CERN)
   Olivier Mattelaer (Universite Catholique de Louvain, original author)
   Stefan Roiser (CERN, original author)
+  Jorgen Teig (CERN)
   Andrea Valassi (CERN, original author)
   Zenny Wettersten (CERN)
 See https://github.com/madgraph5/madgraph4gpu for more details. For the full
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/Bridge.h b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/Bridge.h
index bf8b5e024d..f9ed70dfde 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/Bridge.h
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/Bridge.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Roiser (Nov 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef BRIDGE_H
 #define BRIDGE_H 1
@@ -14,16 +14,23 @@
 #include "MemoryAccessMomenta.h"  // for MemoryAccessMomenta::neppM
 #include "MemoryBuffers.h"        // for HostBufferMomenta, DeviceBufferMomenta etc
 
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
+
+#include <sys/stat.h> // bypass std::filesystem #803
+
 #include <algorithm>
 #include <cassert>
 #include <cmath>
 #include <cstring>
-#include <filesystem>
 #include <iostream>
 #include <memory>
 #include <type_traits>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +90,7 @@ namespace mg5amcCpu
     Bridge& operator=( const Bridge& ) = delete;
     Bridge& operator=( Bridge&& ) = delete;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     /**
      * Set the gpublocks and gputhreads for the gpusequence - throws if evnt != gpublocks*gputhreads
      * (this is needed for BridgeKernel tests rather than for actual production use in Fortran)
@@ -150,7 +157,7 @@ namespace mg5amcCpu
     unsigned int m_nevt; // number of events
     int m_nGoodHel;      // the number of good helicities (-1 initially when they have not yet been calculated)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     int m_gputhreads; // number of gpu threads (default set from number of events, can be modified)
     int m_gpublocks;  // number of gpu blocks (default set from number of events, can be modified)
     DeviceBuffer<FORTRANFPTYPE, sizePerEventMomenta> m_devMomentaF;
@@ -187,12 +194,12 @@ namespace mg5amcCpu
   // Forward declare transposition methods
   //
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
   template<typename Tin, typename Tout>
   __global__ void dev_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt );
 
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
 
   template<typename Tin, typename Tout>
   void hst_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt );
@@ -209,7 +216,7 @@ namespace mg5amcCpu
   Bridge<FORTRANFPTYPE>::Bridge( unsigned int nevtF, unsigned int nparF, unsigned int np4F )
     : m_nevt( nevtF )
     , m_nGoodHel( -1 )
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     , m_gputhreads( 256 )                  // default number of gpu threads
     , m_gpublocks( m_nevt / m_gputhreads ) // this ensures m_nevt <= m_gpublocks*m_gputhreads
     , m_devMomentaF( m_nevt )
@@ -233,7 +240,7 @@ namespace mg5amcCpu
   {
     if( nparF != CPPProcess::npar ) throw std::runtime_error( "Bridge constructor: npar mismatch" );
     if( np4F != CPPProcess::np4 ) throw std::runtime_error( "Bridge constructor: np4 mismatch" );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( ( m_nevt < s_gputhreadsmin ) || ( m_nevt % s_gputhreadsmin != 0 ) )
       throw std::runtime_error( "Bridge constructor: nevt should be a multiple of " + std::to_string( s_gputhreadsmin ) );
     while( m_nevt != m_gpublocks * m_gputhreads )
@@ -249,20 +256,28 @@ namespace mg5amcCpu
 #else
     std::cout << "WARNING! Instantiate host Bridge (nevt=" << m_nevt << ")" << std::endl;
     m_pmek.reset( new MatrixElementKernelHost( m_hstMomentaC, m_hstGs, m_hstRndHel, m_hstRndCol, m_hstMEs, m_hstSelHel, m_hstSelCol, m_nevt ) );
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
     // Create a process object, read param card and set parameters
     // FIXME: the process instance can happily go out of scope because it is only needed to read parameters?
     // FIXME: the CPPProcess should really be a singleton? what if fbridgecreate is called from several Fortran threads?
     CPPProcess process( /*verbose=*/false );
     std::string paramCard = "../../Cards/param_card.dat";
-    if( !std::filesystem::exists( paramCard ) )
-    {
-      paramCard = "../" + paramCard;
-    }
+    /*
+#ifdef __HIPCC__
+    if( !std::experimental::filesystem::exists( paramCard ) ) paramCard = "../" + paramCard;
+#else
+    if( !std::filesystem::exists( paramCard ) ) paramCard = "../" + paramCard;
+#endif
+    */
+    //struct stat dummybuffer; // bypass std::filesystem #803
+    //if( !( stat( paramCard.c_str(), &dummyBuffer ) == 0 ) ) paramCard = "../" + paramCard; //
+    auto fileExists = []( std::string& fileName )
+    { struct stat buffer; return stat( fileName.c_str(), &buffer ) == 0; };
+    if( !fileExists( paramCard ) ) paramCard = "../" + paramCard; // bypass std::filesystem #803
     process.initProc( paramCard );
   }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::set_gpugrid( const int gpublocks, const int gputhreads )
   {
@@ -276,7 +291,7 @@ namespace mg5amcCpu
   }
 #endif
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::gpu_sequence( const FORTRANFPTYPE* momenta,
                                             const FORTRANFPTYPE* gs,
@@ -291,14 +306,14 @@ namespace mg5amcCpu
     constexpr int neppM = MemoryAccessMomenta::neppM;
     if constexpr( neppM == 1 && std::is_same_v<FORTRANFPTYPE, fptype> )
     {
-      checkCuda( cudaMemcpy( m_devMomentaC.data(), momenta, m_devMomentaC.bytes(), cudaMemcpyHostToDevice ) );
+      gpuMemcpy( m_devMomentaC.data(), momenta, m_devMomentaC.bytes(), gpuMemcpyHostToDevice );
     }
     else
     {
-      checkCuda( cudaMemcpy( m_devMomentaF.data(), momenta, m_devMomentaF.bytes(), cudaMemcpyHostToDevice ) );
+      gpuMemcpy( m_devMomentaF.data(), momenta, m_devMomentaF.bytes(), gpuMemcpyHostToDevice );
       const int thrPerEvt = CPPProcess::npar * CPPProcess::np4; // AV: transpose alg does 1 element per thread (NOT 1 event per thread)
       //const int thrPerEvt = 1; // AV: try new alg with 1 event per thread... this seems slower
-      dev_transposeMomentaF2C<<<m_gpublocks * thrPerEvt, m_gputhreads>>>( m_devMomentaF.data(), m_devMomentaC.data(), m_nevt );
+      gpuLaunchKernel( dev_transposeMomentaF2C, m_gpublocks * thrPerEvt, m_gputhreads, m_devMomentaF.data(), m_devMomentaC.data(), m_nevt );
     }
     if constexpr( std::is_same_v<FORTRANFPTYPE, fptype> )
     {
@@ -341,7 +356,7 @@ namespace mg5amcCpu
   }
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::cpu_sequence( const FORTRANFPTYPE* momenta,
                                             const FORTRANFPTYPE* gs,
@@ -396,7 +411,7 @@ namespace mg5amcCpu
   // - C++ array: momenta[npagM][npar][np4][neppM] with nevt=npagM*neppM (AOSOA)
   //
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename Tin, typename Tout>
   __global__ void dev_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt )
   {
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/BridgeKernels.cc b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/BridgeKernels.cc
index d58066c9c1..eaf4037a24 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/BridgeKernels.cc
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/BridgeKernels.cc
@@ -1,17 +1,18 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "BridgeKernels.h"
 
+#include "GpuAbstraction.h"
 #include "MemoryAccessMomenta.h"
 
 #include <sstream>
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -45,7 +46,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu
 {
 
@@ -96,7 +97,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/BridgeKernels.h b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/BridgeKernels.h
index 15eb4bff4d..3efef8ce97 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/BridgeKernels.h
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/BridgeKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef BRIDGEKERNELS_H
 #define BRIDGEKERNELS_H 1
@@ -12,7 +12,7 @@
 #include "MatrixElementKernels.h"
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -49,7 +49,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A Bridge wrapper class encapsulating matrix element calculations on a CPU host
   class BridgeKernelHost final : public BridgeKernelBase
   {
@@ -89,7 +89,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A Bridge wrapper class encapsulating matrix element calculations on a GPU device
   class BridgeKernelDevice : public BridgeKernelBase
   {
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/CommonRandomNumberKernel.cc b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/CommonRandomNumberKernel.cc
index 985b39f576..010bc4cbd0 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/CommonRandomNumberKernel.cc
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/CommonRandomNumberKernel.cc
@@ -1,15 +1,16 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "CommonRandomNumbers.h"
+#include "GpuAbstraction.h"
 #include "MemoryBuffers.h"
 #include "RandomNumberKernels.h"
 
 #include <cassert>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/CrossSectionKernels.cc b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/CrossSectionKernels.cc
index 0b355a3c8d..c15b39844d 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/CrossSectionKernels.cc
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/CrossSectionKernels.cc
@@ -1,10 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "CrossSectionKernels.h"
 
+#include "GpuAbstraction.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessWeights.h"
 #include "MemoryBuffers.h"
@@ -77,7 +78,7 @@ debug_me_is_abnormal( const fptype& me, size_t ievtALL )
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -185,7 +186,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/CrossSectionKernels.h b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/CrossSectionKernels.h
index 7933ca4bbf..4d9659e04e 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/CrossSectionKernels.h
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/CrossSectionKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef CROSSSECTIONKERNELS_H
 #define CROSSSECTIONKERNELS_H 1
@@ -13,7 +13,7 @@
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -96,7 +96,7 @@ namespace mg5amcCpu
   //--------------------------------------------------------------------------
 
   /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating the calculation of event statistics on a GPU device
   class CrossSectionKernelDevice : public CrossSectionKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/CudaRuntime.h b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/CudaRuntime.h
deleted file mode 100644
index 64ce52f4b3..0000000000
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/CudaRuntime.h
+++ /dev/null
@@ -1,85 +0,0 @@
-// Copyright (C) 2020-2023 CERN and UCLouvain.
-// Licensed under the GNU Lesser General Public License (version 3 or later).
-// Created by: S. Roiser (Jul 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
-
-#ifndef MG5AMC_CUDARUNTIME_H
-#define MG5AMC_CUDARUNTIME_H 1
-
-// MG5AMC on GPU uses the CUDA runtime API, not the lower level CUDA driver API
-// See https://docs.nvidia.com/cuda/cuda-runtime-api/driver-vs-runtime-api.html#driver-vs-runtime-api
-
-#include <cassert>
-#include <iostream>
-
-//--------------------------------------------------------------------------
-
-// See https://stackoverflow.com/a/14038590
-#ifdef __CUDACC__ /* clang-format off */
-#define checkCuda( code ) { assertCuda( code, __FILE__, __LINE__ ); }
-inline void assertCuda( cudaError_t code, const char* file, int line, bool abort = true )
-{
-  if( code != cudaSuccess )
-  {
-    printf( "ERROR! assertCuda: '%s' (%d) in %s:%d\n", cudaGetErrorString( code ), code, file, line );
-    if( abort ) assert( code == cudaSuccess );
-  }
-}
-#endif /* clang-format on */
-
-//--------------------------------------------------------------------------
-
-#ifdef __CUDACC__
-namespace mg5amcGpu
-{
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  // *** FIXME! This will all need to be designed differently when going to multi-GPU nodes! ***
-  struct CudaRuntime final
-  {
-    CudaRuntime( const bool debug = true )
-      : m_debug( debug ) { setUp( m_debug ); }
-    ~CudaRuntime() { tearDown( m_debug ); }
-    CudaRuntime( const CudaRuntime& ) = delete;
-    CudaRuntime( CudaRuntime&& ) = delete;
-    CudaRuntime& operator=( const CudaRuntime& ) = delete;
-    CudaRuntime& operator=( CudaRuntime&& ) = delete;
-    bool m_debug;
-
-    // Set up CUDA application
-    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
-    // Calling cudaSetDevice on startup is useful to properly book-keep the time spent in CUDA initialization
-    static void setUp( const bool debug = true )
-    {
-      // ** NB: it is useful to call cudaSetDevice, or cudaFree, to properly book-keep the time spent in CUDA initialization
-      // ** NB: otherwise, the first CUDA operation (eg a cudaMemcpyToSymbol in CPPProcess ctor) appears to take much longer!
-      /*
-      // [We initially added cudaFree(0) to "ease profile analysis" only because it shows up as a big recognizable block!]
-      // No explicit initialization is needed: https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#initialization
-      // It is not clear what cudaFree(0) does at all: https://stackoverflow.com/questions/69967813/
-      if ( debug ) std::cout << "__CudaRuntime: calling cudaFree(0)" << std::endl;
-      checkCuda( cudaFree( 0 ) ); // SLOW!
-      */
-      // Replace cudaFree(0) by cudaSetDevice(0), even if it is not really needed either
-      // (but see https://developer.nvidia.com/blog/cuda-pro-tip-always-set-current-device-avoid-multithreading-bugs)
-      if( debug ) std::cout << "__CudaRuntime: calling cudaSetDevice(0)" << std::endl;
-      checkCuda( cudaSetDevice( 0 ) ); // SLOW!
-    }
-
-    // Tear down CUDA application (call cudaDeviceReset)
-    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
-    // Calling cudaDeviceReset on shutdown is only needed for checking memory leaks in cuda-memcheck
-    // See https://docs.nvidia.com/cuda/cuda-memcheck/index.html#leak-checking
-    static void tearDown( const bool debug = true )
-    {
-      if( debug ) std::cout << "__CudaRuntime: calling cudaDeviceReset()" << std::endl;
-      checkCuda( cudaDeviceReset() );
-    }
-  };
-
-}
-#endif
-
-//--------------------------------------------------------------------------
-
-#endif // MG5AMC_CUDARUNTIME_H
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/CurandRandomNumberKernel.cc b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/CurandRandomNumberKernel.cc
index eb56333b03..08a16f6f2c 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/CurandRandomNumberKernel.cc
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/CurandRandomNumberKernel.cc
@@ -1,9 +1,9 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "MemoryBuffers.h"
 #include "RandomNumberKernels.h"
 
@@ -22,7 +22,7 @@ inline void assertCurand( curandStatus_t code, const char *file, int line, bool
 }
 #endif /* clang-format on */
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -36,7 +36,7 @@ namespace mg5amcCpu
   {
     if( m_isOnDevice )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if( !m_rnarray.isOnDevice() )
         throw std::runtime_error( "CurandRandomNumberKernel on device with a host random number array" );
 #else
@@ -114,7 +114,7 @@ namespace mg5amcCpu
     /*
     printf( "\nCurandRandomNumberKernel::generateRnarray size = %d\n", (int)m_rnarray.size() );
     fptype* data = m_rnarray.data();
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( m_rnarray.isOnDevice() )
     {
       data = new fptype[m_rnarray.size()]();
@@ -123,7 +123,7 @@ namespace mg5amcCpu
 #endif
     for( int i = 0; i < ( (int)m_rnarray.size() / 4 ); i++ )
       printf( "[%4d] %f %f %f %f\n", i * 4, data[i * 4], data[i * 4 + 2], data[i * 4 + 2], data[i * 4 + 3] );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( m_rnarray.isOnDevice() ) delete[] data;
 #endif
     */
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/EventStatistics.h b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/EventStatistics.h
index 48b51e0a49..b425a5bade 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/EventStatistics.h
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/EventStatistics.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef EventStatistics_H
 #define EventStatistics_H 1
@@ -16,7 +16,7 @@
 #include <limits>
 #include <string>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/GpuAbstraction.h b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/GpuAbstraction.h
new file mode 100644
index 0000000000..9c467b1e04
--- /dev/null
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/GpuAbstraction.h
@@ -0,0 +1,69 @@
+// Copyright (C) 2020-2023 CERN and UCLouvain.
+// Licensed under the GNU Lesser General Public License (version 3 or later).
+// Created by: J. Teig (Jul 2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+
+#ifndef MG5AMC_GPUABSTRACTION_H
+#define MG5AMC_GPUABSTRACTION_H 1
+
+#include <cassert>
+
+//--------------------------------------------------------------------------
+
+#ifdef __CUDACC__
+
+#define gpuError_t cudaError_t
+#define gpuPeekAtLastError cudaPeekAtLastError
+#define gpuGetErrorString cudaGetErrorString
+#define gpuSuccess cudaSuccess
+
+#define gpuMallocHost( ptr, size ) checkGpu( cudaMallocHost( ptr, size ) )
+#define gpuMalloc( ptr, size ) checkGpu( cudaMalloc( ptr, size ) )
+
+#define gpuMemcpy( dstData, srcData, srcBytes, func ) checkGpu( cudaMemcpy( dstData, srcData, srcBytes, func ) )
+#define gpuMemcpyHostToDevice cudaMemcpyHostToDevice
+#define gpuMemcpyDeviceToHost cudaMemcpyDeviceToHost
+#define gpuMemcpyToSymbol( type1, type2, size ) checkGpu( cudaMemcpyToSymbol( type1, type2, size ) )
+
+#define gpuFree( ptr ) checkGpu( cudaFree( ptr ) )
+#define gpuFreeHost( ptr ) checkGpu( cudaFreeHost( ptr ) )
+
+#define gpuSetDevice cudaSetDevice
+#define gpuDeviceSynchronize cudaDeviceSynchronize
+#define gpuDeviceReset cudaDeviceReset
+
+#define gpuLaunchKernel( kernel, blocks, threads, ... ) kernel<<<blocks, threads>>>( __VA_ARGS__ )
+#define gpuLaunchKernelSharedMem( kernel, blocks, threads, sharedMem, ... ) kernel<<<blocks, threads, sharedMem>>>( __VA_ARGS__ )
+
+//--------------------------------------------------------------------------
+
+#elif defined __HIPCC__
+
+#define gpuError_t hipError_t
+#define gpuPeekAtLastError hipPeekAtLastError
+#define gpuGetErrorString hipGetErrorString
+#define gpuSuccess hipSuccess
+
+#define gpuMallocHost( ptr, size ) checkGpu( hipHostMalloc( ptr, size ) ) // HostMalloc better
+#define gpuMalloc( ptr, size ) checkGpu( hipMalloc( ptr, size ) )
+
+#define gpuMemcpy( dstData, srcData, srcBytes, func ) checkGpu( hipMemcpy( dstData, srcData, srcBytes, func ) )
+#define gpuMemcpyHostToDevice hipMemcpyHostToDevice
+#define gpuMemcpyDeviceToHost hipMemcpyDeviceToHost
+#define gpuMemcpyToSymbol( type1, type2, size ) checkGpu( hipMemcpyToSymbol( type1, type2, size ) )
+
+#define gpuFree( ptr ) checkGpu( hipFree( ptr ) )
+#define gpuFreeHost( ptr ) checkGpu( hipHostFree( ptr ) )
+
+#define gpuSetDevice hipSetDevice
+#define gpuDeviceSynchronize hipDeviceSynchronize
+#define gpuDeviceReset hipDeviceReset
+
+#define gpuLaunchKernel( kernel, blocks, threads, ... ) kernel<<<blocks, threads>>>( __VA_ARGS__ )
+#define gpuLaunchKernelSharedMem( kernel, blocks, threads, sharedMem, ... ) kernel<<<blocks, threads, sharedMem>>>( __VA_ARGS__ )
+
+//--------------------------------------------------------------------------
+
+#endif
+
+#endif // MG5AMC_GPUABSTRACTION_H
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/GpuRuntime.h b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/GpuRuntime.h
new file mode 100644
index 0000000000..93579ef08b
--- /dev/null
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/GpuRuntime.h
@@ -0,0 +1,85 @@
+// Copyright (C) 2020-2023 CERN and UCLouvain.
+// Licensed under the GNU Lesser General Public License (version 3 or later).
+// Created by: J. Teig (Jun 2023, based on earlier work by S. Roiser) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+
+#ifndef MG5AMC_GPURUNTIME_H
+#define MG5AMC_GPURUNTIME_H 1
+
+// MG5AMC on GPU uses the CUDA runtime API, not the lower level CUDA driver API
+// See https://docs.nvidia.com/cuda/cuda-runtime-api/driver-vs-runtime-api.html#driver-vs-runtime-api
+
+#include "GpuAbstraction.h"
+
+#include <iostream>
+
+//--------------------------------------------------------------------------
+
+// See https://stackoverflow.com/a/14038590
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
+#define checkGpu( code ) { assertGpu( code, __FILE__, __LINE__ ); }
+inline void assertGpu( gpuError_t code, const char* file, int line, bool abort = true )
+{
+  if( code != gpuSuccess )
+  {
+    printf( "ERROR! assertGpu: '%s' (%d) in %s:%d\n", gpuGetErrorString( code ), code, file, line );
+    if( abort ) assert( code == gpuSuccess );
+  }
+}
+#endif /* clang-format on */
+
+//--------------------------------------------------------------------------
+
+#ifdef MGONGPUCPP_GPUIMPL
+namespace mg5amcGpu
+{
+  // Instantiate a GpuRuntime at the beginnining of the application's main to
+  // invoke gpuSetDevice(0) in the constructor and book a gpuDeviceReset() call in the destructor
+  // *** FIXME! This will all need to be designed differently when going to multi-GPU nodes! ***
+  struct GpuRuntime final
+  {
+    GpuRuntime( const bool debug = true )
+      : m_debug( debug ) { setUp( m_debug ); }
+    ~GpuRuntime() { tearDown( m_debug ); }
+    GpuRuntime( const GpuRuntime& ) = delete;
+    GpuRuntime( GpuRuntime&& ) = delete;
+    GpuRuntime& operator=( const GpuRuntime& ) = delete;
+    GpuRuntime& operator=( GpuRuntime&& ) = delete;
+    bool m_debug;
+
+    // Set up CUDA application
+    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
+    // Calling cudaSetDevice on startup is useful to properly book-keep the time spent in CUDA initialization
+    static void setUp( const bool debug = true )
+    {
+      // ** NB: it is useful to call cudaSetDevice, or cudaFree, to properly book-keep the time spent in CUDA initialization
+      // ** NB: otherwise, the first CUDA operation (eg a cudaMemcpyToSymbol in CPPProcess ctor) appears to take much longer!
+      /*
+      // [We initially added cudaFree(0) to "ease profile analysis" only because it shows up as a big recognizable block!]
+      // No explicit initialization is needed: https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#initialization
+      // It is not clear what cudaFree(0) does at all: https://stackoverflow.com/questions/69967813/
+      if ( debug ) std::cout << "__CudaRuntime: calling cudaFree(0)" << std::endl;
+      checkCuda( cudaFree( 0 ) ); // SLOW!
+      */
+      // Replace cudaFree(0) by cudaSetDevice(0), even if it is not really needed either
+      // (but see https://developer.nvidia.com/blog/cuda-pro-tip-always-set-current-device-avoid-multithreading-bugs)
+      if( debug ) std::cout << "__GpuRuntime: calling GpuSetDevice(0)" << std::endl;
+      checkGpu( gpuSetDevice( 0 ) ); // SLOW!
+    }
+
+    // Tear down CUDA application (call cudaDeviceReset)
+    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
+    // Calling cudaDeviceReset on shutdown is only needed for checking memory leaks in cuda-memcheck
+    // See https://docs.nvidia.com/cuda/cuda-memcheck/index.html#leak-checking
+    static void tearDown( const bool debug = true )
+    {
+      if( debug ) std::cout << "__GpuRuntime: calling GpuDeviceReset()" << std::endl;
+      checkGpu( gpuDeviceReset() );
+    }
+  };
+}
+#endif
+
+//--------------------------------------------------------------------------
+
+#endif // MG5AMC_GPURUNTIME_H
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MadgraphTest.h b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MadgraphTest.h
index ef40624c88..6054185300 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MadgraphTest.h
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MadgraphTest.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Hageboeck (Dec 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MADGRAPHTEST_H_
 #define MADGRAPHTEST_H_ 1
@@ -14,7 +14,11 @@
 
 #include <array>
 #include <cmath>
-#include <filesystem>
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
 #include <fstream>
 #include <iomanip>
 #include <memory>
@@ -22,7 +26,7 @@
 #include <string>
 #include <vector>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using mg5amcGpu::CPPProcess;
 #else
 using mg5amcCpu::CPPProcess;
@@ -201,7 +205,7 @@ class MadgraphTest : public testing::TestWithParam<TestDriverBase*>
 
 // Since we link both the CPU-only and GPU tests into the same executable, we prevent
 // a multiply defined symbol by only compiling this in the non-CUDA phase:
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
 /// Compare momenta and matrix elements.
 /// This uses an implementation of TestDriverBase to run a madgraph workflow,
@@ -219,7 +223,14 @@ TEST_P( MadgraphTest, CompareMomentaAndME )
   const char* dumpEventsC = getenv( "CUDACPP_RUNTEST_DUMPEVENTS" );
   const bool dumpEvents = ( dumpEventsC != 0 ) && ( std::string( dumpEventsC ) != "" );
   const std::string refFileName = testDriver->getRefFileName();
+  /*
+#ifdef __HIPCC__
+  const std::string dumpFileName = std::experimental::filesystem::path( refFileName ).filename();
+#else
   const std::string dumpFileName = std::filesystem::path( refFileName ).filename();
+#endif
+  */
+  const std::string dumpFileName = refFileName; // bypass std::filesystem #803
   std::ofstream dumpFile;
   if( dumpEvents )
   {
@@ -307,6 +318,6 @@ TEST_P( MadgraphTest, CompareMomentaAndME )
   }
 }
 
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
 
 #endif /* MADGRAPHTEST_H_ */
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MatrixElementKernels.cc b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MatrixElementKernels.cc
index 74b5239ebf..81699dfea9 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MatrixElementKernels.cc
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MatrixElementKernels.cc
@@ -1,12 +1,12 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "MatrixElementKernels.h"
 
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h" // Includes the abstraction for Nvidia/AMD compilation
 #include "MemoryAccessMomenta.h"
 #include "MemoryBuffers.h"
 
@@ -14,7 +14,7 @@
 
 //============================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu
 {
 
@@ -150,7 +150,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
@@ -209,13 +209,13 @@ namespace mg5amcGpu
     PinnedHostBufferHelicityMask hstIsGoodHel( ncomb );
     DeviceBufferHelicityMask devIsGoodHel( ncomb );
     // ... 0d1. Compute good helicity mask on the device
-    computeDependentCouplings<<<m_gpublocks, m_gputhreads>>>( m_gs.data(), m_couplings.data() );
+    gpuLaunchKernel( computeDependentCouplings, m_gpublocks, m_gputhreads, m_gs.data(), m_couplings.data() );
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-    sigmaKin_getGoodHel<<<m_gpublocks, m_gputhreads>>>( m_momenta.data(), m_couplings.data(), m_matrixElements.data(), m_numerators.data(), m_denominators.data(), devIsGoodHel.data() );
+    gpuLaunchKernel( sigmaKin_getGoodHel, m_gpublocks, m_gputhreads, m_momenta.data(), m_couplings.data(), m_matrixElements.data(), m_numerators.data(), m_denominators.data(), devIsGoodHel.data() );
 #else
-    sigmaKin_getGoodHel<<<m_gpublocks, m_gputhreads>>>( m_momenta.data(), m_couplings.data(), m_matrixElements.data(), devIsGoodHel.data() );
+    gpuLaunchKernel( sigmaKin_getGoodHel, m_gpublocks, m_gputhreads, m_momenta.data(), m_couplings.data(), m_matrixElements.data(), devIsGoodHel.data() );
 #endif
-    checkCuda( cudaPeekAtLastError() );
+    checkGpu( gpuPeekAtLastError() );
     // ... 0d2. Copy back good helicity mask to the host
     copyHostFromDevice( hstIsGoodHel, devIsGoodHel );
     // ... 0d3. Copy back good helicity list to constant memory on the device
@@ -226,19 +226,19 @@ namespace mg5amcGpu
 
   void MatrixElementKernelDevice::computeMatrixElements( const unsigned int channelId )
   {
-    computeDependentCouplings<<<m_gpublocks, m_gputhreads>>>( m_gs.data(), m_couplings.data() );
+    gpuLaunchKernel( computeDependentCouplings, m_gpublocks, m_gputhreads, m_gs.data(), m_couplings.data() );
 #ifndef MGONGPU_NSIGHT_DEBUG
     constexpr unsigned int sharedMemSize = 0;
 #else
     constexpr unsigned int sharedMemSize = ntpbMAX * sizeof( float );
 #endif
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-    sigmaKin<<<m_gpublocks, m_gputhreads, sharedMemSize>>>( m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), channelId, m_numerators.data(), m_denominators.data(), m_selhel.data(), m_selcol.data() );
+    gpuLaunchKernelSharedMem( sigmaKin, m_gpublocks, m_gputhreads, sharedMemSize, m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), channelId, m_numerators.data(), m_denominators.data(), m_selhel.data(), m_selcol.data() );
 #else
-    sigmaKin<<<m_gpublocks, m_gputhreads, sharedMemSize>>>( m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), m_selhel.data(), m_selcol.data() );
+    gpuLaunchKernelSharedMem( sigmaKin, m_gpublocks, m_gputhreads, sharedMemSize, m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), m_selhel.data(), m_selcol.data() );
 #endif
-    checkCuda( cudaPeekAtLastError() );
-    checkCuda( cudaDeviceSynchronize() );
+    checkGpu( gpuPeekAtLastError() );
+    checkGpu( gpuDeviceSynchronize() );
   }
 
   //--------------------------------------------------------------------------
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MatrixElementKernels.h b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MatrixElementKernels.h
index 23e84757a2..72bd8f195b 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MatrixElementKernels.h
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MatrixElementKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MATRIXELEMENTKERNELS_H
 #define MATRIXELEMENTKERNELS_H 1
@@ -10,7 +10,7 @@
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -81,7 +81,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating matrix element calculations on a CPU host
   class MatrixElementKernelHost final : public MatrixElementKernelBase, public NumberOfEvents
   {
@@ -130,7 +130,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating matrix element calculations on a GPU device
   class MatrixElementKernelDevice : public MatrixElementKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessAmplitudes.h b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessAmplitudes.h
index 573b3bbbc9..ffb76e93de 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessAmplitudes.h
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessAmplitudes.h
@@ -15,7 +15,7 @@
 #define MGONGPU_TRIVIAL_AMPLITUDES 1
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessCouplings.h b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessCouplings.h
index 35a3af42e0..3afdf3e554 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessCouplings.h
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessCouplings.h
@@ -15,7 +15,7 @@
 #include "MemoryBuffers.h"       // for HostBufferCouplings::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessCouplingsFixed.h b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessCouplingsFixed.h
index dc0d93afff..ffcdf4dbef 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessCouplingsFixed.h
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessCouplingsFixed.h
@@ -14,7 +14,7 @@
 //#include "MemoryAccessHelpers.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessDenominators.h b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessDenominators.h
index 3bce635718..66f2d32a6b 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessDenominators.h
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessDenominators.h
@@ -10,7 +10,7 @@
 #include "MemoryAccessGs.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessGs.h b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessGs.h
index 31311aa375..4c726b30f3 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessGs.h
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessGs.h
@@ -13,7 +13,7 @@
 #include "MemoryBuffers.h" // for HostBufferMatrixElements::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessHelpers.h b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessHelpers.h
index c82a6c7635..db73e4e064 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessHelpers.h
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessHelpers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessHelpers_H
 #define MemoryAccessHelpers_H 1
@@ -105,7 +105,7 @@ class KernelAccessHelper : public MemoryAccessHelper<T>
     }
     else
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
       //printf( "kernelAccessRecord: ievt=%d threadId=%d\n", ievt, threadIdx.x );
       return T::ieventAccessRecord( buffer, ievt ); // NB fptype and fptype_sv coincide for CUDA
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessMatrixElements.h b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessMatrixElements.h
index f32e6fea5b..3741011971 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessMatrixElements.h
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessMatrixElements.h
@@ -13,7 +13,7 @@
 #include "MemoryBuffers.h" // for HostBufferMatrixElements::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessMomenta.h b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessMomenta.h
index 29266de32c..3be229d392 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessMomenta.h
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessMomenta.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessMomenta_H
 #define MemoryAccessMomenta_H 1
@@ -13,7 +13,7 @@
 #include "MemoryAccessVectors.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -30,7 +30,7 @@ namespace mg5amcCpu
 
     // Number of Events Per Page in the momenta AOSOA memory buffer layout
     // (these are all best kept as a compile-time constants: see issue #23)
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
     // -----------------------------------------------------------------------------------------------
     // --- GPUs: neppM is best set to a power of 2 times the number of fptype's in a 32-byte cacheline
     // --- This is relevant to ensure coalesced access to momenta in global memory
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessNumerators.h b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessNumerators.h
index b152183b28..18991f4fa6 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessNumerators.h
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessNumerators.h
@@ -10,7 +10,7 @@
 #include "MemoryAccessGs.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessRandomNumbers.h b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessRandomNumbers.h
index e2988d39f3..40cb089135 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessRandomNumbers.h
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessRandomNumbers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessRandomNumbers_H
 #define MemoryAccessRandomNumbers_H 1
@@ -11,7 +11,7 @@
 #include "CPPProcess.h"
 #include "MemoryAccessHelpers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using mg5amcGpu::CPPProcess;
 #else
 using mg5amcCpu::CPPProcess;
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessVectors.h b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessVectors.h
index e9b197368e..08faccff0f 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessVectors.h
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessVectors.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessVectors_H
 #define MemoryAccessVectors_H 1
@@ -10,7 +10,7 @@
 
 #include "mgOnGpuVectors.h"
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu // this is only needed for CPU SIMD vectorization
 {
 
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessWavefunctions.h b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessWavefunctions.h
index 5428aaf933..33bef4559e 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessWavefunctions.h
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryAccessWavefunctions.h
@@ -15,7 +15,7 @@
 #define MGONGPU_TRIVIAL_WAVEFUNCTIONS 1
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryBuffers.h b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryBuffers.h
index 3093e6ed18..7756a71621 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryBuffers.h
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/MemoryBuffers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021, based on earlier work by S. Hageboeck) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryBuffers_H
 #define MemoryBuffers_H 1
@@ -11,12 +11,12 @@
 #include "mgOnGpuCxtypes.h"
 
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "Parameters_sm.h"
 
 #include <sstream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -87,7 +87,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   constexpr bool HostBufferALIGNED = false;   // ismisaligned=false
   constexpr bool HostBufferMISALIGNED = true; // ismisaligned=true
 
@@ -119,7 +119,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA pinned host buffer
   template<typename T>
   class PinnedHostBufferBase : public BufferBase<T>
@@ -128,18 +128,18 @@ namespace mg5amcCpu
     PinnedHostBufferBase( const size_t size )
       : BufferBase<T>( size, false )
     {
-      checkCuda( cudaMallocHost( &( this->m_data ), this->bytes() ) );
+      gpuMallocHost( &( this->m_data ), this->bytes() );
     }
     virtual ~PinnedHostBufferBase()
     {
-      checkCuda( cudaFreeHost( this->m_data ) );
+      gpuFreeHost( this->m_data );
     }
   };
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA device buffer
   template<typename T>
   class DeviceBufferBase : public BufferBase<T>
@@ -148,18 +148,18 @@ namespace mg5amcCpu
     DeviceBufferBase( const size_t size )
       : BufferBase<T>( size, true )
     {
-      checkCuda( cudaMalloc( &( this->m_data ), this->bytes() ) );
+      gpuMalloc( &( this->m_data ), this->bytes() );
     }
     virtual ~DeviceBufferBase()
     {
-      checkCuda( cudaFree( this->m_data ) );
+      gpuFree( this->m_data );
     }
   };
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for a given number of events
   template<typename T, size_t sizePerEvent, bool ismisaligned>
   class HostBuffer : public HostBufferBase<T, ismisaligned>, virtual private NumberOfEvents
@@ -175,7 +175,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA pinned host buffer for a given number of events
   template<typename T, size_t sizePerEvent>
   class PinnedHostBuffer : public PinnedHostBufferBase<T>, virtual private NumberOfEvents
@@ -191,7 +191,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA device buffer for a given number of events
   template<typename T, size_t sizePerEvent>
   class DeviceBuffer : public DeviceBufferBase<T>, virtual private NumberOfEvents
@@ -213,7 +213,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for momenta random numbers
   constexpr size_t sizePerEventRndNumMomenta = MemoryBuffers::np4 * MemoryBuffers::nparf;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for momenta random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumMomenta, HostBufferALIGNED> HostBufferRndNumMomenta;
 #else
@@ -232,7 +232,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer with ONE fptype per event
   constexpr size_t sizePerEventOneFp = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer with ONE fptype per event
   typedef HostBuffer<fptype, sizePerEventOneFp, HostBufferALIGNED> HostBufferOneFp;
 #else
@@ -257,7 +257,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for Gs
   constexpr size_t sizePerEventGs = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventGs, HostBufferALIGNED> HostBufferGs;
 #else
@@ -276,7 +276,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for numerators
   constexpr size_t sizePerEventNumerators = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventNumerators, HostBufferALIGNED> HostBufferNumerators;
 #else
@@ -296,7 +296,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for denominators
   constexpr size_t sizePerEventDenominators = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventDenominators, HostBufferALIGNED> HostBufferDenominators;
 #else
@@ -315,7 +315,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for random numbers
   constexpr size_t sizePerEventCouplings = MemoryBuffers::ndcoup * MemoryBuffers::nx2;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventCouplings, HostBufferALIGNED> HostBufferCouplings;
 #else
@@ -333,7 +333,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for momenta
   constexpr size_t sizePerEventMomenta = MemoryBuffers::np4 * MemoryBuffers::npar;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for momenta
   typedef HostBuffer<fptype, sizePerEventMomenta, HostBufferALIGNED> HostBufferMomenta;
   //typedef HostBuffer<fptype, sizePerEventMomenta, HostBufferMISALIGNED> HostBufferMomenta; // TEST MISALIGNMENT!
@@ -352,7 +352,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for sampling weights
   constexpr size_t sizePerEventWeights = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for sampling weights
   typedef HostBuffer<fptype, sizePerEventWeights, HostBufferALIGNED> HostBufferWeights;
 #else
@@ -370,7 +370,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for matrix elements
   constexpr size_t sizePerEventMatrixElements = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for matrix elements
   typedef HostBuffer<fptype, sizePerEventMatrixElements, HostBufferALIGNED> HostBufferMatrixElements;
 #else
@@ -385,7 +385,7 @@ namespace mg5amcCpu
   // A base class encapsulating a memory buffer for the helicity mask
   typedef BufferBase<bool> BufferHelicityMask;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for the helicity mask
   typedef HostBufferBase<bool, HostBufferALIGNED> HostBufferHelicityMask;
 #else
@@ -403,7 +403,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for wavefunctions
   constexpr size_t sizePerEventWavefunctions = MemoryBuffers::nw6 * MemoryBuffers::nx2;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for wavefunctions
   typedef HostBuffer<fptype, sizePerEventWavefunctions, HostBufferALIGNED> HostBufferWavefunctions;
 #else
@@ -421,7 +421,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for helicity random numbers
   constexpr size_t sizePerEventRndNumHelicity = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for helicity random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumHelicity, HostBufferALIGNED> HostBufferRndNumHelicity;
 #else
@@ -439,7 +439,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for color random numbers
   constexpr size_t sizePerEventRndNumColor = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for color random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumColor, HostBufferALIGNED> HostBufferRndNumColor;
 #else
@@ -457,7 +457,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for helicity selection
   constexpr size_t sizePerEventSelectedHelicity = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for helicity selection
   typedef HostBuffer<int, sizePerEventSelectedHelicity, HostBufferALIGNED> HostBufferSelectedHelicity;
 #else
@@ -475,7 +475,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for color selection
   constexpr size_t sizePerEventSelectedColor = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for color selection
   typedef HostBuffer<int, sizePerEventSelectedColor, HostBufferALIGNED> HostBufferSelectedColor;
 #else
@@ -487,7 +487,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<class Tdst, class Tsrc>
   void copyDeviceFromHost( Tdst& dst, const Tsrc& src ) // keep the same order of arguments as in memcpy
   {
@@ -504,13 +504,13 @@ namespace mg5amcCpu
       throw std::runtime_error( sstr.str() );
     }
     // NB (PR #45): cudaMemcpy involves an intermediate memcpy to pinned memory if host array is a not a pinned host array
-    checkCuda( cudaMemcpy( dst.data(), src.data(), src.bytes(), cudaMemcpyHostToDevice ) );
+    gpuMemcpy( dst.data(), src.data(), src.bytes(), gpuMemcpyHostToDevice );
   }
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<class Tdst, class Tsrc>
   void copyHostFromDevice( Tdst& dst, const Tsrc& src ) // keep the same order of arguments as in memcpy
   {
@@ -527,7 +527,7 @@ namespace mg5amcCpu
       throw std::runtime_error( sstr.str() );
     }
     // NB (PR #45): cudaMemcpy involves an intermediate memcpy to pinned memory if host array is a not a pinned host array
-    checkCuda( cudaMemcpy( dst.data(), src.data(), src.bytes(), cudaMemcpyDeviceToHost ) );
+    gpuMemcpy( dst.data(), src.data(), src.bytes(), gpuMemcpyDeviceToHost );
   }
 #endif
 
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/P1_Sigma_sm_gg_ttxggg/CPPProcess.cc b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/P1_Sigma_sm_gg_ttxggg/CPPProcess.cc
index c2f8607428..ff20b7ba63 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/P1_Sigma_sm_gg_ttxggg/CPPProcess.cc
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/P1_Sigma_sm_gg_ttxggg/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_sm.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -46,7 +45,7 @@
 // Class member functions for calculating the matrix elements for
 // Process: g g > t t~ g g g WEIGHTED<=5 @1
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -80,7 +79,7 @@ namespace mg5amcCpu
   __device__ const fptype cIPD[2] = { (fptype)Parameters_sm::mdl_MT, (fptype)Parameters_sm::mdl_WT };
   __device__ const fptype* cIPC = nullptr; // unused as nicoup=0
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ fptype cIPD[2];
   __device__ __constant__ fptype* cIPC = nullptr; // unused as nicoup=0
 #else
@@ -90,7 +89,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -118,13 +117,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -151,7 +150,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -187,7 +186,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -200,8 +199,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -31908,7 +31909,7 @@ namespace mg5amcCpu
         { -116, 442, 442, -134, -134, 505, -44, -134, 28, -224, -62, 496, -53, 19, -62, 496, 10, -80, -62, -71, 10, -80, 1, -8, 136, -116, -116, -44, -44, 514, -116, 442, -44, 28, -53, -62, -44, -53, 514, -62, 100, 10, 28, -62, -62, 10, 10, 1, 514, 505, -62, 496, -71, 568, -44, -134, -53, -62, 19, -71, 10, -80, 100, 10, 640, -80, 1, -8, 10, 1, 64, -8, -62, -71, 10, -80, 1, -8, 28, -62, -62, 10, 10, 1, 1, -8, 10, 1, 64, -8, -8, 64, -80, -8, -512, 64, 496, 568, -80, 640, -8, 64, -224, 496, 496, -80, -80, -8, -8, 64, -80, -8, -512, 64, 64, -512, 640, 64, 4096, -512 },
         { 136, -116, -116, -44, -44, 514, -116, 442, -44, 28, -53, -62, -44, -53, 514, -62, 100, 10, 28, -62, -62, 10, 10, 1, -116, 442, 442, -134, -134, 505, -44, -134, 28, -224, -62, 496, -53, 19, -62, 496, 10, -80, -62, -71, 10, -80, 1, -8, -44, -134, -53, -62, 19, -71, 514, 505, -62, 496, -71, 568, 100, 10, 10, -80, -80, 640, 10, 1, 1, -8, -8, 64, 28, -62, -62, 10, 10, 1, -62, -71, 10, -80, 1, -8, 10, 1, 1, -8, -8, 64, -80, -8, -8, 64, 64, -512, -224, 496, 496, -80, -80, -8, 496, 568, -80, 640, -8, 64, -80, -8, -8, 64, 64, -512, 640, 64, 64, -512, -512, 4096 } }; // 2-D array[120][120]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -31965,7 +31966,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -32024,7 +32025,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -32183,8 +32184,8 @@ namespace mg5amcCpu
       { 1, 1, 1, -1, 1, -1, 1 },
       { 1, 1, 1, -1, 1, 1, -1 },
       { 1, 1, 1, -1, 1, 1, 1 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -32227,9 +32228,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     const fptype tIPD[2] = { (fptype)m_pars->mdl_MT, (fptype)m_pars->mdl_WT };
     //const cxtype tIPC[0] = { ... }; // nicoup=0
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) ) );
-    //checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ) ); // nicoup=0
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) );
+    //gpuMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
 #else
     memcpy( cIPD, tIPD, 2 * sizeof( fptype ) );
     //memcpy( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
@@ -32268,7 +32269,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -32297,6 +32298,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -32310,6 +32315,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -32333,12 +32340,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -32359,7 +32366,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -32485,9 +32492,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -32511,7 +32518,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -32531,7 +32538,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 1536 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -32545,9 +32552,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -32575,7 +32585,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -32785,7 +32795,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/P1_Sigma_sm_gg_ttxggg/CPPProcess.h b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/P1_Sigma_sm_gg_ttxggg/CPPProcess.h
index 2565923dde..fff95b66e2 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/P1_Sigma_sm_gg_ttxggg/CPPProcess.h
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/P1_Sigma_sm_gg_ttxggg/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -107,7 +107,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -120,7 +120,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -150,7 +150,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/P1_Sigma_sm_gg_ttxggg/CudaRuntime.h b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/P1_Sigma_sm_gg_ttxggg/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/P1_Sigma_sm_gg_ttxggg/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/P1_Sigma_sm_gg_ttxggg/GpuAbstraction.h b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/P1_Sigma_sm_gg_ttxggg/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/P1_Sigma_sm_gg_ttxggg/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/P1_Sigma_sm_gg_ttxggg/GpuRuntime.h b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/P1_Sigma_sm_gg_ttxggg/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/P1_Sigma_sm_gg_ttxggg/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/P1_Sigma_sm_gg_ttxggg/check_sa.cc b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/P1_Sigma_sm_gg_ttxggg/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/P1_Sigma_sm_gg_ttxggg/check_sa.cc
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/P1_Sigma_sm_gg_ttxggg/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/RamboSamplingKernels.cc b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/RamboSamplingKernels.cc
index da68aa9255..79abbcc4f8 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/RamboSamplingKernels.cc
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/RamboSamplingKernels.cc
@@ -1,11 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "RamboSamplingKernels.h"
 
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "MemoryAccessMomenta.h"
 #include "MemoryAccessRandomNumbers.h"
 #include "MemoryAccessWeights.h"
@@ -14,7 +14,7 @@
 
 #include <sstream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -92,7 +92,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingKernelDevice::RamboSamplingKernelDevice( const fptype energy,               // input: energy
                                                         const BufferRndNumMomenta& rndmom, // input: random numbers in [0,1]
                                                         BufferMomenta& momenta,            // output: momenta
@@ -135,7 +135,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   getMomentaInitialDevice( const fptype energy,
                            fptype* momenta )
@@ -147,17 +147,17 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   void
   RamboSamplingKernelDevice::getMomentaInitial()
   {
-    getMomentaInitialDevice<<<m_gpublocks, m_gputhreads>>>( m_energy, m_momenta.data() );
+    gpuLaunchKernel( getMomentaInitialDevice, m_gpublocks, m_gputhreads, m_energy, m_momenta.data() );
   }
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   getMomentaFinalDevice( const fptype energy,
                          const fptype* rndmom,
@@ -171,11 +171,11 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   void
   RamboSamplingKernelDevice::getMomentaFinal()
   {
-    getMomentaFinalDevice<<<m_gpublocks, m_gputhreads>>>( m_energy, m_rndmom.data(), m_momenta.data(), m_weights.data() );
+    gpuLaunchKernel( getMomentaFinalDevice, m_gpublocks, m_gputhreads, m_energy, m_rndmom.data(), m_momenta.data(), m_weights.data() );
   }
 #endif
 
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/RamboSamplingKernels.h b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/RamboSamplingKernels.h
index 184089efd7..7c214cd74b 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/RamboSamplingKernels.h
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/RamboSamplingKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef RAMBOSAMPLINGKERNELS_H
 #define RAMBOSAMPLINGKERNELS_H 1
@@ -10,7 +10,7 @@
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -93,7 +93,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating RAMBO phase space sampling on a GPU device
   class RamboSamplingKernelDevice final : public SamplingKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/RandomNumberKernels.h b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/RandomNumberKernels.h
index 188a72c2c9..21d63beeac 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/RandomNumberKernels.h
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/RandomNumberKernels.h
@@ -1,14 +1,14 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef RANDOMNUMBERKERNELS_H
 #define RANDOMNUMBERKERNELS_H 1
 
 #include "mgOnGpuConfig.h"
 
-// NB This must come AFTER mgOnGpuConfig.h which contains our definition of __global__ when __CUDACC__ is not defined
+// NB This must come AFTER mgOnGpuConfig.h which contains our definition of __global__ when MGONGPUCPP_GPUIMPL is not defined
 #ifndef MGONGPU_HAS_NO_CURAND
 //#include "curand.h"
 struct curandGenerator_st; // forward definition from curand.h
@@ -16,7 +16,7 @@ struct curandGenerator_st; // forward definition from curand.h
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/cudacpp.mk b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/cudacpp.mk
index 509307506b..df74dfc284 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/cudacpp.mk
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/cudacpp.mk
@@ -1,7 +1,7 @@
 # Copyright (C) 2020-2023 CERN and UCLouvain.
 # Licensed under the GNU Lesser General Public License (version 3 or later).
 # Created by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-# Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+# Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #=== Determine the name of this makefile (https://ftp.gnu.org/old-gnu/Manuals/make-3.80/html_node/make_17.html)
 #=== NB: use ':=' to ensure that the value of CUDACPP_MAKEFILE is not modified further down after including make_opts
@@ -42,10 +42,10 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure common compiler flags for C++ and CUDA
+#=== Configure common compiler flags for C++ and CUDA/HIP
 
 INCFLAGS = -I.
-OPTFLAGS = -O3 # this ends up in CUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
+OPTFLAGS = -O3 # this ends up in GPUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
 
 # Dependency on src directory
 MG5AMC_COMMONLIB = mg5amc_common
@@ -121,24 +121,45 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure the CUDA compiler
+#=== Configure the GPU compiler (CUDA or HIP)
+
+# FIXME! (AV 24.01.2024)
+# In the current implementation (without separate builds for C++ and CUDA/HIP), we first check for cudacc and hipcc in CUDA_HOME and HIP_HOME.
+# If CUDA_HOME or HIP_HOME are not set, try to determine them from the path to cudacc and hipcc.
+# While convoluted, this is currently necessary to allow disabling CUDA/HIP builds by setting CUDA_HOME or HIP_HOME to invalid paths.
+# This will (probably?) be fixed when separate C++ and CUDA/HIP builds are implemented (PR #775).
 
 # If CXX is not a single word (example "clang++ --gcc-toolchain...") then disable CUDA builds (issue #505)
-# This is because it is impossible to pass this to "CUFLAGS += -ccbin <host-compiler>" below
+# This is because it is impossible to pass this to "GPUFLAGS += -ccbin <host-compiler>" below
 ifneq ($(words $(subst ccache ,,$(CXX))),1) # allow at most "CXX=ccache <host-compiler>" from outside
   $(warning CUDA builds are not supported for multi-word CXX "$(CXX)")
   override CUDA_HOME=disabled
 endif
 
-# If CUDA_HOME is not set, try to set it from the location of nvcc
+# If CUDA_HOME is not set, try to set it from the path to nvcc
 ifndef CUDA_HOME
-  CUDA_HOME = $(patsubst %bin/nvcc,%,$(shell which nvcc 2>/dev/null))
+  CUDA_HOME = $(patsubst %/bin/nvcc,%,$(shell which nvcc 2>/dev/null))
   $(warning CUDA_HOME was not set: using "$(CUDA_HOME)")
 endif
 
-# Set NVCC as $(CUDA_HOME)/bin/nvcc if it exists
+# If HIP_HOME is not set, try to set it from the path to hipcc
+ifndef HIP_HOME
+  HIP_HOME = $(patsubst %/bin/hipcc,%,$(shell which hipcc 2>/dev/null))
+  $(warning HIP_HOME was not set: using "$(HIP_HOME)")
+endif
+
+# FIXME! (AV 24.01.2024)
+# In the current implementation (without separate builds for C++ and CUDA/HIP),
+# builds are performed for HIP only if CUDA is not found in the path.
+# If both CUDA and HIP are installed, HIP builds can be triggered by unsetting CUDA_HOME.
+# This will be fixed when separate C++ and CUDA/HIP builds are implemented (PR #775).
+
+#--- Option 1: CUDA exists -> use CUDA
+
+# Set GPUCC as $(CUDA_HOME)/bin/nvcc if it exists
 ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
-  NVCC = $(CUDA_HOME)/bin/nvcc
+
+  GPUCC = $(CUDA_HOME)/bin/nvcc
   USE_NVTX ?=-DUSE_NVTX
   # See https://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/index.html
   # See https://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/
@@ -148,7 +169,7 @@ ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
   # Examples: use 60 for P100 (Piz Daint), 80 for A100 (Juwels Booster, NVidia raplab/Curiosity).
   MADGRAPH_CUDA_ARCHITECTURE ?= 70
   ###CUARCHFLAGS = -gencode arch=compute_$(MADGRAPH_CUDA_ARCHITECTURE),code=compute_$(MADGRAPH_CUDA_ARCHITECTURE) -gencode arch=compute_$(MADGRAPH_CUDA_ARCHITECTURE),code=sm_$(MADGRAPH_CUDA_ARCHITECTURE) # Older implementation (AV): go back to this one for multi-GPU support #533
-  ###CUARCHFLAGS = --gpu-architecture=compute_$(MADGRAPH_CUDA_ARCHITECTURE) --gpu-code=sm_$(MADGRAPH_CUDA_ARCHITECTURE),compute_$(MADGRAPH_CUDA_ARCHITECTURE) # Newer implementation (SH): cannot use this as-is for multi-GPU support #533
+  ###CUARCHFLAGS = --gpu-architecture=compute_$(MADGRAPH_CUDA_ARCHITECTURE) --gpu-code=sm_$(MADGRAPH_CUDA_ARCHITECTURE),compute_$(MADGRAPH_CUDA_ARCHITECTURE)  # Newer implementation (SH): cannot use this as-is for multi-GPU support #533
   comma:=,
   CUARCHFLAGS = $(foreach arch,$(subst $(comma), ,$(MADGRAPH_CUDA_ARCHITECTURE)),-gencode arch=compute_$(arch),code=compute_$(arch) -gencode arch=compute_$(arch),code=sm_$(arch))
   CUINC = -I$(CUDA_HOME)/include/
@@ -158,41 +179,79 @@ ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
     CURANDLIBFLAGS = -L$(CUDA_HOME)/lib64/ -lcurand # NB: -lcuda is not needed here!
   endif
   CUOPTFLAGS = -lineinfo
-  CUFLAGS = $(foreach opt, $(OPTFLAGS), -Xcompiler $(opt)) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
-  ###CUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
-  ###NVCC_VERSION = $(shell $(NVCC) --version | grep 'Cuda compilation tools' | cut -d' ' -f5 | cut -d, -f1)
-  CUFLAGS += -std=c++17 # need CUDA >= 11.2 (see #333): this is enforced in mgOnGpuConfig.h
+  ###GPUFLAGS = $(OPTFLAGS) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
+  GPUFLAGS = $(foreach opt, $(OPTFLAGS), -Xcompiler $(opt)) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
+  ###GPUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
+  ###GPUCC_VERSION = $(shell $(GPUCC) --version | grep 'Cuda compilation tools' | cut -d' ' -f5 | cut -d, -f1)
+  GPUFLAGS += -std=c++17 # need CUDA >= 11.2 (see #333): this is enforced in mgOnGpuConfig.h
   # Without -maxrregcount: baseline throughput: 6.5E8 (16384 32 12) up to 7.3E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 160 # improves throughput: 6.9E8 (16384 32 12) up to 7.7E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 128 # improves throughput: 7.3E8 (16384 32 12) up to 7.6E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 96 # degrades throughput: 4.1E8 (16384 32 12) up to 4.5E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 64 # degrades throughput: 1.7E8 (16384 32 12) flat at 1.7E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 160 # improves throughput: 6.9E8 (16384 32 12) up to 7.7E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 128 # improves throughput: 7.3E8 (16384 32 12) up to 7.6E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 96 # degrades throughput: 4.1E8 (16384 32 12) up to 4.5E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 64 # degrades throughput: 1.7E8 (16384 32 12) flat at 1.7E8 (65536 128 12)
+  CUBUILDRULEFLAGS = -Xcompiler -fPIC -c
+  CCBUILDRULEFLAGS = -Xcompiler -fPIC -c -x cu
+  CUDATESTFLAGS = -lcuda
+
+  # Set the host C++ compiler for GPUCC via "-ccbin <host-compiler>"
+  # (NB issue #505: this must be a single word, "clang++ --gcc-toolchain..." is not supported)
+  GPUFLAGS += -ccbin $(shell which $(subst ccache ,,$(CXX)))
+
+  # Allow newer (unsupported) C++ compilers with older versions of CUDA if ALLOW_UNSUPPORTED_COMPILER_IN_CUDA is set (#504)
+  ifneq ($(origin ALLOW_UNSUPPORTED_COMPILER_IN_CUDA),undefined)
+  GPUFLAGS += -allow-unsupported-compiler
+  endif
+
 else ifneq ($(origin REQUIRE_CUDA),undefined)
+
   # If REQUIRE_CUDA is set but no cuda is found, stop here (e.g. for CI tests on GPU #443)
-  $(error No cuda installation found (set CUDA_HOME or make nvcc visible in PATH))
+  $(error No cuda installation found (set CUDA_HOME or make GPUCC visible in PATH))
+
+#--- Option 2: CUDA does not exist, HIP exists -> use HIP
+
+# Set GPUCC as $(HIP_HOME)/bin/hipcc if it exists
+else ifneq ($(wildcard $(HIP_HOME)/bin/hipcc),)
+
+  GPUCC = $(HIP_HOME)/bin/hipcc
+  #USE_NVTX ?=-DUSE_NVTX # should maybe find something equivalent to this in HIP?
+  HIPARCHFLAGS = -target x86_64-linux-gnu --offload-arch=gfx90a
+  HIPINC = -I$(HIP_HOME)/include/
+  # Note: -DHIP_FAST_MATH is equivalent to -use_fast_math in HIP 
+  # (but only for single precision line 208: https://rocm-developer-tools.github.io/HIP/hcc__detail_2math__functions_8h_source.html)
+  # Note: CUOPTFLAGS should not be used for HIP, it had been added here but was then removed (#808)
+  GPUFLAGS = $(OPTFLAGS) $(INCFLAGS) $(HIPINC) $(HIPARCHFLAGS) -DHIP_FAST_MATH -DHIP_PLATFORM=amd -fPIC
+  ###GPUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
+  GPUFLAGS += -std=c++17
+  ###GPUFLAGS+= --maxrregcount 255 # (AV: is this option valid on HIP and meaningful on AMD GPUs?)
+  CUBUILDRULEFLAGS = -fPIC -c
+  CCBUILDRULEFLAGS = -fPIC -c -x hip
+
+else ifneq ($(origin REQUIRE_HIP),undefined)
+
+  # If REQUIRE_HIP is set but no HIP is found, stop here (e.g. for CI tests on GPU #443)
+  $(error No hip installation found (set HIP_HOME or make GPUCC visible in PATH))
+
+#--- Option 3: CUDA does not exist, HIP does not exist -> switch off both CUDA and HIP
+
 else
-  # No cuda. Switch cuda compilation off and go to common random numbers in C++
+
+  # No cudacc and no hipcc: switch CUDA and HIP compilation off and go to common random numbers in C++
   $(warning CUDA_HOME is not set or is invalid: export CUDA_HOME to compile with cuda)
-  override NVCC=
+  $(warning HIP_HOME is not set or is invalid: export HIP_HOME to compile with hip)
+  override GPUCC=
   override USE_NVTX=
   override CUINC=
   override CURANDLIBFLAGS=
-endif
-export NVCC
-export CUFLAGS
 
-# Set the host C++ compiler for nvcc via "-ccbin <host-compiler>"
-# (NB issue #505: this must be a single word, "clang++ --gcc-toolchain..." is not supported)
-CUFLAGS += -ccbin $(shell which $(subst ccache ,,$(CXX)))
-
-# Allow newer (unsupported) C++ compilers with older versions of CUDA if ALLOW_UNSUPPORTED_COMPILER_IN_CUDA is set (#504)
-ifneq ($(origin ALLOW_UNSUPPORTED_COMPILER_IN_CUDA),undefined)
-CUFLAGS += -allow-unsupported-compiler
 endif
 
+# Export GPUCC (so that it can also be used in cudacpp_src.mk?)
+export GPUCC
+export GPUFLAGS
+
 #-------------------------------------------------------------------------------
 
-#=== Configure ccache for C++ and CUDA builds
+#=== Configure ccache for C++ and CUDA/HIP builds
 
 # Enable ccache if USECCACHE=1
 ifeq ($(USECCACHE)$(shell echo $(CXX) | grep ccache),1)
@@ -201,15 +260,15 @@ endif
 #ifeq ($(USECCACHE)$(shell echo $(AR) | grep ccache),1)
 #  override AR:=ccache $(AR)
 #endif
-ifneq ($(NVCC),)
-  ifeq ($(USECCACHE)$(shell echo $(NVCC) | grep ccache),1)
-    override NVCC:=ccache $(NVCC)
+ifneq ($(GPUCC),)
+  ifeq ($(USECCACHE)$(shell echo $(GPUCC) | grep ccache),1)
+    override GPUCC:=ccache $(GPUCC)
   endif
 endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure PowerPC-specific compiler flags for C++ and CUDA
+#=== Configure PowerPC-specific compiler flags for C++ and CUDA/HIP
 
 # PowerPC-specific CXX compiler flags (being reviewed)
 ifeq ($(UNAME_P),ppc64le)
@@ -225,9 +284,9 @@ else
   ######CXXFLAGS+= -fno-semantic-interposition # no benefit (neither alone, nor combined with -flto)
 endif
 
-# PowerPC-specific CUDA compiler flags (to be reviewed!)
+# PowerPC-specific CUDA/HIP compiler flags (to be reviewed!)
 ifeq ($(UNAME_P),ppc64le)
-  CUFLAGS+= -Xcompiler -mno-float128
+  GPUFLAGS+= -Xcompiler -mno-float128
 endif
 
 #-------------------------------------------------------------------------------
@@ -235,9 +294,11 @@ endif
 #=== Configure defaults and check if user-defined choices exist for OMPFLAGS, AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
 
 # Set the default OMPFLAGS choice
-ifneq ($(shell $(CXX) --version | egrep '^Intel'),)
+ifneq ($(findstring hipcc,$(GPUCC)),)
+override OMPFLAGS = # disable OpenMP MT when using hipcc #802
+else ifneq ($(shell $(CXX) --version | egrep '^Intel'),)
 override OMPFLAGS = -fopenmp
-###override OMPFLAGS = # disable OpenMP MT on Intel (was ok without nvcc but not ok with nvcc before #578)
+###override OMPFLAGS = # disable OpenMP MT on Intel (was ok without GPUCC but not ok with GPUCC before #578)
 else ifneq ($(shell $(CXX) --version | egrep '^(clang)'),)
 override OMPFLAGS = -fopenmp
 ###override OMPFLAGS = # disable OpenMP MT on clang (was not ok without or with nvcc before #578)
@@ -293,7 +354,10 @@ endif
 
 # Set the default RNDGEN (random number generator) choice
 ifeq ($(RNDGEN),)
-  ifeq ($(NVCC),)
+  ifeq ($(GPUCC),)
+    override RNDGEN = hasNoCurand
+  # Edgecase for HIP compilation
+  else ifeq ($(findstring hipcc,$(GPUCC)),hipcc)
     override RNDGEN = hasNoCurand
   else ifeq ($(RNDGEN),)
     override RNDGEN = hasCurand
@@ -310,7 +374,7 @@ export OMPFLAGS
 
 #-------------------------------------------------------------------------------
 
-#=== Set the CUDA/C++ compiler flags appropriate to user-defined choices of AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
+#=== Set the CUDA/HIP/C++ compiler flags appropriate to user-defined choices of AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
 
 # Set the build flags appropriate to OMPFLAGS
 $(info OMPFLAGS=$(OMPFLAGS))
@@ -368,13 +432,13 @@ CXXFLAGS+= $(AVXFLAGS)
 $(info FPTYPE=$(FPTYPE))
 ifeq ($(FPTYPE),d)
   CXXFLAGS += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
-  CUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
+  GPUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
 else ifeq ($(FPTYPE),f)
   CXXFLAGS += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
-  CUFLAGS  += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
+  GPUFLAGS  += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
 else ifeq ($(FPTYPE),m)
   CXXFLAGS += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
-  CUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
+  GPUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
 else
   $(error Unknown FPTYPE='$(FPTYPE)': only 'd', 'f' and 'm' are supported)
 endif
@@ -383,7 +447,7 @@ endif
 $(info HELINL=$(HELINL))
 ifeq ($(HELINL),1)
   CXXFLAGS += -DMGONGPU_INLINE_HELAMPS
-  CUFLAGS  += -DMGONGPU_INLINE_HELAMPS
+  GPUFLAGS  += -DMGONGPU_INLINE_HELAMPS
 else ifneq ($(HELINL),0)
   $(error Unknown HELINL='$(HELINL)': only '0' and '1' are supported)
 endif
@@ -392,7 +456,7 @@ endif
 $(info HRDCOD=$(HRDCOD))
 ifeq ($(HRDCOD),1)
   CXXFLAGS += -DMGONGPU_HARDCODE_PARAM
-  CUFLAGS  += -DMGONGPU_HARDCODE_PARAM
+  GPUFLAGS  += -DMGONGPU_HARDCODE_PARAM
 else ifneq ($(HRDCOD),0)
   $(error Unknown HRDCOD='$(HRDCOD)': only '0' and '1' are supported)
 endif
@@ -444,11 +508,11 @@ ifeq ($(UNAME_S),Darwin)
   override CULIBFLAGSRPATH2 =
 else
   # RPATH to cuda/cpp libs when linking executables
-  override CXXLIBFLAGSRPATH = -Wl,-rpath,$(LIBDIRRPATH)
-  override CULIBFLAGSRPATH = -Xlinker -rpath,$(LIBDIRRPATH)
+  override CXXLIBFLAGSRPATH = -Wl,-rpath=$(LIBDIRRPATH)
+  override CULIBFLAGSRPATH = -Xlinker -rpath=$(LIBDIRRPATH)
   # RPATH to common lib when linking cuda/cpp libs
-  override CXXLIBFLAGSRPATH2 = -Wl,-rpath,'$$ORIGIN'
-  override CULIBFLAGSRPATH2 = -Xlinker -rpath,'$$ORIGIN'
+  override CXXLIBFLAGSRPATH2 = -Wl,-rpath='$$ORIGIN'
+  override CULIBFLAGSRPATH2 = -Xlinker -rpath='$$ORIGIN'
 endif
 
 # Setting LD_LIBRARY_PATH or DYLD_LIBRARY_PATH in the RUNTIME is no longer necessary (neither on Linux nor on Mac)
@@ -461,7 +525,7 @@ override RUNTIME =
 cxx_main=$(BUILDDIR)/check.exe
 fcxx_main=$(BUILDDIR)/fcheck.exe
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 cu_main=$(BUILDDIR)/gcheck.exe
 fcu_main=$(BUILDDIR)/fgcheck.exe
 else
@@ -492,14 +556,15 @@ $(BUILDDIR)/.build.$(TAG):
 	@touch $(BUILDDIR)/.build.$(TAG)
 
 # Generic target and build rules: objects from CUDA compilation
-ifneq ($(NVCC),)
+# NB: CCBUILDRULEFLAGS includes "-x cu" for nvcc and "-x hip" for hipcc (#810)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/%.o : %.cu *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $(CUBUILDRULEFLAGS) $< -o $@
 
 $(BUILDDIR)/%_cu.o : %.cc *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c -x cu $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $(CCBUILDRULEFLAGS) $< -o $@
 endif
 
 # Generic target and build rules: objects from C++ compilation
@@ -509,11 +574,14 @@ $(BUILDDIR)/%.o : %.cc *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	$(CXX) $(CPPFLAGS) $(CXXFLAGS) -fPIC -c $< -o $@
 
 # Apply special build flags only to CrossSectionKernel.cc and gCrossSectionKernel.cu (no fast math, see #117 and #516)
+# Added edgecase for HIP compilation
 ifeq ($(shell $(CXX) --version | grep ^nvc++),)
 $(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS := $(filter-out -ffast-math,$(CXXFLAGS))
 $(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS += -fno-fast-math
-ifneq ($(NVCC),)
-$(BUILDDIR)/gCrossSectionKernels.o: CUFLAGS += -Xcompiler -fno-fast-math
+ifeq ($(findstring nvcc,$(GPUCC)),nvcc)
+  $(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -Xcompiler -fno-fast-math
+else
+  $(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -fno-fast-math
 endif
 endif
 
@@ -530,10 +598,10 @@ ifeq ($(RNDGEN),hasCurand)
 $(BUILDDIR)/CurandRandomNumberKernel.o: CXXFLAGS += $(CUINC)
 endif
 
-# Avoid "warning: builtin __has_trivial_... is deprecated; use __is_trivially_... instead" in nvcc with icx2023 (#592)
+# Avoid "warning: builtin __has_trivial_... is deprecated; use __is_trivially_... instead" in GPUCC with icx2023 (#592)
 ifneq ($(shell $(CXX) --version | egrep '^(Intel)'),)
-ifneq ($(NVCC),)
-CUFLAGS += -Xcompiler -Wno-deprecated-builtins
+ifneq ($(GPUCC),)
+GPUFLAGS += -Wno-deprecated-builtins
 endif
 endif
 
@@ -541,8 +609,8 @@ endif
 # This patch does remove the warning, but I prefer to keep it disabled for the moment...
 ###ifneq ($(shell $(CXX) --version | egrep '^(clang|Apple clang|Intel)'),)
 ###$(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS += -Wno-overriding-t-option
-###ifneq ($(NVCC),)
-###$(BUILDDIR)/gCrossSectionKernels.o: CUFLAGS += -Xcompiler -Wno-overriding-t-option
+###ifneq ($(GPUCC),)
+###$(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -Xcompiler -Wno-overriding-t-option
 ###endif
 ###endif
 
@@ -569,7 +637,7 @@ MG5AMC_CXXLIB = mg5amc_$(processid_short)_cpp
 cxx_objects_lib=$(BUILDDIR)/CPPProcess.o $(BUILDDIR)/MatrixElementKernels.o $(BUILDDIR)/BridgeKernels.o $(BUILDDIR)/CrossSectionKernels.o
 cxx_objects_exe=$(BUILDDIR)/CommonRandomNumberKernel.o $(BUILDDIR)/RamboSamplingKernels.o
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 MG5AMC_CULIB = mg5amc_$(processid_short)_cuda
 cu_objects_lib=$(BUILDDIR)/gCPPProcess.o $(BUILDDIR)/gMatrixElementKernels.o $(BUILDDIR)/gBridgeKernels.o $(BUILDDIR)/gCrossSectionKernels.o
 cu_objects_exe=$(BUILDDIR)/gCommonRandomNumberKernel.o $(BUILDDIR)/gRamboSamplingKernels.o
@@ -581,11 +649,17 @@ $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: cxx_objects_lib += $(BUILDDIR)/fbridge.o
 $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib)
 	$(CXX) -shared -o $@ $(cxx_objects_lib) $(CXXLIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: $(BUILDDIR)/fbridge_cu.o
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: cu_objects_lib += $(BUILDDIR)/fbridge_cu.o
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cu_objects_lib)
-	$(NVCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+# Bypass std::filesystem completely to ease portability on LUMI #803
+#ifneq ($(findstring hipcc,$(GPUCC)),)
+#	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB) -lstdc++fs
+#else
+#	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+#endif
 endif
 
 #-------------------------------------------------------------------------------
@@ -602,16 +676,16 @@ $(cxx_main): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PAT
 $(cxx_main): $(BUILDDIR)/check_sa.o $(LIBDIR)/lib$(MG5AMC_CXXLIB).so $(cxx_objects_exe) $(BUILDDIR)/CurandRandomNumberKernel.o
 	$(CXX) -o $@ $(BUILDDIR)/check_sa.o $(OMPFLAGS) -ldl -pthread $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe) $(BUILDDIR)/CurandRandomNumberKernel.o $(CURANDLIBFLAGS)
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(cu_main): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(cu_main): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(cu_main): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(cu_main): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 else ifneq ($(shell $(CXX) --version | grep ^nvc++),) # support nvc++ #531
 $(cu_main): LIBFLAGS += -L$(patsubst %bin/nvc++,%lib,$(subst ccache ,,$(CXX))) -lnvhpcatm -lnvcpumath -lnvc
 endif
 $(cu_main): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(cu_main): $(BUILDDIR)/gcheck_sa.o $(LIBDIR)/lib$(MG5AMC_CULIB).so $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o
-	$(NVCC) -o $@ $(BUILDDIR)/gcheck_sa.o $(CUARCHFLAGS) $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o $(CURANDLIBFLAGS)
+	$(GPUCC) -o $@ $(BUILDDIR)/gcheck_sa.o $(CUARCHFLAGS) $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o $(CURANDLIBFLAGS)
 endif
 
 #-------------------------------------------------------------------------------
@@ -635,19 +709,27 @@ $(fcxx_main): LIBFLAGS += -L$(shell dirname $(shell $(FC) --print-file-name libg
 endif
 $(fcxx_main): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(fcxx_main): $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler.o $(LIBDIR)/lib$(MG5AMC_CXXLIB).so $(cxx_objects_exe)
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(BUILDDIR)/fcheck_sa.o $(OMPFLAGS) $(BUILDDIR)/fsampler.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe) -lstdc++
+else
 	$(CXX) -o $@ $(BUILDDIR)/fcheck_sa.o $(OMPFLAGS) $(BUILDDIR)/fsampler.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe)
+endif
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(fcu_main): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(fcu_main): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(fcu_main): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(fcu_main): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 endif
 ifeq ($(UNAME_S),Darwin)
 $(fcu_main): LIBFLAGS += -L$(shell dirname $(shell $(FC) --print-file-name libgfortran.dylib)) # add path to libgfortran on Mac #375
 endif
 $(fcu_main): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(fcu_main): $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBDIR)/lib$(MG5AMC_CULIB).so $(cu_objects_exe)
-	$(NVCC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe)
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) -lstdc++ -L$(shell dirname $(shell $(GPUCC) -print-prog-name=clang))/../../lib -lamdhip64
+else
+	$(GPUCC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe)
+endif
 endif
 
 #-------------------------------------------------------------------------------
@@ -659,7 +741,7 @@ $(BUILDDIR)/testxxx.o: testxxx_cc_ref.txt
 $(testmain): $(BUILDDIR)/testxxx.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/testxxx.o # Comment out this line to skip the C++ test of xxx functions
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/testxxx_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/testxxx_cu.o: INCFLAGS += $(GTESTINC)
 $(BUILDDIR)/testxxx_cu.o: testxxx_cc_ref.txt
@@ -672,7 +754,7 @@ $(BUILDDIR)/testmisc.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/testmisc.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/testmisc.o # Comment out this line to skip the C++ miscellaneous tests
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/testmisc_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/testmisc_cu.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/testmisc_cu.o
@@ -684,12 +766,12 @@ $(BUILDDIR)/runTest.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/runTest.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/runTest.o
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/runTest_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/runTest_cu.o: INCFLAGS += $(GTESTINC)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(testmain): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(testmain): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(testmain): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(testmain): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 else ifneq ($(shell $(CXX) --version | grep ^nvc++),) # support nvc++ #531
 $(testmain): LIBFLAGS += -L$(patsubst %bin/nvc++,%lib,$(subst ccache ,,$(CXX))) -lnvhpcatm -lnvcpumath -lnvc
 endif
@@ -713,14 +795,23 @@ $(testmain): LIBFLAGS += -lgomp
 endif
 endif
 
-ifeq ($(NVCC),) # link only runTest.o
+# Bypass std::filesystem completely to ease portability on LUMI #803
+#ifneq ($(findstring hipcc,$(GPUCC)),)
+#$(testmain): LIBFLAGS += -lstdc++fs
+#endif
+
+ifeq ($(GPUCC),) # link only runTest.o
 $(testmain): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(testmain): $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib) $(cxx_objects_exe) $(GTESTLIBS)
 	$(CXX) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) -ldl -pthread $(LIBFLAGS)
 else # link both runTest.o and runTest_cu.o
 $(testmain): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(testmain): $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) $(GTESTLIBS)
-	$(NVCC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) -lcuda
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) $(CUDATESTFLAGS) -lstdc++ -lpthread  -L$(shell dirname $(shell $(GPUCC) -print-prog-name=clang))/../../lib -lamdhip64
+else
+	$(GPUCC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) $(CUDATESTFLAGS)
+endif
 endif
 
 # Use target gtestlibs to build only googletest
@@ -829,9 +920,9 @@ ifeq ($(USECCACHE),1)
 	ccache --version | head -1
 endif
 	@echo ""
-	@echo NVCC=$(NVCC)
-ifneq ($(NVCC),)
-	$(NVCC) --version
+	@echo GPUCC=$(GPUCC)
+ifneq ($(GPUCC),)
+	$(GPUCC) --version
 endif
 	@echo ""
 	@echo CXX=$(CXX)
@@ -850,7 +941,7 @@ endif
 
 # Target: check (run the C++ test executable)
 # [NB THIS IS WHAT IS USED IN THE GITHUB CI!]
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 check: runTest cmpFcheck cmpFGcheck
 else
 check: runTest cmpFcheck
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/fbridge.cc b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/fbridge.cc
index 2d2b36d560..22ce3f5115 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/fbridge.cc
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/fbridge.cc
@@ -1,11 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Roiser (Oct 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "Bridge.h"
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 
 extern "C"
 {
@@ -22,7 +22,7 @@ extern "C"
    * Using the same Fortran MadEvent code, linking to the hetrerogeneous library would allow access to both CPU and GPU implementations.
    * The specific heterogeneous configuration (how many GPUs, how many threads on each CPU, etc) could be loaded in CUDA/C++ from a data file.
    */
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -46,8 +46,8 @@ extern "C"
    */
   void fbridgecreate_( CppObjectInFortran** ppbridge, const int* pnevtF, const int* pnparF, const int* pnp4F )
   {
-#ifdef __CUDACC__
-    CudaRuntime::setUp();
+#ifdef MGONGPUCPP_GPUIMPL
+    GpuRuntime::setUp();
 #endif
     // (NB: CPPProcess::initProc no longer needs to be executed here because it is called in the Bridge constructor)
     // FIXME: disable OMP in Bridge when called from Fortran
@@ -65,8 +65,8 @@ extern "C"
     Bridge<FORTRANFPTYPE>* pbridge = dynamic_cast<Bridge<FORTRANFPTYPE>*>( *ppbridge );
     if( pbridge == 0 ) throw std::runtime_error( "fbridgedelete_: invalid Bridge address" );
     delete pbridge;
-#ifdef __CUDACC__
-    CudaRuntime::tearDown();
+#ifdef MGONGPUCPP_GPUIMPL
+    GpuRuntime::tearDown();
 #endif
   }
 
@@ -96,7 +96,7 @@ extern "C"
   {
     Bridge<FORTRANFPTYPE>* pbridge = dynamic_cast<Bridge<FORTRANFPTYPE>*>( *ppbridge );
     if( pbridge == 0 ) throw std::runtime_error( "fbridgesequence_: invalid Bridge address" );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Use the device/GPU implementation in the CUDA library
     // (there is also a host implementation in this library)
     pbridge->gpu_sequence( momenta, gs, rndhel, rndcol, *pchannelId, mes, selhel, selcol );
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/fsampler.cc b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/fsampler.cc
index 2fb445372d..3743934f41 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/fsampler.cc
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/fsampler.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Feb 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "mgOnGpuConfig.h"
 
@@ -13,7 +13,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -40,7 +40,7 @@ namespace mg5amcCpu
   private:
     const int m_nevt; // The number of events in each iteration
     int m_iiter;      // The iteration counter (for random number seeding)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     HostBufferRndNumMomenta m_hstRndmom; // Memory buffers for random numbers
     HostBufferMomenta m_hstMomenta;      // Memory buffers for momenta
     HostBufferWeights m_hstWeights;      // Memory buffers for sampling weights
@@ -105,7 +105,7 @@ namespace mg5amcCpu
 
 extern "C"
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/runTest.cc b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/runTest.cc
index d4a760a71b..de327f2321 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/runTest.cc
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/runTest.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Hageboeck (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*xxx to run only testxxx.cc tests
 //----------------------------------------------------------------------------
@@ -18,7 +18,7 @@
 #include "RandomNumberKernels.h"
 #include "epoch_process_id.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using namespace mg5amcGpu;
 #else
 using namespace mg5amcCpu;
@@ -35,7 +35,7 @@ struct CUDA_CPU_TestBase : public TestDriverBase
     : TestDriverBase( npar, refFileName ) {}
 };
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 struct CPUTest : public CUDA_CPU_TestBase
 {
   // Struct data members (process, and memory structures for random numbers, momenta, matrix elements and weights on host and device)
@@ -119,7 +119,7 @@ struct CPUTest : public CUDA_CPU_TestBase
 };
 #endif
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 struct CUDATest : public CUDA_CPU_TestBase
 {
   // Reset the device when our test goes out of scope. Note that this should happen after
@@ -128,7 +128,7 @@ struct CUDATest : public CUDA_CPU_TestBase
   {
     ~DeviceReset()
     {
-      checkCuda( cudaDeviceReset() ); // this is needed by cuda-memcheck --leak-check full
+      checkGpu( gpuDeviceReset() ); // this is needed by cuda-memcheck --leak-check full
     }
   } deviceResetter;
 
@@ -256,7 +256,7 @@ INSTANTIATE_TEST_SUITE_P( prefix, \
                           test_suite_name, \
                           testing::Values( new CUDATest( MG_EPOCH_REFERENCE_FILE_NAME ) ) );
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 MG_INSTANTIATE_TEST_SUITE_GPU( XTESTID_GPU( MG_EPOCH_PROCESS_ID ), MadgraphTest );
 #else
 MG_INSTANTIATE_TEST_SUITE_CPU( XTESTID_CPU( MG_EPOCH_PROCESS_ID ), MadgraphTest );
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/testmisc.cc b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/testmisc.cc
index 895d6eeb56..ba9e59a8a3 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/testmisc.cc
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/testmisc.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*misc to run only testmisc.cc tests
 //----------------------------------------------------------------------------
@@ -17,7 +17,7 @@
 #include <sstream>
 #include <typeinfo>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #define TESTID( s ) s##_GPU_MISC
 #else
 #define TESTID( s ) s##_CPU_MISC
@@ -26,7 +26,7 @@
 #define XTESTID( s ) TESTID( s )
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -59,7 +59,7 @@ namespace mg5amcCpu
 
 TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testmisc )
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/testxxx.cc b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/testxxx.cc
index 3361fe5aa9..e5167de00c 100644
--- a/epochX/cudacpp/gg_ttggg.sa/SubProcesses/testxxx.cc
+++ b/epochX/cudacpp/gg_ttggg.sa/SubProcesses/testxxx.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Apr 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*xxx to run only testxxx.cc tests
 //----------------------------------------------------------------------------
@@ -24,7 +24,7 @@
 #include <iomanip>
 #include <iostream>
 #include <vector>
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #define TESTID( s ) s##_GPU_XXX
 #else
 #define TESTID( s ) s##_CPU_XXX
@@ -32,7 +32,7 @@
 
 #define XTESTID( s ) TESTID( s )
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -42,7 +42,7 @@ namespace mg5amcCpu
   int FPEhandlerIevt = -1;
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU): '" << FPEhandlerMessage << "' ievt=" << FPEhandlerIevt << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU neppV=" << neppV << "): '" << FPEhandlerMessage << "' ievt=" << FPEhandlerIevt << std::endl;
@@ -53,7 +53,7 @@ namespace mg5amcCpu
 
 TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -77,7 +77,7 @@ TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
   assert( nevt % neppM == 0 ); // nevt must be a multiple of neppM
   assert( nevt % neppV == 0 ); // nevt must be a multiple of neppV
   // Fill in the input momenta
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   mg5amcGpu::PinnedHostBufferMomenta hstMomenta( nevt ); // AOSOA[npagM][npar=4][np4=4][neppM]
 #else
   mg5amcCpu::HostBufferMomenta hstMomenta( nevt ); // AOSOA[npagM][npar=4][np4=4][neppM]
@@ -322,7 +322,7 @@ TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
   {
     for( int ievt = 0; ievt < nevt; ievt++ )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       using namespace mg5amcGpu;
 #else
       using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_ttggg.sa/src/HelAmps_sm.h b/epochX/cudacpp/gg_ttggg.sa/src/HelAmps_sm.h
index 8df465ad6d..8b4ad719be 100644
--- a/epochX/cudacpp/gg_ttggg.sa/src/HelAmps_sm.h
+++ b/epochX/cudacpp/gg_ttggg.sa/src/HelAmps_sm.h
@@ -5,7 +5,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -28,7 +28,7 @@
 //#include <iomanip>
 //#include <iostream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttggg.sa/src/Parameters_sm.cc b/epochX/cudacpp/gg_ttggg.sa/src/Parameters_sm.cc
index 64fc3fea62..067445b198 100644
--- a/epochX/cudacpp/gg_ttggg.sa/src/Parameters_sm.cc
+++ b/epochX/cudacpp/gg_ttggg.sa/src/Parameters_sm.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -17,7 +17,7 @@
 #include <iomanip>
 #include <iostream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using namespace mg5amcGpu;
 #else
 using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gg_ttggg.sa/src/Parameters_sm.h b/epochX/cudacpp/gg_ttggg.sa/src/Parameters_sm.h
index b6568d3761..9581d66e0e 100644
--- a/epochX/cudacpp/gg_ttggg.sa/src/Parameters_sm.h
+++ b/epochX/cudacpp/gg_ttggg.sa/src/Parameters_sm.h
@@ -27,7 +27,7 @@
 #include "read_slha.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -93,7 +93,7 @@ namespace mg5amcCpu
 #include <limits>
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -218,7 +218,7 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -239,7 +239,7 @@ namespace mg5amcCpu
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wunused-variable"  // e.g. <<warning: unused variable ‘mdl_G__exp__2’ [-Wunused-variable]>>
 #pragma GCC diagnostic ignored "-Wunused-parameter" // e.g. <<warning: unused parameter ‘G’ [-Wunused-parameter]>>
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #pragma nv_diagnostic push
 #pragma nv_diag_suppress 177 // e.g. <<warning #177-D: variable "mdl_G__exp__2" was declared but never referenced>>
 #endif
@@ -267,7 +267,7 @@ namespace mg5amcCpu
       // End SM implementation - no special handling of vectors of floats as in EFT (#439)
       return out;
     }
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #pragma GCC diagnostic pop
 #pragma nv_diagnostic pop
 #endif
diff --git a/epochX/cudacpp/gg_ttggg.sa/src/cudacpp_src.mk b/epochX/cudacpp/gg_ttggg.sa/src/cudacpp_src.mk
index d4cc628aec..b2b9da5288 100644
--- a/epochX/cudacpp/gg_ttggg.sa/src/cudacpp_src.mk
+++ b/epochX/cudacpp/gg_ttggg.sa/src/cudacpp_src.mk
@@ -19,7 +19,7 @@ SHELL := /bin/bash
 #=== Configure common compiler flags for CUDA and C++
 
 INCFLAGS = -I.
-OPTFLAGS = -O3 # this ends up in CUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
+OPTFLAGS = -O3 # this ends up in GPUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
 
 #-------------------------------------------------------------------------------
 
@@ -45,13 +45,13 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure the CUDA compiler (note: NVCC is already exported including ccache)
+#=== Configure the CUDA compiler (note: GPUCC is already exported including ccache)
 
-###$(info NVCC=$(NVCC))
+###$(info GPUCC=$(GPUCC))
 
 #-------------------------------------------------------------------------------
 
-#=== Configure ccache for C++ builds (note: NVCC is already exported including ccache)
+#=== Configure ccache for C++ builds (note: GPUCC is already exported including ccache)
 
 # Enable ccache if USECCACHE=1
 ifeq ($(USECCACHE)$(shell echo $(CXX) | grep ccache),1)
@@ -92,6 +92,13 @@ endif
 ###$(info OMPFLAGS=$(OMPFLAGS))
 CXXFLAGS += $(OMPFLAGS)
 
+# Add correct flags for nvcc (-x cu) and hipcc (-x hip) for GPU code (see #810)
+ifeq ($(findstring nvcc,$(GPUCC)),nvcc)
+  GPUFLAGS += -Xcompiler -fPIC -c -x cu
+else ifeq ($(findstring hipcc,$(GPUCC)),hipcc)
+  GPUFLAGS += -fPIC -c -x hip
+endif
+
 # Set the build flags appropriate to each AVX choice (example: "make AVX=none")
 # [NB MGONGPU_PVW512 is needed because "-mprefer-vector-width=256" is not exposed in a macro]
 # [See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96476]
@@ -253,20 +260,20 @@ $(BUILDDIR)/%.o : %.cc *.h $(BUILDDIR)/.build.$(TAG)
 # Generic target and build rules: objects from CUDA compilation
 $(BUILDDIR)/%_cu.o : %.cc *.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c -x cu $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $< -o $@
 
 #-------------------------------------------------------------------------------
 
 cxx_objects=$(addprefix $(BUILDDIR)/, Parameters_sm.o read_slha.o)
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 cu_objects=$(addprefix $(BUILDDIR)/, Parameters_sm_cu.o)
 endif
 
 # Target (and build rules): common (src) library
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so : $(cxx_objects) $(cu_objects)
 	@if [ ! -d $(LIBDIR) ]; then echo "mkdir -p $(LIBDIR)"; mkdir -p $(LIBDIR); fi
-	$(NVCC) -shared -o $@ $(cxx_objects) $(cu_objects) $(LDFLAGS)
+	$(GPUCC) -shared -o $@ $(cxx_objects) $(cu_objects) $(LDFLAGS)
 else
 $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so : $(cxx_objects)
 	@if [ ! -d $(LIBDIR) ]; then echo "mkdir -p $(LIBDIR)"; mkdir -p $(LIBDIR); fi
diff --git a/epochX/cudacpp/gg_ttggg.sa/src/mgOnGpuConfig.h b/epochX/cudacpp/gg_ttggg.sa/src/mgOnGpuConfig.h
index b247654dcf..475749ca7c 100644
--- a/epochX/cudacpp/gg_ttggg.sa/src/mgOnGpuConfig.h
+++ b/epochX/cudacpp/gg_ttggg.sa/src/mgOnGpuConfig.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jul 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUCONFIG_H
 #define MGONGPUCONFIG_H 1
@@ -10,12 +10,26 @@
 // There are two different code bases for standalone_cudacpp (without multichannel) and madevent+cudacpp (with multichannel)
 #undef MGONGPU_SUPPORTS_MULTICHANNEL
 
+// Is this a GPU (CUDA, HIP) or CPU implementation?
+#ifdef __CUDACC__
+#define MGONGPUCPP_GPUIMPL cuda
+#elif defined __HIPCC__
+#define MGONGPUCPP_GPUIMPL hip
+#include "hip/hip_runtime.h" // needed for blockDim, blockIdx, threadIdx: better in mgOnGpuConfig.h than in GpuAbstraction.h
+#else
+#undef MGONGPUCPP_GPUIMPL
+#endif
+
 // ** NB1 Throughputs (e.g. 6.8E8) are events/sec for "./gcheck.exe -p 65536 128 12"
 // ** NB2 Baseline on b7g47n0004 fluctuates (probably depends on load on other VMs)
 
 // Choose if curand is supported for generating random numbers
+// For HIP, by default, do not use curand (common random numbers will be used instead)
 // For both CUDA and C++, by default, do not inline, but allow this macro to be set from outside with e.g. -DMGONGPU_HAS_NO_CURAND
-// (there exist CUDA installations, e.g. using the HPC package, which do not include curand - see PR #784)
+// (there exist CUDA installations, e.g. using the HPC package, which do not include curand - see PR #784 and #785)
+#if defined __HIPCC__
+#define MGONGPU_HAS_NO_CURAND 1
+#else
 //#ifdef __CUDACC__
 //#undef MGONGPU_HAS_NO_CURAND // default
 ////#define MGONGPU_HAS_NO_CURAND 1
@@ -23,6 +37,7 @@
 //#undef MGONGPU_HAS_NO_CURAND // default
 ////#define MGONGPU_HAS_NO_CURAND 1
 //#endif
+#endif
 
 // Choose floating point precision (for everything but color algebra #537)
 // If one of these macros has been set from outside with e.g. -DMGONGPU_FPTYPE_FLOAT, nothing happens (issue #167)
@@ -54,23 +69,31 @@
 //#undef MGONGPU_HARDCODE_PARAM // default
 ////#define MGONGPU_HARDCODE_PARAM 1
 
-// Complex type in c++: std::complex or cxsmpl (CHOOSE ONLY ONE)
-#ifndef __CUDACC__
-//#define MGONGPU_CPPCXTYPE_STDCOMPLEX 1 // ~8 percent slower on float, same on double (5.1E6/double, 9.4E6/float)
-#define MGONGPU_CPPCXTYPE_CXSMPL 1 // new default (5.1E6/double, 10.2E6/float)
-#endif
-
-// Complex type in cuda: thrust or cucomplex or cxsmpl (CHOOSE ONLY ONE)
+// Complex type in CUDA: thrust or cucomplex or cxsmpl (CHOOSE ONLY ONE)
+// (NB THIS IS MGONGPU_*CU*CXTYPE_xxx)
 #ifdef __CUDACC__
 #define MGONGPU_CUCXTYPE_THRUST 1 // default (~1.15E9/double, ~3.2E9/float)
 //#define MGONGPU_CUCXTYPE_CUCOMPLEX 1 // ~10 percent slower (1.03E9/double, ~2.8E9/float)
 //#define MGONGPU_CUCXTYPE_CXSMPL 1 // ~10 percent slower (1.00E9/double, ~2.9E9/float)
+
+// Complex type in HIP: cxsmpl (ONLY ONE OPTION POSSIBLE? #810)
+// (NB THIS IS MGONGPU_*HIP*CXTYPE_xxx)
+#elif defined __HIPCC__
+#define MGONGPU_HIPCXTYPE_CXSMPL 1 // default for HIP
+
+// Complex type in C++: std::complex or cxsmpl (CHOOSE ONLY ONE)
+// (NB THIS IS MGONGPU_*CPP*CXTYPE_xxx)
+#else
+//#define MGONGPU_CPPCXTYPE_STDCOMPLEX 1 // ~8 percent slower on float, same on double (5.1E6/double, 9.4E6/float)
+#define MGONGPU_CPPCXTYPE_CXSMPL 1 // new default (5.1E6/double, 10.2E6/float)
 #endif
 
-// Cuda nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
+// CUDA nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
 #ifdef __CUDACC__
-#undef MGONGPU_NSIGHT_DEBUG // default
+#undef MGONGPU_NSIGHT_DEBUG // default in CUDA
 //#define MGONGPU_NSIGHT_DEBUG 1
+#else
+#undef MGONGPU_NSIGHT_DEBUG // only option in HIP or C++
 #endif
 
 // SANITY CHECKS (floating point precision for everything but color algebra #537)
@@ -86,17 +109,21 @@
 #error You cannot use double precision for color algebra and single precision elsewhere
 #endif
 
-// SANITY CHECKS (c++ complex number implementation)
-#ifndef __CUDACC__
-#if defined MGONGPU_CPPCXTYPE_STDCOMPLEX and defined MGONGPU_CPPCXTYPE_CXSMPL
-#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CPPCXTYPE_STDCOMPLEX or MGONGPU_CPPCXTYPE_CXSMPL
+// SANITY CHECKS (CUDA complex number implementation)
+#ifdef __CUDACC__
+#if defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CUCOMPLEX
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX for CUDA
+#elif defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CXSMPL for CUDA
+#elif defined MGONGPU_CUCXTYPE_CUCOMPLEX and defined MGONGPU_CUCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE OF MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL for CUDA
 #endif
 #endif
 
-// SANITY CHECKS (cuda complex number implementation)
-#ifdef __CUDACC__
-#if defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CUCOMPLEX and defined MGONGPU_CUCXTYPE_CXSMPL
-#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL
+// SANITY CHECKS (C++ complex number implementation)
+#ifndef MGONGPUCPP_GPUIMPL
+#if defined MGONGPU_CPPCXTYPE_STDCOMPLEX and defined MGONGPU_CPPCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CPPCXTYPE_STDCOMPLEX or MGONGPU_CPPCXTYPE_CXSMPL for C++
 #endif
 #endif
 
@@ -134,7 +161,7 @@ namespace mgOnGpu
   // Alignment requirement for using reinterpret_cast with SIMD vectorized code
   // (using reinterpret_cast with non aligned memory may lead to segmentation faults!)
   // Only needed for C++ code but can be enforced also in NVCC builds of C++ code using CUDA>=11.2 and C++17 (#318, #319, #333)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   constexpr int cppAlign = 64; // alignment requirement for SIMD vectorization (64-byte i.e. 512-bit)
 #endif
 
@@ -145,7 +172,7 @@ using mgOnGpu::fptype;
 using mgOnGpu::fptype2;
 
 // C++ SIMD vectorization width (this will be used to set neppV)
-#ifdef __CUDACC__ // CUDA implementation has no SIMD
+#ifdef MGONGPUCPP_GPUIMPL // CUDA and HIP implementations have no SIMD
 #undef MGONGPU_CPPSIMD
 #elif defined __AVX512VL__ && defined MGONGPU_PVW512 // C++ "512z" AVX512 with 512 width (512-bit ie 64-byte): 8 (DOUBLE) or 16 (FLOAT)
 #ifdef MGONGPU_FPTYPE_DOUBLE
@@ -175,9 +202,9 @@ using mgOnGpu::fptype2;
 #undef MGONGPU_CPPSIMD
 #endif
 
-// Cuda nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
+// CUDA nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
 // Arguments (not used so far): text is __FUNCTION__, code is 0 (start) or 1 (end)
-#if defined __CUDACC__ && defined MGONGPU_NSIGHT_DEBUG /* clang-format off */
+#if defined __CUDA__ && defined MGONGPU_NSIGHT_DEBUG /* clang-format off */
 #define mgDebugDeclare() __shared__ float mgDebugCounter[mgOnGpu::ntpbMAX];
 #define mgDebugInitialise() { mgDebugCounter[threadIdx.x] = 0; }
 #define mgDebug( code, text ) { mgDebugCounter[threadIdx.x] += 1; }
@@ -189,8 +216,8 @@ using mgOnGpu::fptype2;
 #define mgDebugFinalise() { /*noop*/ }
 #endif /* clang-format on */
 
-// Define empty CUDA declaration specifiers for C++
-#ifndef __CUDACC__
+// Define empty CUDA/HIP declaration specifiers for C++
+#ifndef MGONGPUCPP_GPUIMPL
 #define __global__
 #define __host__
 #define __device__
diff --git a/epochX/cudacpp/gg_ttggg.sa/src/mgOnGpuCxtypes.h b/epochX/cudacpp/gg_ttggg.sa/src/mgOnGpuCxtypes.h
index ca9a9f00c0..7ede1dbfae 100644
--- a/epochX/cudacpp/gg_ttggg.sa/src/mgOnGpuCxtypes.h
+++ b/epochX/cudacpp/gg_ttggg.sa/src/mgOnGpuCxtypes.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022, based on earlier work by D. Smith) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUCXTYPES_H
 #define MGONGPUCXTYPES_H 1
@@ -19,7 +19,7 @@
 #include <complex>
 
 // Complex type in cuda: thrust or cucomplex or cxsmpl
-#ifdef __CUDACC__
+#ifdef __CUDACC__ // this must be __CUDAC__ (not MGONGPUCPP_GPUIMPL)
 #if defined MGONGPU_CUCXTYPE_THRUST
 #pragma clang diagnostic push
 #pragma clang diagnostic ignored "-Wtautological-compare" // for icpx2021/clang13 (https://stackoverflow.com/a/15864661)
@@ -30,8 +30,13 @@
 #elif not defined MGONGPU_CUCXTYPE_CXSMPL
 #error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL
 #endif
+// Complex type in HIP: cxsmpl
+#elif defined __HIPCC__
+#if not defined MGONGPU_HIPCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_HIPCXTYPE_CXSMPL
+#endif
 #else
-// Complex type in c++: std::complex or cxsmpl
+// Complex type in c++ or HIP: std::complex or cxsmpl
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
 #include <cmath>
 #elif not defined MGONGPU_CPPCXTYPE_CXSMPL
@@ -82,7 +87,7 @@ namespace mgOnGpu /* clang-format off */
 using mgOnGpu::cxsmpl;
 
 // Printout to stream for user defined types
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -92,7 +97,7 @@ namespace mg5amcCpu
   inline __host__ std::ostream&
   operator<<( std::ostream& out, const cxsmpl<FP>& c )
   {
-    out << std::complex( c.real(), c.imag() );
+    out << std::complex<FP>( c.real(), c.imag() );
     return out;
   }
 
@@ -215,14 +220,14 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
   // --- Type definitions (complex type: cxtype)
-#ifdef __CUDACC__ // cuda
+#ifdef __CUDACC__ // this must be __CUDAC__ (not MGONGPUCPP_GPUIMPL)
 #if defined MGONGPU_CUCXTYPE_THRUST
   typedef thrust::complex<fptype> cxtype;
 #elif defined MGONGPU_CUCXTYPE_CUCOMPLEX
@@ -255,13 +260,13 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
-#if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+#if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
 
   //------------------------------
   // CUDA or C++ - using cxsmpl
@@ -303,11 +308,11 @@ namespace mg5amcCpu
     return cxmake( c.real(), c.imag() );
   }
 
-#endif // #if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+#endif // #if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
 
   //==========================================================================
 
-#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_THRUST // cuda + thrust
+#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_THRUST // cuda + thrust (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // CUDA - using thrust::complex
@@ -347,7 +352,7 @@ namespace mg5amcCpu
 
   //==========================================================================
 
-#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_CUCOMPLEX // cuda + cucomplex
+#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_CUCOMPLEX // cuda + cucomplex (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // CUDA - using cuComplex
@@ -566,7 +571,7 @@ namespace mg5amcCpu
 
   //==========================================================================
 
-#if not defined __CUDACC__ and defined MGONGPU_CPPCXTYPE_STDCOMPLEX // c++ + stdcomplex
+#if not defined __CUDACC__ and defined MGONGPU_CPPCXTYPE_STDCOMPLEX // c++/hip + stdcomplex (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // C++ - using std::complex
@@ -633,7 +638,7 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gg_ttggg.sa/src/mgOnGpuFptypes.h b/epochX/cudacpp/gg_ttggg.sa/src/mgOnGpuFptypes.h
index 905c97d700..fa3a02664b 100644
--- a/epochX/cudacpp/gg_ttggg.sa/src/mgOnGpuFptypes.h
+++ b/epochX/cudacpp/gg_ttggg.sa/src/mgOnGpuFptypes.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUFPTYPES_H
 #define MGONGPUFPTYPES_H 1
@@ -12,7 +12,7 @@
 #include <cmath>
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL // cuda
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -20,7 +20,7 @@ namespace mg5amcCpu
 {
   //==========================================================================
 
-#ifdef __CUDACC__ // cuda
+#ifdef MGONGPUCPP_GPUIMPL // cuda
 
   //------------------------------
   // Floating point types - Cuda
@@ -64,11 +64,11 @@ namespace mg5amcCpu
 #endif
   }
 
-#endif // #ifdef __CUDACC__
+#endif // #ifdef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
   //------------------------------
   // Floating point types - C++
@@ -92,7 +92,7 @@ namespace mg5amcCpu
     return std::sqrt( f );
   }
 
-#endif // #ifndef __CUDACC__
+#endif // #ifndef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
diff --git a/epochX/cudacpp/gg_ttggg.sa/src/mgOnGpuVectors.h b/epochX/cudacpp/gg_ttggg.sa/src/mgOnGpuVectors.h
index e1299ba81e..cdae04326b 100644
--- a/epochX/cudacpp/gg_ttggg.sa/src/mgOnGpuVectors.h
+++ b/epochX/cudacpp/gg_ttggg.sa/src/mgOnGpuVectors.h
@@ -32,7 +32,7 @@
 #endif
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -131,7 +131,7 @@ namespace mg5amcCpu
 #endif
 #endif
 
-#else // i.e #ifndef MGONGPU_CPPSIMD (this includes #ifdef __CUDACC__)
+#else // i.e #ifndef MGONGPU_CPPSIMD (this includes #ifdef MGONGPUCPP_GPUIMPL)
 
   const int neppV = 1;
 
@@ -153,13 +153,13 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
   // Printout to stream for user defined types
 
@@ -805,11 +805,11 @@ namespace mg5amcCpu
 
 #endif // #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
 
-#endif // #ifndef __CUDACC__
+#endif // #ifndef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
   //------------------------------
   // Vector types - CUDA
@@ -853,12 +853,12 @@ namespace mg5amcCpu
     return mask;
   }
 
-#endif // #ifdef __CUDACC__
+#endif // #ifdef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
   // Scalar-or-vector types: scalar in CUDA, vector or scalar in C++
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   typedef bool bool_sv;
   typedef fptype fptype_sv;
   typedef fptype2 fptype2_sv;
@@ -879,7 +879,7 @@ namespace mg5amcCpu
 #endif
 
   // Scalar-or-vector zeros: scalar in CUDA, vector or scalar in C++
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   inline __host__ __device__ cxtype cxzero_sv(){ return cxtype( 0, 0 ); }
 #elif defined MGONGPU_CPPSIMD
   inline cxtype_v cxzero_sv() { return cxtype_v(); } // RRRR=0000 IIII=0000
diff --git a/epochX/cudacpp/gg_ttggg.sa/src/rambo.h b/epochX/cudacpp/gg_ttggg.sa/src/rambo.h
index e02ea52496..cd7e1008ea 100644
--- a/epochX/cudacpp/gg_ttggg.sa/src/rambo.h
+++ b/epochX/cudacpp/gg_ttggg.sa/src/rambo.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -18,7 +18,7 @@
 #include <iostream>
 
 // Simplified rambo version for 2 to N (with N>=2) processes with massless particles
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +83,7 @@ namespace mg5amcCpu
       static bool first = true;
       if( first )
       {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
         if constexpr( M_ACCESS::isOnDevice() ) // avoid
         {
           const int ievt0 = 0;
@@ -166,7 +166,7 @@ namespace mg5amcCpu
     wt = po2log;
     if( nparf != 2 ) wt = ( 2. * nparf - 4. ) * log( energy ) + z[nparf - 1];
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     // issue warnings if weight is too small or too large
     static int iwarn[5] = { 0, 0, 0, 0, 0 };
     if( wt < -180. )
diff --git a/epochX/cudacpp/gg_ttggg.sa/src/read_slha.cc b/epochX/cudacpp/gg_ttggg.sa/src/read_slha.cc
index 055b19a779..f8e46f2e66 100644
--- a/epochX/cudacpp/gg_ttggg.sa/src/read_slha.cc
+++ b/epochX/cudacpp/gg_ttggg.sa/src/read_slha.cc
@@ -11,7 +11,11 @@
 
 #include <algorithm>
 #include <cstdlib>
-#include <filesystem>
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
 #include <fstream>
 #include <iostream>
 
@@ -60,7 +64,14 @@ SLHAReader::read_slha_file( std::string file_name, bool verbose )
     {
       std::cout << "WARNING! Card file '" << file_name << "' does not exist:"
                 << " look for the file in directory $" << envpath << "='" << getenv( envpath ) << "'" << std::endl;
+      /*
+#ifdef __HIPCC__
+      const std::string file_name2 = std::experimental::filesystem::path( getenv( envpath ) ) / std::experimental::filesystem::path( file_name ).filename();
+#else
       const std::string file_name2 = std::filesystem::path( getenv( envpath ) ) / std::filesystem::path( file_name ).filename();
+#endif
+      */
+      const std::string file_name2 = std::string( getenv( envpath ) ) + "/" + file_name; // bypass std::filesystem #803
       param_card.open( file_name2.c_str(), std::ifstream::in );
       if( param_card.good() )
       {
diff --git a/epochX/cudacpp/gq_ttq.mad/CODEGEN_mad_gq_ttq_log.txt b/epochX/cudacpp/gq_ttq.mad/CODEGEN_mad_gq_ttq_log.txt
index 2c0e77fafd..cfbc521449 100644
--- a/epochX/cudacpp/gq_ttq.mad/CODEGEN_mad_gq_ttq_log.txt
+++ b/epochX/cudacpp/gq_ttq.mad/CODEGEN_mad_gq_ttq_log.txt
@@ -52,7 +52,7 @@ Note that you can still compile and run aMC@NLO with the built-in PDFs
 
 Using default text editor "vi". Set another one in ./input/mg5_configuration.txt
 Using default eps viewer "evince". Set another one in ./input/mg5_configuration.txt
-No valid web browser found. Please set in ./input/mg5_configuration.txt
+Using default web browser "firefox". Set another one in ./input/mg5_configuration.txt
 import /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_mad_gq_ttq.mg
 The import format was not given, so we guess it as command
 set stdout_level DEBUG
@@ -61,7 +61,7 @@ set zerowidth_tchannel F
 define q = u c d s u~ c~ d~ s~
 INFO: load particles 
 INFO: load vertices 
-DEBUG: model prefixing  takes 0.005677223205566406 
+DEBUG: model prefixing  takes 0.0053598880767822266 
 INFO: Restrict model sm with file models/sm/restrict_default.dat . 
 DEBUG: Simplifying conditional expressions 
 DEBUG: remove interactions: u s w+ at order: QED=1 
@@ -170,7 +170,7 @@ INFO: Crossed process found for g u~ > t t~ u~, reuse diagrams.
 INFO: Crossed process found for g c~ > t t~ c~, reuse diagrams. 
 INFO: Crossed process found for g d~ > t t~ d~, reuse diagrams. 
 INFO: Crossed process found for g s~ > t t~ s~, reuse diagrams. 
-8 processes with 40 diagrams generated in 0.080 s
+8 processes with 40 diagrams generated in 0.079 s
 Total: 8 processes with 40 diagrams
 output madevent ../TMPOUT/CODEGEN_mad_gq_ttq --hel_recycling=False --vector_size=32 --me_exporter=standalone_cudacpp
 Load PLUGIN.CUDACPP_OUTPUT
@@ -198,7 +198,7 @@ INFO: Combined process g d~ > t t~ d~ WEIGHTED<=3 @1 with process g u~ > t t~ u~
 INFO: Combined process g s~ > t t~ s~ WEIGHTED<=3 @1 with process g u~ > t t~ u~ WEIGHTED<=3 @1 
 INFO: Creating files in directory P1_gu_ttxu 
 DEBUG:  kwargs[prefix] = 0 [model_handling.py at line 1058] 
-DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7f914b38cc40> [export_v4.py at line 6262] 
+DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7fa8ada4ebe0> [export_v4.py at line 6262] 
 INFO: Creating files in directory . 
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.h
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.cc
@@ -215,7 +215,7 @@ INFO: Generating Feynman diagrams for Process: g u > t t~ u WEIGHTED<=3 @1
 INFO: Finding symmetric diagrams for subprocess group gu_ttxu 
 INFO: Creating files in directory P1_gux_ttxux 
 DEBUG:  kwargs[prefix] = 0 [model_handling.py at line 1058] 
-DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7f914b238c10> [export_v4.py at line 6262] 
+DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7fa8ada4ebe0> [export_v4.py at line 6262] 
 INFO: Creating files in directory . 
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.h
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.cc
@@ -230,17 +230,17 @@ INFO: Created files CPPProcess.h and CPPProcess.cc in directory ./.
 DEBUG:  vector, subproc_group,self.opt['vector_size'] =  32 True 32 [export_v4.py at line 1872] 
 INFO: Generating Feynman diagrams for Process: g u~ > t t~ u~ WEIGHTED<=3 @1 
 INFO: Finding symmetric diagrams for subprocess group gux_ttxux 
-Generated helas calls for 2 subprocesses (10 diagrams) in 0.032 s
-Wrote files for 32 helas calls in 0.231 s
+Generated helas calls for 2 subprocesses (10 diagrams) in 0.043 s
+Wrote files for 32 helas calls in 0.217 s
 ALOHA: aloha starts to compute helicity amplitudes
 ALOHA: aloha creates FFV1 routines
 ALOHA: aloha creates VVV1 routines
-ALOHA: aloha creates 2 routines in  0.364 s
+ALOHA: aloha creates 2 routines in  0.144 s
 DEBUG:  Entering PLUGIN_ProcessExporter.convert_model (create the model) [output.py at line 202] 
 ALOHA: aloha starts to compute helicity amplitudes
 ALOHA: aloha creates FFV1 routines
 ALOHA: aloha creates VVV1 routines
-ALOHA: aloha creates 4 routines in  0.137 s
+ALOHA: aloha creates 4 routines in  0.130 s
 <class 'aloha.create_aloha.AbstractRoutine'> FFV1
 <class 'aloha.create_aloha.AbstractRoutine'> FFV1
 <class 'aloha.create_aloha.AbstractRoutine'> FFV1
@@ -294,10 +294,10 @@ Type "launch" to generate events from this process, or see
 Run "open index.html" to see more information about this process.
 quit
 
-real	0m2.934s
-user	0m1.748s
-sys	0m0.220s
-Code generation completed in 3 seconds
+real	0m1.946s
+user	0m1.693s
+sys	0m0.230s
+Code generation completed in 2 seconds
 ************************************************************
 *                                                          *
 *                      W E L C O M E to                    *
@@ -323,7 +323,7 @@ INFO: load configuration from /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/mg5amc
 INFO: load configuration from /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_mad_gq_ttq/Cards/me5_configuration.txt  
 Using default text editor "vi". Set another one in ./input/mg5_configuration.txt
 Using default eps viewer "evince". Set another one in ./input/mg5_configuration.txt
-No valid web browser found. Please set in ./input/mg5_configuration.txt
+Using default web browser "firefox". Set another one in ./input/mg5_configuration.txt
 treatcards run
 quit
 INFO:  
@@ -353,7 +353,7 @@ INFO: load configuration from /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/mg5amc
 INFO: load configuration from /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_mad_gq_ttq/Cards/me5_configuration.txt  
 Using default text editor "vi". Set another one in ./input/mg5_configuration.txt
 Using default eps viewer "evince". Set another one in ./input/mg5_configuration.txt
-No valid web browser found. Please set in ./input/mg5_configuration.txt
+Using default web browser "firefox". Set another one in ./input/mg5_configuration.txt
 treatcards param
 quit
 INFO:  
diff --git a/epochX/cudacpp/gq_ttq.mad/COPYRIGHT b/epochX/cudacpp/gq_ttq.mad/COPYRIGHT
index a134b5fef9..9036d9260a 100644
--- a/epochX/cudacpp/gq_ttq.mad/COPYRIGHT
+++ b/epochX/cudacpp/gq_ttq.mad/COPYRIGHT
@@ -15,6 +15,7 @@ The full development team currently includes the following authors :
   Stephan Hageboeck (CERN)
   Olivier Mattelaer (Universite Catholique de Louvain, original author)
   Stefan Roiser (CERN, original author)
+  Jorgen Teig (CERN)
   Andrea Valassi (CERN, original author)
   Zenny Wettersten (CERN)
 See https://github.com/madgraph5/madgraph4gpu for more details. For the full
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/Bridge.h b/epochX/cudacpp/gq_ttq.mad/SubProcesses/Bridge.h
index bf8b5e024d..f9ed70dfde 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/Bridge.h
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/Bridge.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Roiser (Nov 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef BRIDGE_H
 #define BRIDGE_H 1
@@ -14,16 +14,23 @@
 #include "MemoryAccessMomenta.h"  // for MemoryAccessMomenta::neppM
 #include "MemoryBuffers.h"        // for HostBufferMomenta, DeviceBufferMomenta etc
 
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
+
+#include <sys/stat.h> // bypass std::filesystem #803
+
 #include <algorithm>
 #include <cassert>
 #include <cmath>
 #include <cstring>
-#include <filesystem>
 #include <iostream>
 #include <memory>
 #include <type_traits>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +90,7 @@ namespace mg5amcCpu
     Bridge& operator=( const Bridge& ) = delete;
     Bridge& operator=( Bridge&& ) = delete;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     /**
      * Set the gpublocks and gputhreads for the gpusequence - throws if evnt != gpublocks*gputhreads
      * (this is needed for BridgeKernel tests rather than for actual production use in Fortran)
@@ -150,7 +157,7 @@ namespace mg5amcCpu
     unsigned int m_nevt; // number of events
     int m_nGoodHel;      // the number of good helicities (-1 initially when they have not yet been calculated)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     int m_gputhreads; // number of gpu threads (default set from number of events, can be modified)
     int m_gpublocks;  // number of gpu blocks (default set from number of events, can be modified)
     DeviceBuffer<FORTRANFPTYPE, sizePerEventMomenta> m_devMomentaF;
@@ -187,12 +194,12 @@ namespace mg5amcCpu
   // Forward declare transposition methods
   //
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
   template<typename Tin, typename Tout>
   __global__ void dev_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt );
 
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
 
   template<typename Tin, typename Tout>
   void hst_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt );
@@ -209,7 +216,7 @@ namespace mg5amcCpu
   Bridge<FORTRANFPTYPE>::Bridge( unsigned int nevtF, unsigned int nparF, unsigned int np4F )
     : m_nevt( nevtF )
     , m_nGoodHel( -1 )
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     , m_gputhreads( 256 )                  // default number of gpu threads
     , m_gpublocks( m_nevt / m_gputhreads ) // this ensures m_nevt <= m_gpublocks*m_gputhreads
     , m_devMomentaF( m_nevt )
@@ -233,7 +240,7 @@ namespace mg5amcCpu
   {
     if( nparF != CPPProcess::npar ) throw std::runtime_error( "Bridge constructor: npar mismatch" );
     if( np4F != CPPProcess::np4 ) throw std::runtime_error( "Bridge constructor: np4 mismatch" );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( ( m_nevt < s_gputhreadsmin ) || ( m_nevt % s_gputhreadsmin != 0 ) )
       throw std::runtime_error( "Bridge constructor: nevt should be a multiple of " + std::to_string( s_gputhreadsmin ) );
     while( m_nevt != m_gpublocks * m_gputhreads )
@@ -249,20 +256,28 @@ namespace mg5amcCpu
 #else
     std::cout << "WARNING! Instantiate host Bridge (nevt=" << m_nevt << ")" << std::endl;
     m_pmek.reset( new MatrixElementKernelHost( m_hstMomentaC, m_hstGs, m_hstRndHel, m_hstRndCol, m_hstMEs, m_hstSelHel, m_hstSelCol, m_nevt ) );
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
     // Create a process object, read param card and set parameters
     // FIXME: the process instance can happily go out of scope because it is only needed to read parameters?
     // FIXME: the CPPProcess should really be a singleton? what if fbridgecreate is called from several Fortran threads?
     CPPProcess process( /*verbose=*/false );
     std::string paramCard = "../../Cards/param_card.dat";
-    if( !std::filesystem::exists( paramCard ) )
-    {
-      paramCard = "../" + paramCard;
-    }
+    /*
+#ifdef __HIPCC__
+    if( !std::experimental::filesystem::exists( paramCard ) ) paramCard = "../" + paramCard;
+#else
+    if( !std::filesystem::exists( paramCard ) ) paramCard = "../" + paramCard;
+#endif
+    */
+    //struct stat dummybuffer; // bypass std::filesystem #803
+    //if( !( stat( paramCard.c_str(), &dummyBuffer ) == 0 ) ) paramCard = "../" + paramCard; //
+    auto fileExists = []( std::string& fileName )
+    { struct stat buffer; return stat( fileName.c_str(), &buffer ) == 0; };
+    if( !fileExists( paramCard ) ) paramCard = "../" + paramCard; // bypass std::filesystem #803
     process.initProc( paramCard );
   }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::set_gpugrid( const int gpublocks, const int gputhreads )
   {
@@ -276,7 +291,7 @@ namespace mg5amcCpu
   }
 #endif
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::gpu_sequence( const FORTRANFPTYPE* momenta,
                                             const FORTRANFPTYPE* gs,
@@ -291,14 +306,14 @@ namespace mg5amcCpu
     constexpr int neppM = MemoryAccessMomenta::neppM;
     if constexpr( neppM == 1 && std::is_same_v<FORTRANFPTYPE, fptype> )
     {
-      checkCuda( cudaMemcpy( m_devMomentaC.data(), momenta, m_devMomentaC.bytes(), cudaMemcpyHostToDevice ) );
+      gpuMemcpy( m_devMomentaC.data(), momenta, m_devMomentaC.bytes(), gpuMemcpyHostToDevice );
     }
     else
     {
-      checkCuda( cudaMemcpy( m_devMomentaF.data(), momenta, m_devMomentaF.bytes(), cudaMemcpyHostToDevice ) );
+      gpuMemcpy( m_devMomentaF.data(), momenta, m_devMomentaF.bytes(), gpuMemcpyHostToDevice );
       const int thrPerEvt = CPPProcess::npar * CPPProcess::np4; // AV: transpose alg does 1 element per thread (NOT 1 event per thread)
       //const int thrPerEvt = 1; // AV: try new alg with 1 event per thread... this seems slower
-      dev_transposeMomentaF2C<<<m_gpublocks * thrPerEvt, m_gputhreads>>>( m_devMomentaF.data(), m_devMomentaC.data(), m_nevt );
+      gpuLaunchKernel( dev_transposeMomentaF2C, m_gpublocks * thrPerEvt, m_gputhreads, m_devMomentaF.data(), m_devMomentaC.data(), m_nevt );
     }
     if constexpr( std::is_same_v<FORTRANFPTYPE, fptype> )
     {
@@ -341,7 +356,7 @@ namespace mg5amcCpu
   }
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::cpu_sequence( const FORTRANFPTYPE* momenta,
                                             const FORTRANFPTYPE* gs,
@@ -396,7 +411,7 @@ namespace mg5amcCpu
   // - C++ array: momenta[npagM][npar][np4][neppM] with nevt=npagM*neppM (AOSOA)
   //
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename Tin, typename Tout>
   __global__ void dev_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt )
   {
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/BridgeKernels.cc b/epochX/cudacpp/gq_ttq.mad/SubProcesses/BridgeKernels.cc
index d58066c9c1..eaf4037a24 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/BridgeKernels.cc
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/BridgeKernels.cc
@@ -1,17 +1,18 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "BridgeKernels.h"
 
+#include "GpuAbstraction.h"
 #include "MemoryAccessMomenta.h"
 
 #include <sstream>
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -45,7 +46,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu
 {
 
@@ -96,7 +97,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/BridgeKernels.h b/epochX/cudacpp/gq_ttq.mad/SubProcesses/BridgeKernels.h
index 15eb4bff4d..3efef8ce97 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/BridgeKernels.h
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/BridgeKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef BRIDGEKERNELS_H
 #define BRIDGEKERNELS_H 1
@@ -12,7 +12,7 @@
 #include "MatrixElementKernels.h"
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -49,7 +49,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A Bridge wrapper class encapsulating matrix element calculations on a CPU host
   class BridgeKernelHost final : public BridgeKernelBase
   {
@@ -89,7 +89,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A Bridge wrapper class encapsulating matrix element calculations on a GPU device
   class BridgeKernelDevice : public BridgeKernelBase
   {
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/CommonRandomNumberKernel.cc b/epochX/cudacpp/gq_ttq.mad/SubProcesses/CommonRandomNumberKernel.cc
index 985b39f576..010bc4cbd0 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/CommonRandomNumberKernel.cc
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/CommonRandomNumberKernel.cc
@@ -1,15 +1,16 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "CommonRandomNumbers.h"
+#include "GpuAbstraction.h"
 #include "MemoryBuffers.h"
 #include "RandomNumberKernels.h"
 
 #include <cassert>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/CrossSectionKernels.cc b/epochX/cudacpp/gq_ttq.mad/SubProcesses/CrossSectionKernels.cc
index 0b355a3c8d..c15b39844d 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/CrossSectionKernels.cc
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/CrossSectionKernels.cc
@@ -1,10 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "CrossSectionKernels.h"
 
+#include "GpuAbstraction.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessWeights.h"
 #include "MemoryBuffers.h"
@@ -77,7 +78,7 @@ debug_me_is_abnormal( const fptype& me, size_t ievtALL )
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -185,7 +186,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/CrossSectionKernels.h b/epochX/cudacpp/gq_ttq.mad/SubProcesses/CrossSectionKernels.h
index 7933ca4bbf..4d9659e04e 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/CrossSectionKernels.h
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/CrossSectionKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef CROSSSECTIONKERNELS_H
 #define CROSSSECTIONKERNELS_H 1
@@ -13,7 +13,7 @@
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -96,7 +96,7 @@ namespace mg5amcCpu
   //--------------------------------------------------------------------------
 
   /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating the calculation of event statistics on a GPU device
   class CrossSectionKernelDevice : public CrossSectionKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/CudaRuntime.h b/epochX/cudacpp/gq_ttq.mad/SubProcesses/CudaRuntime.h
deleted file mode 100644
index 64ce52f4b3..0000000000
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/CudaRuntime.h
+++ /dev/null
@@ -1,85 +0,0 @@
-// Copyright (C) 2020-2023 CERN and UCLouvain.
-// Licensed under the GNU Lesser General Public License (version 3 or later).
-// Created by: S. Roiser (Jul 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
-
-#ifndef MG5AMC_CUDARUNTIME_H
-#define MG5AMC_CUDARUNTIME_H 1
-
-// MG5AMC on GPU uses the CUDA runtime API, not the lower level CUDA driver API
-// See https://docs.nvidia.com/cuda/cuda-runtime-api/driver-vs-runtime-api.html#driver-vs-runtime-api
-
-#include <cassert>
-#include <iostream>
-
-//--------------------------------------------------------------------------
-
-// See https://stackoverflow.com/a/14038590
-#ifdef __CUDACC__ /* clang-format off */
-#define checkCuda( code ) { assertCuda( code, __FILE__, __LINE__ ); }
-inline void assertCuda( cudaError_t code, const char* file, int line, bool abort = true )
-{
-  if( code != cudaSuccess )
-  {
-    printf( "ERROR! assertCuda: '%s' (%d) in %s:%d\n", cudaGetErrorString( code ), code, file, line );
-    if( abort ) assert( code == cudaSuccess );
-  }
-}
-#endif /* clang-format on */
-
-//--------------------------------------------------------------------------
-
-#ifdef __CUDACC__
-namespace mg5amcGpu
-{
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  // *** FIXME! This will all need to be designed differently when going to multi-GPU nodes! ***
-  struct CudaRuntime final
-  {
-    CudaRuntime( const bool debug = true )
-      : m_debug( debug ) { setUp( m_debug ); }
-    ~CudaRuntime() { tearDown( m_debug ); }
-    CudaRuntime( const CudaRuntime& ) = delete;
-    CudaRuntime( CudaRuntime&& ) = delete;
-    CudaRuntime& operator=( const CudaRuntime& ) = delete;
-    CudaRuntime& operator=( CudaRuntime&& ) = delete;
-    bool m_debug;
-
-    // Set up CUDA application
-    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
-    // Calling cudaSetDevice on startup is useful to properly book-keep the time spent in CUDA initialization
-    static void setUp( const bool debug = true )
-    {
-      // ** NB: it is useful to call cudaSetDevice, or cudaFree, to properly book-keep the time spent in CUDA initialization
-      // ** NB: otherwise, the first CUDA operation (eg a cudaMemcpyToSymbol in CPPProcess ctor) appears to take much longer!
-      /*
-      // [We initially added cudaFree(0) to "ease profile analysis" only because it shows up as a big recognizable block!]
-      // No explicit initialization is needed: https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#initialization
-      // It is not clear what cudaFree(0) does at all: https://stackoverflow.com/questions/69967813/
-      if ( debug ) std::cout << "__CudaRuntime: calling cudaFree(0)" << std::endl;
-      checkCuda( cudaFree( 0 ) ); // SLOW!
-      */
-      // Replace cudaFree(0) by cudaSetDevice(0), even if it is not really needed either
-      // (but see https://developer.nvidia.com/blog/cuda-pro-tip-always-set-current-device-avoid-multithreading-bugs)
-      if( debug ) std::cout << "__CudaRuntime: calling cudaSetDevice(0)" << std::endl;
-      checkCuda( cudaSetDevice( 0 ) ); // SLOW!
-    }
-
-    // Tear down CUDA application (call cudaDeviceReset)
-    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
-    // Calling cudaDeviceReset on shutdown is only needed for checking memory leaks in cuda-memcheck
-    // See https://docs.nvidia.com/cuda/cuda-memcheck/index.html#leak-checking
-    static void tearDown( const bool debug = true )
-    {
-      if( debug ) std::cout << "__CudaRuntime: calling cudaDeviceReset()" << std::endl;
-      checkCuda( cudaDeviceReset() );
-    }
-  };
-
-}
-#endif
-
-//--------------------------------------------------------------------------
-
-#endif // MG5AMC_CUDARUNTIME_H
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/CurandRandomNumberKernel.cc b/epochX/cudacpp/gq_ttq.mad/SubProcesses/CurandRandomNumberKernel.cc
index eb56333b03..08a16f6f2c 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/CurandRandomNumberKernel.cc
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/CurandRandomNumberKernel.cc
@@ -1,9 +1,9 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "MemoryBuffers.h"
 #include "RandomNumberKernels.h"
 
@@ -22,7 +22,7 @@ inline void assertCurand( curandStatus_t code, const char *file, int line, bool
 }
 #endif /* clang-format on */
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -36,7 +36,7 @@ namespace mg5amcCpu
   {
     if( m_isOnDevice )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if( !m_rnarray.isOnDevice() )
         throw std::runtime_error( "CurandRandomNumberKernel on device with a host random number array" );
 #else
@@ -114,7 +114,7 @@ namespace mg5amcCpu
     /*
     printf( "\nCurandRandomNumberKernel::generateRnarray size = %d\n", (int)m_rnarray.size() );
     fptype* data = m_rnarray.data();
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( m_rnarray.isOnDevice() )
     {
       data = new fptype[m_rnarray.size()]();
@@ -123,7 +123,7 @@ namespace mg5amcCpu
 #endif
     for( int i = 0; i < ( (int)m_rnarray.size() / 4 ); i++ )
       printf( "[%4d] %f %f %f %f\n", i * 4, data[i * 4], data[i * 4 + 2], data[i * 4 + 2], data[i * 4 + 3] );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( m_rnarray.isOnDevice() ) delete[] data;
 #endif
     */
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/EventStatistics.h b/epochX/cudacpp/gq_ttq.mad/SubProcesses/EventStatistics.h
index 48b51e0a49..b425a5bade 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/EventStatistics.h
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/EventStatistics.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef EventStatistics_H
 #define EventStatistics_H 1
@@ -16,7 +16,7 @@
 #include <limits>
 #include <string>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/GpuAbstraction.h b/epochX/cudacpp/gq_ttq.mad/SubProcesses/GpuAbstraction.h
new file mode 100644
index 0000000000..9c467b1e04
--- /dev/null
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/GpuAbstraction.h
@@ -0,0 +1,69 @@
+// Copyright (C) 2020-2023 CERN and UCLouvain.
+// Licensed under the GNU Lesser General Public License (version 3 or later).
+// Created by: J. Teig (Jul 2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+
+#ifndef MG5AMC_GPUABSTRACTION_H
+#define MG5AMC_GPUABSTRACTION_H 1
+
+#include <cassert>
+
+//--------------------------------------------------------------------------
+
+#ifdef __CUDACC__
+
+#define gpuError_t cudaError_t
+#define gpuPeekAtLastError cudaPeekAtLastError
+#define gpuGetErrorString cudaGetErrorString
+#define gpuSuccess cudaSuccess
+
+#define gpuMallocHost( ptr, size ) checkGpu( cudaMallocHost( ptr, size ) )
+#define gpuMalloc( ptr, size ) checkGpu( cudaMalloc( ptr, size ) )
+
+#define gpuMemcpy( dstData, srcData, srcBytes, func ) checkGpu( cudaMemcpy( dstData, srcData, srcBytes, func ) )
+#define gpuMemcpyHostToDevice cudaMemcpyHostToDevice
+#define gpuMemcpyDeviceToHost cudaMemcpyDeviceToHost
+#define gpuMemcpyToSymbol( type1, type2, size ) checkGpu( cudaMemcpyToSymbol( type1, type2, size ) )
+
+#define gpuFree( ptr ) checkGpu( cudaFree( ptr ) )
+#define gpuFreeHost( ptr ) checkGpu( cudaFreeHost( ptr ) )
+
+#define gpuSetDevice cudaSetDevice
+#define gpuDeviceSynchronize cudaDeviceSynchronize
+#define gpuDeviceReset cudaDeviceReset
+
+#define gpuLaunchKernel( kernel, blocks, threads, ... ) kernel<<<blocks, threads>>>( __VA_ARGS__ )
+#define gpuLaunchKernelSharedMem( kernel, blocks, threads, sharedMem, ... ) kernel<<<blocks, threads, sharedMem>>>( __VA_ARGS__ )
+
+//--------------------------------------------------------------------------
+
+#elif defined __HIPCC__
+
+#define gpuError_t hipError_t
+#define gpuPeekAtLastError hipPeekAtLastError
+#define gpuGetErrorString hipGetErrorString
+#define gpuSuccess hipSuccess
+
+#define gpuMallocHost( ptr, size ) checkGpu( hipHostMalloc( ptr, size ) ) // HostMalloc better
+#define gpuMalloc( ptr, size ) checkGpu( hipMalloc( ptr, size ) )
+
+#define gpuMemcpy( dstData, srcData, srcBytes, func ) checkGpu( hipMemcpy( dstData, srcData, srcBytes, func ) )
+#define gpuMemcpyHostToDevice hipMemcpyHostToDevice
+#define gpuMemcpyDeviceToHost hipMemcpyDeviceToHost
+#define gpuMemcpyToSymbol( type1, type2, size ) checkGpu( hipMemcpyToSymbol( type1, type2, size ) )
+
+#define gpuFree( ptr ) checkGpu( hipFree( ptr ) )
+#define gpuFreeHost( ptr ) checkGpu( hipHostFree( ptr ) )
+
+#define gpuSetDevice hipSetDevice
+#define gpuDeviceSynchronize hipDeviceSynchronize
+#define gpuDeviceReset hipDeviceReset
+
+#define gpuLaunchKernel( kernel, blocks, threads, ... ) kernel<<<blocks, threads>>>( __VA_ARGS__ )
+#define gpuLaunchKernelSharedMem( kernel, blocks, threads, sharedMem, ... ) kernel<<<blocks, threads, sharedMem>>>( __VA_ARGS__ )
+
+//--------------------------------------------------------------------------
+
+#endif
+
+#endif // MG5AMC_GPUABSTRACTION_H
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/GpuRuntime.h b/epochX/cudacpp/gq_ttq.mad/SubProcesses/GpuRuntime.h
new file mode 100644
index 0000000000..93579ef08b
--- /dev/null
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/GpuRuntime.h
@@ -0,0 +1,85 @@
+// Copyright (C) 2020-2023 CERN and UCLouvain.
+// Licensed under the GNU Lesser General Public License (version 3 or later).
+// Created by: J. Teig (Jun 2023, based on earlier work by S. Roiser) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+
+#ifndef MG5AMC_GPURUNTIME_H
+#define MG5AMC_GPURUNTIME_H 1
+
+// MG5AMC on GPU uses the CUDA runtime API, not the lower level CUDA driver API
+// See https://docs.nvidia.com/cuda/cuda-runtime-api/driver-vs-runtime-api.html#driver-vs-runtime-api
+
+#include "GpuAbstraction.h"
+
+#include <iostream>
+
+//--------------------------------------------------------------------------
+
+// See https://stackoverflow.com/a/14038590
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
+#define checkGpu( code ) { assertGpu( code, __FILE__, __LINE__ ); }
+inline void assertGpu( gpuError_t code, const char* file, int line, bool abort = true )
+{
+  if( code != gpuSuccess )
+  {
+    printf( "ERROR! assertGpu: '%s' (%d) in %s:%d\n", gpuGetErrorString( code ), code, file, line );
+    if( abort ) assert( code == gpuSuccess );
+  }
+}
+#endif /* clang-format on */
+
+//--------------------------------------------------------------------------
+
+#ifdef MGONGPUCPP_GPUIMPL
+namespace mg5amcGpu
+{
+  // Instantiate a GpuRuntime at the beginnining of the application's main to
+  // invoke gpuSetDevice(0) in the constructor and book a gpuDeviceReset() call in the destructor
+  // *** FIXME! This will all need to be designed differently when going to multi-GPU nodes! ***
+  struct GpuRuntime final
+  {
+    GpuRuntime( const bool debug = true )
+      : m_debug( debug ) { setUp( m_debug ); }
+    ~GpuRuntime() { tearDown( m_debug ); }
+    GpuRuntime( const GpuRuntime& ) = delete;
+    GpuRuntime( GpuRuntime&& ) = delete;
+    GpuRuntime& operator=( const GpuRuntime& ) = delete;
+    GpuRuntime& operator=( GpuRuntime&& ) = delete;
+    bool m_debug;
+
+    // Set up CUDA application
+    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
+    // Calling cudaSetDevice on startup is useful to properly book-keep the time spent in CUDA initialization
+    static void setUp( const bool debug = true )
+    {
+      // ** NB: it is useful to call cudaSetDevice, or cudaFree, to properly book-keep the time spent in CUDA initialization
+      // ** NB: otherwise, the first CUDA operation (eg a cudaMemcpyToSymbol in CPPProcess ctor) appears to take much longer!
+      /*
+      // [We initially added cudaFree(0) to "ease profile analysis" only because it shows up as a big recognizable block!]
+      // No explicit initialization is needed: https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#initialization
+      // It is not clear what cudaFree(0) does at all: https://stackoverflow.com/questions/69967813/
+      if ( debug ) std::cout << "__CudaRuntime: calling cudaFree(0)" << std::endl;
+      checkCuda( cudaFree( 0 ) ); // SLOW!
+      */
+      // Replace cudaFree(0) by cudaSetDevice(0), even if it is not really needed either
+      // (but see https://developer.nvidia.com/blog/cuda-pro-tip-always-set-current-device-avoid-multithreading-bugs)
+      if( debug ) std::cout << "__GpuRuntime: calling GpuSetDevice(0)" << std::endl;
+      checkGpu( gpuSetDevice( 0 ) ); // SLOW!
+    }
+
+    // Tear down CUDA application (call cudaDeviceReset)
+    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
+    // Calling cudaDeviceReset on shutdown is only needed for checking memory leaks in cuda-memcheck
+    // See https://docs.nvidia.com/cuda/cuda-memcheck/index.html#leak-checking
+    static void tearDown( const bool debug = true )
+    {
+      if( debug ) std::cout << "__GpuRuntime: calling GpuDeviceReset()" << std::endl;
+      checkGpu( gpuDeviceReset() );
+    }
+  };
+}
+#endif
+
+//--------------------------------------------------------------------------
+
+#endif // MG5AMC_GPURUNTIME_H
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/MadgraphTest.h b/epochX/cudacpp/gq_ttq.mad/SubProcesses/MadgraphTest.h
index ef40624c88..6054185300 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/MadgraphTest.h
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/MadgraphTest.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Hageboeck (Dec 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MADGRAPHTEST_H_
 #define MADGRAPHTEST_H_ 1
@@ -14,7 +14,11 @@
 
 #include <array>
 #include <cmath>
-#include <filesystem>
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
 #include <fstream>
 #include <iomanip>
 #include <memory>
@@ -22,7 +26,7 @@
 #include <string>
 #include <vector>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using mg5amcGpu::CPPProcess;
 #else
 using mg5amcCpu::CPPProcess;
@@ -201,7 +205,7 @@ class MadgraphTest : public testing::TestWithParam<TestDriverBase*>
 
 // Since we link both the CPU-only and GPU tests into the same executable, we prevent
 // a multiply defined symbol by only compiling this in the non-CUDA phase:
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
 /// Compare momenta and matrix elements.
 /// This uses an implementation of TestDriverBase to run a madgraph workflow,
@@ -219,7 +223,14 @@ TEST_P( MadgraphTest, CompareMomentaAndME )
   const char* dumpEventsC = getenv( "CUDACPP_RUNTEST_DUMPEVENTS" );
   const bool dumpEvents = ( dumpEventsC != 0 ) && ( std::string( dumpEventsC ) != "" );
   const std::string refFileName = testDriver->getRefFileName();
+  /*
+#ifdef __HIPCC__
+  const std::string dumpFileName = std::experimental::filesystem::path( refFileName ).filename();
+#else
   const std::string dumpFileName = std::filesystem::path( refFileName ).filename();
+#endif
+  */
+  const std::string dumpFileName = refFileName; // bypass std::filesystem #803
   std::ofstream dumpFile;
   if( dumpEvents )
   {
@@ -307,6 +318,6 @@ TEST_P( MadgraphTest, CompareMomentaAndME )
   }
 }
 
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
 
 #endif /* MADGRAPHTEST_H_ */
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/MatrixElementKernels.cc b/epochX/cudacpp/gq_ttq.mad/SubProcesses/MatrixElementKernels.cc
index 74b5239ebf..81699dfea9 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/MatrixElementKernels.cc
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/MatrixElementKernels.cc
@@ -1,12 +1,12 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "MatrixElementKernels.h"
 
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h" // Includes the abstraction for Nvidia/AMD compilation
 #include "MemoryAccessMomenta.h"
 #include "MemoryBuffers.h"
 
@@ -14,7 +14,7 @@
 
 //============================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu
 {
 
@@ -150,7 +150,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
@@ -209,13 +209,13 @@ namespace mg5amcGpu
     PinnedHostBufferHelicityMask hstIsGoodHel( ncomb );
     DeviceBufferHelicityMask devIsGoodHel( ncomb );
     // ... 0d1. Compute good helicity mask on the device
-    computeDependentCouplings<<<m_gpublocks, m_gputhreads>>>( m_gs.data(), m_couplings.data() );
+    gpuLaunchKernel( computeDependentCouplings, m_gpublocks, m_gputhreads, m_gs.data(), m_couplings.data() );
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-    sigmaKin_getGoodHel<<<m_gpublocks, m_gputhreads>>>( m_momenta.data(), m_couplings.data(), m_matrixElements.data(), m_numerators.data(), m_denominators.data(), devIsGoodHel.data() );
+    gpuLaunchKernel( sigmaKin_getGoodHel, m_gpublocks, m_gputhreads, m_momenta.data(), m_couplings.data(), m_matrixElements.data(), m_numerators.data(), m_denominators.data(), devIsGoodHel.data() );
 #else
-    sigmaKin_getGoodHel<<<m_gpublocks, m_gputhreads>>>( m_momenta.data(), m_couplings.data(), m_matrixElements.data(), devIsGoodHel.data() );
+    gpuLaunchKernel( sigmaKin_getGoodHel, m_gpublocks, m_gputhreads, m_momenta.data(), m_couplings.data(), m_matrixElements.data(), devIsGoodHel.data() );
 #endif
-    checkCuda( cudaPeekAtLastError() );
+    checkGpu( gpuPeekAtLastError() );
     // ... 0d2. Copy back good helicity mask to the host
     copyHostFromDevice( hstIsGoodHel, devIsGoodHel );
     // ... 0d3. Copy back good helicity list to constant memory on the device
@@ -226,19 +226,19 @@ namespace mg5amcGpu
 
   void MatrixElementKernelDevice::computeMatrixElements( const unsigned int channelId )
   {
-    computeDependentCouplings<<<m_gpublocks, m_gputhreads>>>( m_gs.data(), m_couplings.data() );
+    gpuLaunchKernel( computeDependentCouplings, m_gpublocks, m_gputhreads, m_gs.data(), m_couplings.data() );
 #ifndef MGONGPU_NSIGHT_DEBUG
     constexpr unsigned int sharedMemSize = 0;
 #else
     constexpr unsigned int sharedMemSize = ntpbMAX * sizeof( float );
 #endif
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-    sigmaKin<<<m_gpublocks, m_gputhreads, sharedMemSize>>>( m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), channelId, m_numerators.data(), m_denominators.data(), m_selhel.data(), m_selcol.data() );
+    gpuLaunchKernelSharedMem( sigmaKin, m_gpublocks, m_gputhreads, sharedMemSize, m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), channelId, m_numerators.data(), m_denominators.data(), m_selhel.data(), m_selcol.data() );
 #else
-    sigmaKin<<<m_gpublocks, m_gputhreads, sharedMemSize>>>( m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), m_selhel.data(), m_selcol.data() );
+    gpuLaunchKernelSharedMem( sigmaKin, m_gpublocks, m_gputhreads, sharedMemSize, m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), m_selhel.data(), m_selcol.data() );
 #endif
-    checkCuda( cudaPeekAtLastError() );
-    checkCuda( cudaDeviceSynchronize() );
+    checkGpu( gpuPeekAtLastError() );
+    checkGpu( gpuDeviceSynchronize() );
   }
 
   //--------------------------------------------------------------------------
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/MatrixElementKernels.h b/epochX/cudacpp/gq_ttq.mad/SubProcesses/MatrixElementKernels.h
index 23e84757a2..72bd8f195b 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/MatrixElementKernels.h
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/MatrixElementKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MATRIXELEMENTKERNELS_H
 #define MATRIXELEMENTKERNELS_H 1
@@ -10,7 +10,7 @@
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -81,7 +81,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating matrix element calculations on a CPU host
   class MatrixElementKernelHost final : public MatrixElementKernelBase, public NumberOfEvents
   {
@@ -130,7 +130,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating matrix element calculations on a GPU device
   class MatrixElementKernelDevice : public MatrixElementKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessAmplitudes.h b/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessAmplitudes.h
index 573b3bbbc9..ffb76e93de 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessAmplitudes.h
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessAmplitudes.h
@@ -15,7 +15,7 @@
 #define MGONGPU_TRIVIAL_AMPLITUDES 1
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessCouplings.h b/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessCouplings.h
index 35a3af42e0..3afdf3e554 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessCouplings.h
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessCouplings.h
@@ -15,7 +15,7 @@
 #include "MemoryBuffers.h"       // for HostBufferCouplings::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessCouplingsFixed.h b/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessCouplingsFixed.h
index dc0d93afff..ffcdf4dbef 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessCouplingsFixed.h
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessCouplingsFixed.h
@@ -14,7 +14,7 @@
 //#include "MemoryAccessHelpers.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessDenominators.h b/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessDenominators.h
index 3bce635718..66f2d32a6b 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessDenominators.h
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessDenominators.h
@@ -10,7 +10,7 @@
 #include "MemoryAccessGs.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessGs.h b/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessGs.h
index 31311aa375..4c726b30f3 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessGs.h
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessGs.h
@@ -13,7 +13,7 @@
 #include "MemoryBuffers.h" // for HostBufferMatrixElements::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessHelpers.h b/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessHelpers.h
index c82a6c7635..db73e4e064 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessHelpers.h
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessHelpers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessHelpers_H
 #define MemoryAccessHelpers_H 1
@@ -105,7 +105,7 @@ class KernelAccessHelper : public MemoryAccessHelper<T>
     }
     else
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
       //printf( "kernelAccessRecord: ievt=%d threadId=%d\n", ievt, threadIdx.x );
       return T::ieventAccessRecord( buffer, ievt ); // NB fptype and fptype_sv coincide for CUDA
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessMatrixElements.h b/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessMatrixElements.h
index f32e6fea5b..3741011971 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessMatrixElements.h
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessMatrixElements.h
@@ -13,7 +13,7 @@
 #include "MemoryBuffers.h" // for HostBufferMatrixElements::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessMomenta.h b/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessMomenta.h
index 29266de32c..3be229d392 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessMomenta.h
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessMomenta.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessMomenta_H
 #define MemoryAccessMomenta_H 1
@@ -13,7 +13,7 @@
 #include "MemoryAccessVectors.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -30,7 +30,7 @@ namespace mg5amcCpu
 
     // Number of Events Per Page in the momenta AOSOA memory buffer layout
     // (these are all best kept as a compile-time constants: see issue #23)
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
     // -----------------------------------------------------------------------------------------------
     // --- GPUs: neppM is best set to a power of 2 times the number of fptype's in a 32-byte cacheline
     // --- This is relevant to ensure coalesced access to momenta in global memory
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessNumerators.h b/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessNumerators.h
index b152183b28..18991f4fa6 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessNumerators.h
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessNumerators.h
@@ -10,7 +10,7 @@
 #include "MemoryAccessGs.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessRandomNumbers.h b/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessRandomNumbers.h
index e2988d39f3..40cb089135 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessRandomNumbers.h
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessRandomNumbers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessRandomNumbers_H
 #define MemoryAccessRandomNumbers_H 1
@@ -11,7 +11,7 @@
 #include "CPPProcess.h"
 #include "MemoryAccessHelpers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using mg5amcGpu::CPPProcess;
 #else
 using mg5amcCpu::CPPProcess;
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessVectors.h b/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessVectors.h
index e9b197368e..08faccff0f 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessVectors.h
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessVectors.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessVectors_H
 #define MemoryAccessVectors_H 1
@@ -10,7 +10,7 @@
 
 #include "mgOnGpuVectors.h"
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu // this is only needed for CPU SIMD vectorization
 {
 
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessWavefunctions.h b/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessWavefunctions.h
index 5428aaf933..33bef4559e 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessWavefunctions.h
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryAccessWavefunctions.h
@@ -15,7 +15,7 @@
 #define MGONGPU_TRIVIAL_WAVEFUNCTIONS 1
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryBuffers.h b/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryBuffers.h
index 3093e6ed18..7756a71621 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryBuffers.h
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/MemoryBuffers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021, based on earlier work by S. Hageboeck) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryBuffers_H
 #define MemoryBuffers_H 1
@@ -11,12 +11,12 @@
 #include "mgOnGpuCxtypes.h"
 
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "Parameters_sm.h"
 
 #include <sstream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -87,7 +87,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   constexpr bool HostBufferALIGNED = false;   // ismisaligned=false
   constexpr bool HostBufferMISALIGNED = true; // ismisaligned=true
 
@@ -119,7 +119,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA pinned host buffer
   template<typename T>
   class PinnedHostBufferBase : public BufferBase<T>
@@ -128,18 +128,18 @@ namespace mg5amcCpu
     PinnedHostBufferBase( const size_t size )
       : BufferBase<T>( size, false )
     {
-      checkCuda( cudaMallocHost( &( this->m_data ), this->bytes() ) );
+      gpuMallocHost( &( this->m_data ), this->bytes() );
     }
     virtual ~PinnedHostBufferBase()
     {
-      checkCuda( cudaFreeHost( this->m_data ) );
+      gpuFreeHost( this->m_data );
     }
   };
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA device buffer
   template<typename T>
   class DeviceBufferBase : public BufferBase<T>
@@ -148,18 +148,18 @@ namespace mg5amcCpu
     DeviceBufferBase( const size_t size )
       : BufferBase<T>( size, true )
     {
-      checkCuda( cudaMalloc( &( this->m_data ), this->bytes() ) );
+      gpuMalloc( &( this->m_data ), this->bytes() );
     }
     virtual ~DeviceBufferBase()
     {
-      checkCuda( cudaFree( this->m_data ) );
+      gpuFree( this->m_data );
     }
   };
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for a given number of events
   template<typename T, size_t sizePerEvent, bool ismisaligned>
   class HostBuffer : public HostBufferBase<T, ismisaligned>, virtual private NumberOfEvents
@@ -175,7 +175,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA pinned host buffer for a given number of events
   template<typename T, size_t sizePerEvent>
   class PinnedHostBuffer : public PinnedHostBufferBase<T>, virtual private NumberOfEvents
@@ -191,7 +191,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA device buffer for a given number of events
   template<typename T, size_t sizePerEvent>
   class DeviceBuffer : public DeviceBufferBase<T>, virtual private NumberOfEvents
@@ -213,7 +213,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for momenta random numbers
   constexpr size_t sizePerEventRndNumMomenta = MemoryBuffers::np4 * MemoryBuffers::nparf;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for momenta random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumMomenta, HostBufferALIGNED> HostBufferRndNumMomenta;
 #else
@@ -232,7 +232,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer with ONE fptype per event
   constexpr size_t sizePerEventOneFp = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer with ONE fptype per event
   typedef HostBuffer<fptype, sizePerEventOneFp, HostBufferALIGNED> HostBufferOneFp;
 #else
@@ -257,7 +257,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for Gs
   constexpr size_t sizePerEventGs = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventGs, HostBufferALIGNED> HostBufferGs;
 #else
@@ -276,7 +276,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for numerators
   constexpr size_t sizePerEventNumerators = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventNumerators, HostBufferALIGNED> HostBufferNumerators;
 #else
@@ -296,7 +296,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for denominators
   constexpr size_t sizePerEventDenominators = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventDenominators, HostBufferALIGNED> HostBufferDenominators;
 #else
@@ -315,7 +315,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for random numbers
   constexpr size_t sizePerEventCouplings = MemoryBuffers::ndcoup * MemoryBuffers::nx2;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventCouplings, HostBufferALIGNED> HostBufferCouplings;
 #else
@@ -333,7 +333,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for momenta
   constexpr size_t sizePerEventMomenta = MemoryBuffers::np4 * MemoryBuffers::npar;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for momenta
   typedef HostBuffer<fptype, sizePerEventMomenta, HostBufferALIGNED> HostBufferMomenta;
   //typedef HostBuffer<fptype, sizePerEventMomenta, HostBufferMISALIGNED> HostBufferMomenta; // TEST MISALIGNMENT!
@@ -352,7 +352,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for sampling weights
   constexpr size_t sizePerEventWeights = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for sampling weights
   typedef HostBuffer<fptype, sizePerEventWeights, HostBufferALIGNED> HostBufferWeights;
 #else
@@ -370,7 +370,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for matrix elements
   constexpr size_t sizePerEventMatrixElements = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for matrix elements
   typedef HostBuffer<fptype, sizePerEventMatrixElements, HostBufferALIGNED> HostBufferMatrixElements;
 #else
@@ -385,7 +385,7 @@ namespace mg5amcCpu
   // A base class encapsulating a memory buffer for the helicity mask
   typedef BufferBase<bool> BufferHelicityMask;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for the helicity mask
   typedef HostBufferBase<bool, HostBufferALIGNED> HostBufferHelicityMask;
 #else
@@ -403,7 +403,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for wavefunctions
   constexpr size_t sizePerEventWavefunctions = MemoryBuffers::nw6 * MemoryBuffers::nx2;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for wavefunctions
   typedef HostBuffer<fptype, sizePerEventWavefunctions, HostBufferALIGNED> HostBufferWavefunctions;
 #else
@@ -421,7 +421,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for helicity random numbers
   constexpr size_t sizePerEventRndNumHelicity = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for helicity random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumHelicity, HostBufferALIGNED> HostBufferRndNumHelicity;
 #else
@@ -439,7 +439,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for color random numbers
   constexpr size_t sizePerEventRndNumColor = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for color random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumColor, HostBufferALIGNED> HostBufferRndNumColor;
 #else
@@ -457,7 +457,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for helicity selection
   constexpr size_t sizePerEventSelectedHelicity = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for helicity selection
   typedef HostBuffer<int, sizePerEventSelectedHelicity, HostBufferALIGNED> HostBufferSelectedHelicity;
 #else
@@ -475,7 +475,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for color selection
   constexpr size_t sizePerEventSelectedColor = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for color selection
   typedef HostBuffer<int, sizePerEventSelectedColor, HostBufferALIGNED> HostBufferSelectedColor;
 #else
@@ -487,7 +487,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<class Tdst, class Tsrc>
   void copyDeviceFromHost( Tdst& dst, const Tsrc& src ) // keep the same order of arguments as in memcpy
   {
@@ -504,13 +504,13 @@ namespace mg5amcCpu
       throw std::runtime_error( sstr.str() );
     }
     // NB (PR #45): cudaMemcpy involves an intermediate memcpy to pinned memory if host array is a not a pinned host array
-    checkCuda( cudaMemcpy( dst.data(), src.data(), src.bytes(), cudaMemcpyHostToDevice ) );
+    gpuMemcpy( dst.data(), src.data(), src.bytes(), gpuMemcpyHostToDevice );
   }
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<class Tdst, class Tsrc>
   void copyHostFromDevice( Tdst& dst, const Tsrc& src ) // keep the same order of arguments as in memcpy
   {
@@ -527,7 +527,7 @@ namespace mg5amcCpu
       throw std::runtime_error( sstr.str() );
     }
     // NB (PR #45): cudaMemcpy involves an intermediate memcpy to pinned memory if host array is a not a pinned host array
-    checkCuda( cudaMemcpy( dst.data(), src.data(), src.bytes(), cudaMemcpyDeviceToHost ) );
+    gpuMemcpy( dst.data(), src.data(), src.bytes(), gpuMemcpyDeviceToHost );
   }
 #endif
 
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu/CPPProcess.cc b/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu/CPPProcess.cc
index 6242b019fa..4ece50575c 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu/CPPProcess.cc
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_sm.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -49,7 +48,7 @@
 // Process: g d > t t~ d WEIGHTED<=3 @1
 // Process: g s > t t~ s WEIGHTED<=3 @1
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +82,7 @@ namespace mg5amcCpu
   __device__ const fptype cIPD[2] = { (fptype)Parameters_sm::mdl_MT, (fptype)Parameters_sm::mdl_WT };
   __device__ const fptype* cIPC = nullptr; // unused as nicoup=0
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ fptype cIPD[2];
   __device__ __constant__ fptype* cIPC = nullptr; // unused as nicoup=0
 #else
@@ -93,7 +92,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -121,13 +120,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -154,7 +153,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -190,7 +189,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -203,8 +202,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -341,7 +342,7 @@ namespace mg5amcCpu
         { 4, 0, 12, 4 },
         { 0, 4, 4, 12 } }; // 2-D array[4][4]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -398,7 +399,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -457,7 +458,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -520,8 +521,8 @@ namespace mg5amcCpu
       { 1, -1, 1, 1, 1 },
       { 1, -1, 1, -1, -1 },
       { 1, -1, 1, -1, 1 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -562,9 +563,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     const fptype tIPD[2] = { (fptype)m_pars->mdl_MT, (fptype)m_pars->mdl_WT };
     //const cxtype tIPC[0] = { ... }; // nicoup=0
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) ) );
-    //checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ) ); // nicoup=0
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) );
+    //gpuMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
 #else
     memcpy( cIPD, tIPD, 2 * sizeof( fptype ) );
     //memcpy( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
@@ -601,7 +602,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -630,6 +631,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -643,6 +648,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -666,12 +673,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -692,7 +699,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -818,9 +825,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -844,7 +851,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -864,7 +871,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 96 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -878,9 +885,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -908,7 +918,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -1118,7 +1128,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu/CPPProcess.h b/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu/CPPProcess.h
index bf037c6c28..ce22572055 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu/CPPProcess.h
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -110,7 +110,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -123,7 +123,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -153,7 +153,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu/CudaRuntime.h b/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu/GpuAbstraction.h b/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu/GpuRuntime.h b/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu/check_sa.cc b/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu/check_sa.cc
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/CPPProcess.cc b/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/CPPProcess.cc
index 90788b2c75..fee492fbc1 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/CPPProcess.cc
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_sm.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -49,7 +48,7 @@
 // Process: g d~ > t t~ d~ WEIGHTED<=3 @1
 // Process: g s~ > t t~ s~ WEIGHTED<=3 @1
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +82,7 @@ namespace mg5amcCpu
   __device__ const fptype cIPD[2] = { (fptype)Parameters_sm::mdl_MT, (fptype)Parameters_sm::mdl_WT };
   __device__ const fptype* cIPC = nullptr; // unused as nicoup=0
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ fptype cIPD[2];
   __device__ __constant__ fptype* cIPC = nullptr; // unused as nicoup=0
 #else
@@ -93,7 +92,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -121,13 +120,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -154,7 +153,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -190,7 +189,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -203,8 +202,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -341,7 +342,7 @@ namespace mg5amcCpu
         { 4, 0, 12, 4 },
         { 0, 4, 4, 12 } }; // 2-D array[4][4]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -398,7 +399,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -457,7 +458,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -520,8 +521,8 @@ namespace mg5amcCpu
       { 1, 1, 1, 1, -1 },
       { 1, 1, 1, -1, 1 },
       { 1, 1, 1, -1, -1 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -562,9 +563,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     const fptype tIPD[2] = { (fptype)m_pars->mdl_MT, (fptype)m_pars->mdl_WT };
     //const cxtype tIPC[0] = { ... }; // nicoup=0
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) ) );
-    //checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ) ); // nicoup=0
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) );
+    //gpuMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
 #else
     memcpy( cIPD, tIPD, 2 * sizeof( fptype ) );
     //memcpy( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
@@ -601,7 +602,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -630,6 +631,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -643,6 +648,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -666,12 +673,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -692,7 +699,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -818,9 +825,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -844,7 +851,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -864,7 +871,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 96 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -878,9 +885,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -908,7 +918,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -1118,7 +1128,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/CPPProcess.h b/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/CPPProcess.h
index 0f49f5247b..46c4347506 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/CPPProcess.h
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -110,7 +110,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -123,7 +123,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -153,7 +153,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/CudaRuntime.h b/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/GpuAbstraction.h b/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/GpuRuntime.h b/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/check_sa.cc b/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/check_sa.cc
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/RamboSamplingKernels.cc b/epochX/cudacpp/gq_ttq.mad/SubProcesses/RamboSamplingKernels.cc
index da68aa9255..79abbcc4f8 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/RamboSamplingKernels.cc
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/RamboSamplingKernels.cc
@@ -1,11 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "RamboSamplingKernels.h"
 
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "MemoryAccessMomenta.h"
 #include "MemoryAccessRandomNumbers.h"
 #include "MemoryAccessWeights.h"
@@ -14,7 +14,7 @@
 
 #include <sstream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -92,7 +92,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingKernelDevice::RamboSamplingKernelDevice( const fptype energy,               // input: energy
                                                         const BufferRndNumMomenta& rndmom, // input: random numbers in [0,1]
                                                         BufferMomenta& momenta,            // output: momenta
@@ -135,7 +135,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   getMomentaInitialDevice( const fptype energy,
                            fptype* momenta )
@@ -147,17 +147,17 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   void
   RamboSamplingKernelDevice::getMomentaInitial()
   {
-    getMomentaInitialDevice<<<m_gpublocks, m_gputhreads>>>( m_energy, m_momenta.data() );
+    gpuLaunchKernel( getMomentaInitialDevice, m_gpublocks, m_gputhreads, m_energy, m_momenta.data() );
   }
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   getMomentaFinalDevice( const fptype energy,
                          const fptype* rndmom,
@@ -171,11 +171,11 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   void
   RamboSamplingKernelDevice::getMomentaFinal()
   {
-    getMomentaFinalDevice<<<m_gpublocks, m_gputhreads>>>( m_energy, m_rndmom.data(), m_momenta.data(), m_weights.data() );
+    gpuLaunchKernel( getMomentaFinalDevice, m_gpublocks, m_gputhreads, m_energy, m_rndmom.data(), m_momenta.data(), m_weights.data() );
   }
 #endif
 
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/RamboSamplingKernels.h b/epochX/cudacpp/gq_ttq.mad/SubProcesses/RamboSamplingKernels.h
index 184089efd7..7c214cd74b 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/RamboSamplingKernels.h
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/RamboSamplingKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef RAMBOSAMPLINGKERNELS_H
 #define RAMBOSAMPLINGKERNELS_H 1
@@ -10,7 +10,7 @@
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -93,7 +93,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating RAMBO phase space sampling on a GPU device
   class RamboSamplingKernelDevice final : public SamplingKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/RandomNumberKernels.h b/epochX/cudacpp/gq_ttq.mad/SubProcesses/RandomNumberKernels.h
index 188a72c2c9..21d63beeac 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/RandomNumberKernels.h
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/RandomNumberKernels.h
@@ -1,14 +1,14 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef RANDOMNUMBERKERNELS_H
 #define RANDOMNUMBERKERNELS_H 1
 
 #include "mgOnGpuConfig.h"
 
-// NB This must come AFTER mgOnGpuConfig.h which contains our definition of __global__ when __CUDACC__ is not defined
+// NB This must come AFTER mgOnGpuConfig.h which contains our definition of __global__ when MGONGPUCPP_GPUIMPL is not defined
 #ifndef MGONGPU_HAS_NO_CURAND
 //#include "curand.h"
 struct curandGenerator_st; // forward definition from curand.h
@@ -16,7 +16,7 @@ struct curandGenerator_st; // forward definition from curand.h
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/cudacpp.mk b/epochX/cudacpp/gq_ttq.mad/SubProcesses/cudacpp.mk
index 509307506b..df74dfc284 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/cudacpp.mk
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/cudacpp.mk
@@ -1,7 +1,7 @@
 # Copyright (C) 2020-2023 CERN and UCLouvain.
 # Licensed under the GNU Lesser General Public License (version 3 or later).
 # Created by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-# Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+# Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #=== Determine the name of this makefile (https://ftp.gnu.org/old-gnu/Manuals/make-3.80/html_node/make_17.html)
 #=== NB: use ':=' to ensure that the value of CUDACPP_MAKEFILE is not modified further down after including make_opts
@@ -42,10 +42,10 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure common compiler flags for C++ and CUDA
+#=== Configure common compiler flags for C++ and CUDA/HIP
 
 INCFLAGS = -I.
-OPTFLAGS = -O3 # this ends up in CUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
+OPTFLAGS = -O3 # this ends up in GPUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
 
 # Dependency on src directory
 MG5AMC_COMMONLIB = mg5amc_common
@@ -121,24 +121,45 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure the CUDA compiler
+#=== Configure the GPU compiler (CUDA or HIP)
+
+# FIXME! (AV 24.01.2024)
+# In the current implementation (without separate builds for C++ and CUDA/HIP), we first check for cudacc and hipcc in CUDA_HOME and HIP_HOME.
+# If CUDA_HOME or HIP_HOME are not set, try to determine them from the path to cudacc and hipcc.
+# While convoluted, this is currently necessary to allow disabling CUDA/HIP builds by setting CUDA_HOME or HIP_HOME to invalid paths.
+# This will (probably?) be fixed when separate C++ and CUDA/HIP builds are implemented (PR #775).
 
 # If CXX is not a single word (example "clang++ --gcc-toolchain...") then disable CUDA builds (issue #505)
-# This is because it is impossible to pass this to "CUFLAGS += -ccbin <host-compiler>" below
+# This is because it is impossible to pass this to "GPUFLAGS += -ccbin <host-compiler>" below
 ifneq ($(words $(subst ccache ,,$(CXX))),1) # allow at most "CXX=ccache <host-compiler>" from outside
   $(warning CUDA builds are not supported for multi-word CXX "$(CXX)")
   override CUDA_HOME=disabled
 endif
 
-# If CUDA_HOME is not set, try to set it from the location of nvcc
+# If CUDA_HOME is not set, try to set it from the path to nvcc
 ifndef CUDA_HOME
-  CUDA_HOME = $(patsubst %bin/nvcc,%,$(shell which nvcc 2>/dev/null))
+  CUDA_HOME = $(patsubst %/bin/nvcc,%,$(shell which nvcc 2>/dev/null))
   $(warning CUDA_HOME was not set: using "$(CUDA_HOME)")
 endif
 
-# Set NVCC as $(CUDA_HOME)/bin/nvcc if it exists
+# If HIP_HOME is not set, try to set it from the path to hipcc
+ifndef HIP_HOME
+  HIP_HOME = $(patsubst %/bin/hipcc,%,$(shell which hipcc 2>/dev/null))
+  $(warning HIP_HOME was not set: using "$(HIP_HOME)")
+endif
+
+# FIXME! (AV 24.01.2024)
+# In the current implementation (without separate builds for C++ and CUDA/HIP),
+# builds are performed for HIP only if CUDA is not found in the path.
+# If both CUDA and HIP are installed, HIP builds can be triggered by unsetting CUDA_HOME.
+# This will be fixed when separate C++ and CUDA/HIP builds are implemented (PR #775).
+
+#--- Option 1: CUDA exists -> use CUDA
+
+# Set GPUCC as $(CUDA_HOME)/bin/nvcc if it exists
 ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
-  NVCC = $(CUDA_HOME)/bin/nvcc
+
+  GPUCC = $(CUDA_HOME)/bin/nvcc
   USE_NVTX ?=-DUSE_NVTX
   # See https://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/index.html
   # See https://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/
@@ -148,7 +169,7 @@ ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
   # Examples: use 60 for P100 (Piz Daint), 80 for A100 (Juwels Booster, NVidia raplab/Curiosity).
   MADGRAPH_CUDA_ARCHITECTURE ?= 70
   ###CUARCHFLAGS = -gencode arch=compute_$(MADGRAPH_CUDA_ARCHITECTURE),code=compute_$(MADGRAPH_CUDA_ARCHITECTURE) -gencode arch=compute_$(MADGRAPH_CUDA_ARCHITECTURE),code=sm_$(MADGRAPH_CUDA_ARCHITECTURE) # Older implementation (AV): go back to this one for multi-GPU support #533
-  ###CUARCHFLAGS = --gpu-architecture=compute_$(MADGRAPH_CUDA_ARCHITECTURE) --gpu-code=sm_$(MADGRAPH_CUDA_ARCHITECTURE),compute_$(MADGRAPH_CUDA_ARCHITECTURE) # Newer implementation (SH): cannot use this as-is for multi-GPU support #533
+  ###CUARCHFLAGS = --gpu-architecture=compute_$(MADGRAPH_CUDA_ARCHITECTURE) --gpu-code=sm_$(MADGRAPH_CUDA_ARCHITECTURE),compute_$(MADGRAPH_CUDA_ARCHITECTURE)  # Newer implementation (SH): cannot use this as-is for multi-GPU support #533
   comma:=,
   CUARCHFLAGS = $(foreach arch,$(subst $(comma), ,$(MADGRAPH_CUDA_ARCHITECTURE)),-gencode arch=compute_$(arch),code=compute_$(arch) -gencode arch=compute_$(arch),code=sm_$(arch))
   CUINC = -I$(CUDA_HOME)/include/
@@ -158,41 +179,79 @@ ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
     CURANDLIBFLAGS = -L$(CUDA_HOME)/lib64/ -lcurand # NB: -lcuda is not needed here!
   endif
   CUOPTFLAGS = -lineinfo
-  CUFLAGS = $(foreach opt, $(OPTFLAGS), -Xcompiler $(opt)) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
-  ###CUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
-  ###NVCC_VERSION = $(shell $(NVCC) --version | grep 'Cuda compilation tools' | cut -d' ' -f5 | cut -d, -f1)
-  CUFLAGS += -std=c++17 # need CUDA >= 11.2 (see #333): this is enforced in mgOnGpuConfig.h
+  ###GPUFLAGS = $(OPTFLAGS) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
+  GPUFLAGS = $(foreach opt, $(OPTFLAGS), -Xcompiler $(opt)) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
+  ###GPUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
+  ###GPUCC_VERSION = $(shell $(GPUCC) --version | grep 'Cuda compilation tools' | cut -d' ' -f5 | cut -d, -f1)
+  GPUFLAGS += -std=c++17 # need CUDA >= 11.2 (see #333): this is enforced in mgOnGpuConfig.h
   # Without -maxrregcount: baseline throughput: 6.5E8 (16384 32 12) up to 7.3E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 160 # improves throughput: 6.9E8 (16384 32 12) up to 7.7E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 128 # improves throughput: 7.3E8 (16384 32 12) up to 7.6E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 96 # degrades throughput: 4.1E8 (16384 32 12) up to 4.5E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 64 # degrades throughput: 1.7E8 (16384 32 12) flat at 1.7E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 160 # improves throughput: 6.9E8 (16384 32 12) up to 7.7E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 128 # improves throughput: 7.3E8 (16384 32 12) up to 7.6E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 96 # degrades throughput: 4.1E8 (16384 32 12) up to 4.5E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 64 # degrades throughput: 1.7E8 (16384 32 12) flat at 1.7E8 (65536 128 12)
+  CUBUILDRULEFLAGS = -Xcompiler -fPIC -c
+  CCBUILDRULEFLAGS = -Xcompiler -fPIC -c -x cu
+  CUDATESTFLAGS = -lcuda
+
+  # Set the host C++ compiler for GPUCC via "-ccbin <host-compiler>"
+  # (NB issue #505: this must be a single word, "clang++ --gcc-toolchain..." is not supported)
+  GPUFLAGS += -ccbin $(shell which $(subst ccache ,,$(CXX)))
+
+  # Allow newer (unsupported) C++ compilers with older versions of CUDA if ALLOW_UNSUPPORTED_COMPILER_IN_CUDA is set (#504)
+  ifneq ($(origin ALLOW_UNSUPPORTED_COMPILER_IN_CUDA),undefined)
+  GPUFLAGS += -allow-unsupported-compiler
+  endif
+
 else ifneq ($(origin REQUIRE_CUDA),undefined)
+
   # If REQUIRE_CUDA is set but no cuda is found, stop here (e.g. for CI tests on GPU #443)
-  $(error No cuda installation found (set CUDA_HOME or make nvcc visible in PATH))
+  $(error No cuda installation found (set CUDA_HOME or make GPUCC visible in PATH))
+
+#--- Option 2: CUDA does not exist, HIP exists -> use HIP
+
+# Set GPUCC as $(HIP_HOME)/bin/hipcc if it exists
+else ifneq ($(wildcard $(HIP_HOME)/bin/hipcc),)
+
+  GPUCC = $(HIP_HOME)/bin/hipcc
+  #USE_NVTX ?=-DUSE_NVTX # should maybe find something equivalent to this in HIP?
+  HIPARCHFLAGS = -target x86_64-linux-gnu --offload-arch=gfx90a
+  HIPINC = -I$(HIP_HOME)/include/
+  # Note: -DHIP_FAST_MATH is equivalent to -use_fast_math in HIP 
+  # (but only for single precision line 208: https://rocm-developer-tools.github.io/HIP/hcc__detail_2math__functions_8h_source.html)
+  # Note: CUOPTFLAGS should not be used for HIP, it had been added here but was then removed (#808)
+  GPUFLAGS = $(OPTFLAGS) $(INCFLAGS) $(HIPINC) $(HIPARCHFLAGS) -DHIP_FAST_MATH -DHIP_PLATFORM=amd -fPIC
+  ###GPUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
+  GPUFLAGS += -std=c++17
+  ###GPUFLAGS+= --maxrregcount 255 # (AV: is this option valid on HIP and meaningful on AMD GPUs?)
+  CUBUILDRULEFLAGS = -fPIC -c
+  CCBUILDRULEFLAGS = -fPIC -c -x hip
+
+else ifneq ($(origin REQUIRE_HIP),undefined)
+
+  # If REQUIRE_HIP is set but no HIP is found, stop here (e.g. for CI tests on GPU #443)
+  $(error No hip installation found (set HIP_HOME or make GPUCC visible in PATH))
+
+#--- Option 3: CUDA does not exist, HIP does not exist -> switch off both CUDA and HIP
+
 else
-  # No cuda. Switch cuda compilation off and go to common random numbers in C++
+
+  # No cudacc and no hipcc: switch CUDA and HIP compilation off and go to common random numbers in C++
   $(warning CUDA_HOME is not set or is invalid: export CUDA_HOME to compile with cuda)
-  override NVCC=
+  $(warning HIP_HOME is not set or is invalid: export HIP_HOME to compile with hip)
+  override GPUCC=
   override USE_NVTX=
   override CUINC=
   override CURANDLIBFLAGS=
-endif
-export NVCC
-export CUFLAGS
 
-# Set the host C++ compiler for nvcc via "-ccbin <host-compiler>"
-# (NB issue #505: this must be a single word, "clang++ --gcc-toolchain..." is not supported)
-CUFLAGS += -ccbin $(shell which $(subst ccache ,,$(CXX)))
-
-# Allow newer (unsupported) C++ compilers with older versions of CUDA if ALLOW_UNSUPPORTED_COMPILER_IN_CUDA is set (#504)
-ifneq ($(origin ALLOW_UNSUPPORTED_COMPILER_IN_CUDA),undefined)
-CUFLAGS += -allow-unsupported-compiler
 endif
 
+# Export GPUCC (so that it can also be used in cudacpp_src.mk?)
+export GPUCC
+export GPUFLAGS
+
 #-------------------------------------------------------------------------------
 
-#=== Configure ccache for C++ and CUDA builds
+#=== Configure ccache for C++ and CUDA/HIP builds
 
 # Enable ccache if USECCACHE=1
 ifeq ($(USECCACHE)$(shell echo $(CXX) | grep ccache),1)
@@ -201,15 +260,15 @@ endif
 #ifeq ($(USECCACHE)$(shell echo $(AR) | grep ccache),1)
 #  override AR:=ccache $(AR)
 #endif
-ifneq ($(NVCC),)
-  ifeq ($(USECCACHE)$(shell echo $(NVCC) | grep ccache),1)
-    override NVCC:=ccache $(NVCC)
+ifneq ($(GPUCC),)
+  ifeq ($(USECCACHE)$(shell echo $(GPUCC) | grep ccache),1)
+    override GPUCC:=ccache $(GPUCC)
   endif
 endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure PowerPC-specific compiler flags for C++ and CUDA
+#=== Configure PowerPC-specific compiler flags for C++ and CUDA/HIP
 
 # PowerPC-specific CXX compiler flags (being reviewed)
 ifeq ($(UNAME_P),ppc64le)
@@ -225,9 +284,9 @@ else
   ######CXXFLAGS+= -fno-semantic-interposition # no benefit (neither alone, nor combined with -flto)
 endif
 
-# PowerPC-specific CUDA compiler flags (to be reviewed!)
+# PowerPC-specific CUDA/HIP compiler flags (to be reviewed!)
 ifeq ($(UNAME_P),ppc64le)
-  CUFLAGS+= -Xcompiler -mno-float128
+  GPUFLAGS+= -Xcompiler -mno-float128
 endif
 
 #-------------------------------------------------------------------------------
@@ -235,9 +294,11 @@ endif
 #=== Configure defaults and check if user-defined choices exist for OMPFLAGS, AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
 
 # Set the default OMPFLAGS choice
-ifneq ($(shell $(CXX) --version | egrep '^Intel'),)
+ifneq ($(findstring hipcc,$(GPUCC)),)
+override OMPFLAGS = # disable OpenMP MT when using hipcc #802
+else ifneq ($(shell $(CXX) --version | egrep '^Intel'),)
 override OMPFLAGS = -fopenmp
-###override OMPFLAGS = # disable OpenMP MT on Intel (was ok without nvcc but not ok with nvcc before #578)
+###override OMPFLAGS = # disable OpenMP MT on Intel (was ok without GPUCC but not ok with GPUCC before #578)
 else ifneq ($(shell $(CXX) --version | egrep '^(clang)'),)
 override OMPFLAGS = -fopenmp
 ###override OMPFLAGS = # disable OpenMP MT on clang (was not ok without or with nvcc before #578)
@@ -293,7 +354,10 @@ endif
 
 # Set the default RNDGEN (random number generator) choice
 ifeq ($(RNDGEN),)
-  ifeq ($(NVCC),)
+  ifeq ($(GPUCC),)
+    override RNDGEN = hasNoCurand
+  # Edgecase for HIP compilation
+  else ifeq ($(findstring hipcc,$(GPUCC)),hipcc)
     override RNDGEN = hasNoCurand
   else ifeq ($(RNDGEN),)
     override RNDGEN = hasCurand
@@ -310,7 +374,7 @@ export OMPFLAGS
 
 #-------------------------------------------------------------------------------
 
-#=== Set the CUDA/C++ compiler flags appropriate to user-defined choices of AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
+#=== Set the CUDA/HIP/C++ compiler flags appropriate to user-defined choices of AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
 
 # Set the build flags appropriate to OMPFLAGS
 $(info OMPFLAGS=$(OMPFLAGS))
@@ -368,13 +432,13 @@ CXXFLAGS+= $(AVXFLAGS)
 $(info FPTYPE=$(FPTYPE))
 ifeq ($(FPTYPE),d)
   CXXFLAGS += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
-  CUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
+  GPUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
 else ifeq ($(FPTYPE),f)
   CXXFLAGS += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
-  CUFLAGS  += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
+  GPUFLAGS  += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
 else ifeq ($(FPTYPE),m)
   CXXFLAGS += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
-  CUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
+  GPUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
 else
   $(error Unknown FPTYPE='$(FPTYPE)': only 'd', 'f' and 'm' are supported)
 endif
@@ -383,7 +447,7 @@ endif
 $(info HELINL=$(HELINL))
 ifeq ($(HELINL),1)
   CXXFLAGS += -DMGONGPU_INLINE_HELAMPS
-  CUFLAGS  += -DMGONGPU_INLINE_HELAMPS
+  GPUFLAGS  += -DMGONGPU_INLINE_HELAMPS
 else ifneq ($(HELINL),0)
   $(error Unknown HELINL='$(HELINL)': only '0' and '1' are supported)
 endif
@@ -392,7 +456,7 @@ endif
 $(info HRDCOD=$(HRDCOD))
 ifeq ($(HRDCOD),1)
   CXXFLAGS += -DMGONGPU_HARDCODE_PARAM
-  CUFLAGS  += -DMGONGPU_HARDCODE_PARAM
+  GPUFLAGS  += -DMGONGPU_HARDCODE_PARAM
 else ifneq ($(HRDCOD),0)
   $(error Unknown HRDCOD='$(HRDCOD)': only '0' and '1' are supported)
 endif
@@ -444,11 +508,11 @@ ifeq ($(UNAME_S),Darwin)
   override CULIBFLAGSRPATH2 =
 else
   # RPATH to cuda/cpp libs when linking executables
-  override CXXLIBFLAGSRPATH = -Wl,-rpath,$(LIBDIRRPATH)
-  override CULIBFLAGSRPATH = -Xlinker -rpath,$(LIBDIRRPATH)
+  override CXXLIBFLAGSRPATH = -Wl,-rpath=$(LIBDIRRPATH)
+  override CULIBFLAGSRPATH = -Xlinker -rpath=$(LIBDIRRPATH)
   # RPATH to common lib when linking cuda/cpp libs
-  override CXXLIBFLAGSRPATH2 = -Wl,-rpath,'$$ORIGIN'
-  override CULIBFLAGSRPATH2 = -Xlinker -rpath,'$$ORIGIN'
+  override CXXLIBFLAGSRPATH2 = -Wl,-rpath='$$ORIGIN'
+  override CULIBFLAGSRPATH2 = -Xlinker -rpath='$$ORIGIN'
 endif
 
 # Setting LD_LIBRARY_PATH or DYLD_LIBRARY_PATH in the RUNTIME is no longer necessary (neither on Linux nor on Mac)
@@ -461,7 +525,7 @@ override RUNTIME =
 cxx_main=$(BUILDDIR)/check.exe
 fcxx_main=$(BUILDDIR)/fcheck.exe
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 cu_main=$(BUILDDIR)/gcheck.exe
 fcu_main=$(BUILDDIR)/fgcheck.exe
 else
@@ -492,14 +556,15 @@ $(BUILDDIR)/.build.$(TAG):
 	@touch $(BUILDDIR)/.build.$(TAG)
 
 # Generic target and build rules: objects from CUDA compilation
-ifneq ($(NVCC),)
+# NB: CCBUILDRULEFLAGS includes "-x cu" for nvcc and "-x hip" for hipcc (#810)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/%.o : %.cu *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $(CUBUILDRULEFLAGS) $< -o $@
 
 $(BUILDDIR)/%_cu.o : %.cc *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c -x cu $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $(CCBUILDRULEFLAGS) $< -o $@
 endif
 
 # Generic target and build rules: objects from C++ compilation
@@ -509,11 +574,14 @@ $(BUILDDIR)/%.o : %.cc *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	$(CXX) $(CPPFLAGS) $(CXXFLAGS) -fPIC -c $< -o $@
 
 # Apply special build flags only to CrossSectionKernel.cc and gCrossSectionKernel.cu (no fast math, see #117 and #516)
+# Added edgecase for HIP compilation
 ifeq ($(shell $(CXX) --version | grep ^nvc++),)
 $(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS := $(filter-out -ffast-math,$(CXXFLAGS))
 $(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS += -fno-fast-math
-ifneq ($(NVCC),)
-$(BUILDDIR)/gCrossSectionKernels.o: CUFLAGS += -Xcompiler -fno-fast-math
+ifeq ($(findstring nvcc,$(GPUCC)),nvcc)
+  $(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -Xcompiler -fno-fast-math
+else
+  $(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -fno-fast-math
 endif
 endif
 
@@ -530,10 +598,10 @@ ifeq ($(RNDGEN),hasCurand)
 $(BUILDDIR)/CurandRandomNumberKernel.o: CXXFLAGS += $(CUINC)
 endif
 
-# Avoid "warning: builtin __has_trivial_... is deprecated; use __is_trivially_... instead" in nvcc with icx2023 (#592)
+# Avoid "warning: builtin __has_trivial_... is deprecated; use __is_trivially_... instead" in GPUCC with icx2023 (#592)
 ifneq ($(shell $(CXX) --version | egrep '^(Intel)'),)
-ifneq ($(NVCC),)
-CUFLAGS += -Xcompiler -Wno-deprecated-builtins
+ifneq ($(GPUCC),)
+GPUFLAGS += -Wno-deprecated-builtins
 endif
 endif
 
@@ -541,8 +609,8 @@ endif
 # This patch does remove the warning, but I prefer to keep it disabled for the moment...
 ###ifneq ($(shell $(CXX) --version | egrep '^(clang|Apple clang|Intel)'),)
 ###$(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS += -Wno-overriding-t-option
-###ifneq ($(NVCC),)
-###$(BUILDDIR)/gCrossSectionKernels.o: CUFLAGS += -Xcompiler -Wno-overriding-t-option
+###ifneq ($(GPUCC),)
+###$(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -Xcompiler -Wno-overriding-t-option
 ###endif
 ###endif
 
@@ -569,7 +637,7 @@ MG5AMC_CXXLIB = mg5amc_$(processid_short)_cpp
 cxx_objects_lib=$(BUILDDIR)/CPPProcess.o $(BUILDDIR)/MatrixElementKernels.o $(BUILDDIR)/BridgeKernels.o $(BUILDDIR)/CrossSectionKernels.o
 cxx_objects_exe=$(BUILDDIR)/CommonRandomNumberKernel.o $(BUILDDIR)/RamboSamplingKernels.o
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 MG5AMC_CULIB = mg5amc_$(processid_short)_cuda
 cu_objects_lib=$(BUILDDIR)/gCPPProcess.o $(BUILDDIR)/gMatrixElementKernels.o $(BUILDDIR)/gBridgeKernels.o $(BUILDDIR)/gCrossSectionKernels.o
 cu_objects_exe=$(BUILDDIR)/gCommonRandomNumberKernel.o $(BUILDDIR)/gRamboSamplingKernels.o
@@ -581,11 +649,17 @@ $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: cxx_objects_lib += $(BUILDDIR)/fbridge.o
 $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib)
 	$(CXX) -shared -o $@ $(cxx_objects_lib) $(CXXLIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: $(BUILDDIR)/fbridge_cu.o
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: cu_objects_lib += $(BUILDDIR)/fbridge_cu.o
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cu_objects_lib)
-	$(NVCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+# Bypass std::filesystem completely to ease portability on LUMI #803
+#ifneq ($(findstring hipcc,$(GPUCC)),)
+#	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB) -lstdc++fs
+#else
+#	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+#endif
 endif
 
 #-------------------------------------------------------------------------------
@@ -602,16 +676,16 @@ $(cxx_main): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PAT
 $(cxx_main): $(BUILDDIR)/check_sa.o $(LIBDIR)/lib$(MG5AMC_CXXLIB).so $(cxx_objects_exe) $(BUILDDIR)/CurandRandomNumberKernel.o
 	$(CXX) -o $@ $(BUILDDIR)/check_sa.o $(OMPFLAGS) -ldl -pthread $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe) $(BUILDDIR)/CurandRandomNumberKernel.o $(CURANDLIBFLAGS)
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(cu_main): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(cu_main): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(cu_main): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(cu_main): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 else ifneq ($(shell $(CXX) --version | grep ^nvc++),) # support nvc++ #531
 $(cu_main): LIBFLAGS += -L$(patsubst %bin/nvc++,%lib,$(subst ccache ,,$(CXX))) -lnvhpcatm -lnvcpumath -lnvc
 endif
 $(cu_main): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(cu_main): $(BUILDDIR)/gcheck_sa.o $(LIBDIR)/lib$(MG5AMC_CULIB).so $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o
-	$(NVCC) -o $@ $(BUILDDIR)/gcheck_sa.o $(CUARCHFLAGS) $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o $(CURANDLIBFLAGS)
+	$(GPUCC) -o $@ $(BUILDDIR)/gcheck_sa.o $(CUARCHFLAGS) $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o $(CURANDLIBFLAGS)
 endif
 
 #-------------------------------------------------------------------------------
@@ -635,19 +709,27 @@ $(fcxx_main): LIBFLAGS += -L$(shell dirname $(shell $(FC) --print-file-name libg
 endif
 $(fcxx_main): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(fcxx_main): $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler.o $(LIBDIR)/lib$(MG5AMC_CXXLIB).so $(cxx_objects_exe)
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(BUILDDIR)/fcheck_sa.o $(OMPFLAGS) $(BUILDDIR)/fsampler.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe) -lstdc++
+else
 	$(CXX) -o $@ $(BUILDDIR)/fcheck_sa.o $(OMPFLAGS) $(BUILDDIR)/fsampler.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe)
+endif
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(fcu_main): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(fcu_main): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(fcu_main): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(fcu_main): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 endif
 ifeq ($(UNAME_S),Darwin)
 $(fcu_main): LIBFLAGS += -L$(shell dirname $(shell $(FC) --print-file-name libgfortran.dylib)) # add path to libgfortran on Mac #375
 endif
 $(fcu_main): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(fcu_main): $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBDIR)/lib$(MG5AMC_CULIB).so $(cu_objects_exe)
-	$(NVCC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe)
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) -lstdc++ -L$(shell dirname $(shell $(GPUCC) -print-prog-name=clang))/../../lib -lamdhip64
+else
+	$(GPUCC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe)
+endif
 endif
 
 #-------------------------------------------------------------------------------
@@ -659,7 +741,7 @@ $(BUILDDIR)/testxxx.o: testxxx_cc_ref.txt
 $(testmain): $(BUILDDIR)/testxxx.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/testxxx.o # Comment out this line to skip the C++ test of xxx functions
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/testxxx_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/testxxx_cu.o: INCFLAGS += $(GTESTINC)
 $(BUILDDIR)/testxxx_cu.o: testxxx_cc_ref.txt
@@ -672,7 +754,7 @@ $(BUILDDIR)/testmisc.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/testmisc.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/testmisc.o # Comment out this line to skip the C++ miscellaneous tests
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/testmisc_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/testmisc_cu.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/testmisc_cu.o
@@ -684,12 +766,12 @@ $(BUILDDIR)/runTest.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/runTest.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/runTest.o
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/runTest_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/runTest_cu.o: INCFLAGS += $(GTESTINC)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(testmain): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(testmain): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(testmain): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(testmain): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 else ifneq ($(shell $(CXX) --version | grep ^nvc++),) # support nvc++ #531
 $(testmain): LIBFLAGS += -L$(patsubst %bin/nvc++,%lib,$(subst ccache ,,$(CXX))) -lnvhpcatm -lnvcpumath -lnvc
 endif
@@ -713,14 +795,23 @@ $(testmain): LIBFLAGS += -lgomp
 endif
 endif
 
-ifeq ($(NVCC),) # link only runTest.o
+# Bypass std::filesystem completely to ease portability on LUMI #803
+#ifneq ($(findstring hipcc,$(GPUCC)),)
+#$(testmain): LIBFLAGS += -lstdc++fs
+#endif
+
+ifeq ($(GPUCC),) # link only runTest.o
 $(testmain): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(testmain): $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib) $(cxx_objects_exe) $(GTESTLIBS)
 	$(CXX) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) -ldl -pthread $(LIBFLAGS)
 else # link both runTest.o and runTest_cu.o
 $(testmain): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(testmain): $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) $(GTESTLIBS)
-	$(NVCC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) -lcuda
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) $(CUDATESTFLAGS) -lstdc++ -lpthread  -L$(shell dirname $(shell $(GPUCC) -print-prog-name=clang))/../../lib -lamdhip64
+else
+	$(GPUCC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) $(CUDATESTFLAGS)
+endif
 endif
 
 # Use target gtestlibs to build only googletest
@@ -829,9 +920,9 @@ ifeq ($(USECCACHE),1)
 	ccache --version | head -1
 endif
 	@echo ""
-	@echo NVCC=$(NVCC)
-ifneq ($(NVCC),)
-	$(NVCC) --version
+	@echo GPUCC=$(GPUCC)
+ifneq ($(GPUCC),)
+	$(GPUCC) --version
 endif
 	@echo ""
 	@echo CXX=$(CXX)
@@ -850,7 +941,7 @@ endif
 
 # Target: check (run the C++ test executable)
 # [NB THIS IS WHAT IS USED IN THE GITHUB CI!]
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 check: runTest cmpFcheck cmpFGcheck
 else
 check: runTest cmpFcheck
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/fbridge.cc b/epochX/cudacpp/gq_ttq.mad/SubProcesses/fbridge.cc
index 2d2b36d560..22ce3f5115 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/fbridge.cc
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/fbridge.cc
@@ -1,11 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Roiser (Oct 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "Bridge.h"
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 
 extern "C"
 {
@@ -22,7 +22,7 @@ extern "C"
    * Using the same Fortran MadEvent code, linking to the hetrerogeneous library would allow access to both CPU and GPU implementations.
    * The specific heterogeneous configuration (how many GPUs, how many threads on each CPU, etc) could be loaded in CUDA/C++ from a data file.
    */
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -46,8 +46,8 @@ extern "C"
    */
   void fbridgecreate_( CppObjectInFortran** ppbridge, const int* pnevtF, const int* pnparF, const int* pnp4F )
   {
-#ifdef __CUDACC__
-    CudaRuntime::setUp();
+#ifdef MGONGPUCPP_GPUIMPL
+    GpuRuntime::setUp();
 #endif
     // (NB: CPPProcess::initProc no longer needs to be executed here because it is called in the Bridge constructor)
     // FIXME: disable OMP in Bridge when called from Fortran
@@ -65,8 +65,8 @@ extern "C"
     Bridge<FORTRANFPTYPE>* pbridge = dynamic_cast<Bridge<FORTRANFPTYPE>*>( *ppbridge );
     if( pbridge == 0 ) throw std::runtime_error( "fbridgedelete_: invalid Bridge address" );
     delete pbridge;
-#ifdef __CUDACC__
-    CudaRuntime::tearDown();
+#ifdef MGONGPUCPP_GPUIMPL
+    GpuRuntime::tearDown();
 #endif
   }
 
@@ -96,7 +96,7 @@ extern "C"
   {
     Bridge<FORTRANFPTYPE>* pbridge = dynamic_cast<Bridge<FORTRANFPTYPE>*>( *ppbridge );
     if( pbridge == 0 ) throw std::runtime_error( "fbridgesequence_: invalid Bridge address" );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Use the device/GPU implementation in the CUDA library
     // (there is also a host implementation in this library)
     pbridge->gpu_sequence( momenta, gs, rndhel, rndcol, *pchannelId, mes, selhel, selcol );
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/fsampler.cc b/epochX/cudacpp/gq_ttq.mad/SubProcesses/fsampler.cc
index 2fb445372d..3743934f41 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/fsampler.cc
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/fsampler.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Feb 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "mgOnGpuConfig.h"
 
@@ -13,7 +13,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -40,7 +40,7 @@ namespace mg5amcCpu
   private:
     const int m_nevt; // The number of events in each iteration
     int m_iiter;      // The iteration counter (for random number seeding)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     HostBufferRndNumMomenta m_hstRndmom; // Memory buffers for random numbers
     HostBufferMomenta m_hstMomenta;      // Memory buffers for momenta
     HostBufferWeights m_hstWeights;      // Memory buffers for sampling weights
@@ -105,7 +105,7 @@ namespace mg5amcCpu
 
 extern "C"
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/runTest.cc b/epochX/cudacpp/gq_ttq.mad/SubProcesses/runTest.cc
index d4a760a71b..de327f2321 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/runTest.cc
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/runTest.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Hageboeck (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*xxx to run only testxxx.cc tests
 //----------------------------------------------------------------------------
@@ -18,7 +18,7 @@
 #include "RandomNumberKernels.h"
 #include "epoch_process_id.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using namespace mg5amcGpu;
 #else
 using namespace mg5amcCpu;
@@ -35,7 +35,7 @@ struct CUDA_CPU_TestBase : public TestDriverBase
     : TestDriverBase( npar, refFileName ) {}
 };
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 struct CPUTest : public CUDA_CPU_TestBase
 {
   // Struct data members (process, and memory structures for random numbers, momenta, matrix elements and weights on host and device)
@@ -119,7 +119,7 @@ struct CPUTest : public CUDA_CPU_TestBase
 };
 #endif
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 struct CUDATest : public CUDA_CPU_TestBase
 {
   // Reset the device when our test goes out of scope. Note that this should happen after
@@ -128,7 +128,7 @@ struct CUDATest : public CUDA_CPU_TestBase
   {
     ~DeviceReset()
     {
-      checkCuda( cudaDeviceReset() ); // this is needed by cuda-memcheck --leak-check full
+      checkGpu( gpuDeviceReset() ); // this is needed by cuda-memcheck --leak-check full
     }
   } deviceResetter;
 
@@ -256,7 +256,7 @@ INSTANTIATE_TEST_SUITE_P( prefix, \
                           test_suite_name, \
                           testing::Values( new CUDATest( MG_EPOCH_REFERENCE_FILE_NAME ) ) );
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 MG_INSTANTIATE_TEST_SUITE_GPU( XTESTID_GPU( MG_EPOCH_PROCESS_ID ), MadgraphTest );
 #else
 MG_INSTANTIATE_TEST_SUITE_CPU( XTESTID_CPU( MG_EPOCH_PROCESS_ID ), MadgraphTest );
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/testmisc.cc b/epochX/cudacpp/gq_ttq.mad/SubProcesses/testmisc.cc
index 895d6eeb56..ba9e59a8a3 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/testmisc.cc
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/testmisc.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*misc to run only testmisc.cc tests
 //----------------------------------------------------------------------------
@@ -17,7 +17,7 @@
 #include <sstream>
 #include <typeinfo>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #define TESTID( s ) s##_GPU_MISC
 #else
 #define TESTID( s ) s##_CPU_MISC
@@ -26,7 +26,7 @@
 #define XTESTID( s ) TESTID( s )
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -59,7 +59,7 @@ namespace mg5amcCpu
 
 TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testmisc )
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gq_ttq.mad/SubProcesses/testxxx.cc b/epochX/cudacpp/gq_ttq.mad/SubProcesses/testxxx.cc
index 3361fe5aa9..e5167de00c 100644
--- a/epochX/cudacpp/gq_ttq.mad/SubProcesses/testxxx.cc
+++ b/epochX/cudacpp/gq_ttq.mad/SubProcesses/testxxx.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Apr 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*xxx to run only testxxx.cc tests
 //----------------------------------------------------------------------------
@@ -24,7 +24,7 @@
 #include <iomanip>
 #include <iostream>
 #include <vector>
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #define TESTID( s ) s##_GPU_XXX
 #else
 #define TESTID( s ) s##_CPU_XXX
@@ -32,7 +32,7 @@
 
 #define XTESTID( s ) TESTID( s )
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -42,7 +42,7 @@ namespace mg5amcCpu
   int FPEhandlerIevt = -1;
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU): '" << FPEhandlerMessage << "' ievt=" << FPEhandlerIevt << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU neppV=" << neppV << "): '" << FPEhandlerMessage << "' ievt=" << FPEhandlerIevt << std::endl;
@@ -53,7 +53,7 @@ namespace mg5amcCpu
 
 TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -77,7 +77,7 @@ TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
   assert( nevt % neppM == 0 ); // nevt must be a multiple of neppM
   assert( nevt % neppV == 0 ); // nevt must be a multiple of neppV
   // Fill in the input momenta
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   mg5amcGpu::PinnedHostBufferMomenta hstMomenta( nevt ); // AOSOA[npagM][npar=4][np4=4][neppM]
 #else
   mg5amcCpu::HostBufferMomenta hstMomenta( nevt ); // AOSOA[npagM][npar=4][np4=4][neppM]
@@ -322,7 +322,7 @@ TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
   {
     for( int ievt = 0; ievt < nevt; ievt++ )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       using namespace mg5amcGpu;
 #else
       using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gq_ttq.mad/bin/internal/launch_plugin.py b/epochX/cudacpp/gq_ttq.mad/bin/internal/launch_plugin.py
index 3b09713e12..90af6b7053 100644
--- a/epochX/cudacpp/gq_ttq.mad/bin/internal/launch_plugin.py
+++ b/epochX/cudacpp/gq_ttq.mad/bin/internal/launch_plugin.py
@@ -103,7 +103,7 @@ def default_setup(self):
                        fct_mod=(self.reset_makeopts,(),{}),
                        allowed=['auto', 'none', 'sse4', 'avx2','512y','512z'])
         self.add_param('cudacpp_backend', 'CPP', include=False, hidden=False,
-                       allowed=['Fortan', 'CPP', 'CUDA'])
+                       allowed=['Fortran', 'CPP', 'CUDA'])
         self['vector_size'] = 16 # already setup in default class (just change value)
         self['aloha_flag'] = '--fast-math'
         self['matrix_flag'] = '-O3'
diff --git a/epochX/cudacpp/gq_ttq.mad/src/HelAmps_sm.h b/epochX/cudacpp/gq_ttq.mad/src/HelAmps_sm.h
index cd4e6de668..45000c7246 100644
--- a/epochX/cudacpp/gq_ttq.mad/src/HelAmps_sm.h
+++ b/epochX/cudacpp/gq_ttq.mad/src/HelAmps_sm.h
@@ -5,7 +5,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -28,7 +28,7 @@
 //#include <iomanip>
 //#include <iostream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gq_ttq.mad/src/Parameters_sm.cc b/epochX/cudacpp/gq_ttq.mad/src/Parameters_sm.cc
index c06dcbb252..8b92ea0bd6 100644
--- a/epochX/cudacpp/gq_ttq.mad/src/Parameters_sm.cc
+++ b/epochX/cudacpp/gq_ttq.mad/src/Parameters_sm.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -17,7 +17,7 @@
 #include <iomanip>
 #include <iostream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using namespace mg5amcGpu;
 #else
 using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gq_ttq.mad/src/Parameters_sm.h b/epochX/cudacpp/gq_ttq.mad/src/Parameters_sm.h
index a6eb185434..a3615ec77a 100644
--- a/epochX/cudacpp/gq_ttq.mad/src/Parameters_sm.h
+++ b/epochX/cudacpp/gq_ttq.mad/src/Parameters_sm.h
@@ -27,7 +27,7 @@
 #include "read_slha.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -93,7 +93,7 @@ namespace mg5amcCpu
 #include <limits>
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -217,7 +217,7 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -236,7 +236,7 @@ namespace mg5amcCpu
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wunused-variable"  // e.g. <<warning: unused variable ‘mdl_G__exp__2’ [-Wunused-variable]>>
 #pragma GCC diagnostic ignored "-Wunused-parameter" // e.g. <<warning: unused parameter ‘G’ [-Wunused-parameter]>>
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #pragma nv_diagnostic push
 #pragma nv_diag_suppress 177 // e.g. <<warning #177-D: variable "mdl_G__exp__2" was declared but never referenced>>
 #endif
@@ -263,7 +263,7 @@ namespace mg5amcCpu
       // End SM implementation - no special handling of vectors of floats as in EFT (#439)
       return out;
     }
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #pragma GCC diagnostic pop
 #pragma nv_diagnostic pop
 #endif
diff --git a/epochX/cudacpp/gq_ttq.mad/src/cudacpp_src.mk b/epochX/cudacpp/gq_ttq.mad/src/cudacpp_src.mk
index d4cc628aec..b2b9da5288 100644
--- a/epochX/cudacpp/gq_ttq.mad/src/cudacpp_src.mk
+++ b/epochX/cudacpp/gq_ttq.mad/src/cudacpp_src.mk
@@ -19,7 +19,7 @@ SHELL := /bin/bash
 #=== Configure common compiler flags for CUDA and C++
 
 INCFLAGS = -I.
-OPTFLAGS = -O3 # this ends up in CUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
+OPTFLAGS = -O3 # this ends up in GPUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
 
 #-------------------------------------------------------------------------------
 
@@ -45,13 +45,13 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure the CUDA compiler (note: NVCC is already exported including ccache)
+#=== Configure the CUDA compiler (note: GPUCC is already exported including ccache)
 
-###$(info NVCC=$(NVCC))
+###$(info GPUCC=$(GPUCC))
 
 #-------------------------------------------------------------------------------
 
-#=== Configure ccache for C++ builds (note: NVCC is already exported including ccache)
+#=== Configure ccache for C++ builds (note: GPUCC is already exported including ccache)
 
 # Enable ccache if USECCACHE=1
 ifeq ($(USECCACHE)$(shell echo $(CXX) | grep ccache),1)
@@ -92,6 +92,13 @@ endif
 ###$(info OMPFLAGS=$(OMPFLAGS))
 CXXFLAGS += $(OMPFLAGS)
 
+# Add correct flags for nvcc (-x cu) and hipcc (-x hip) for GPU code (see #810)
+ifeq ($(findstring nvcc,$(GPUCC)),nvcc)
+  GPUFLAGS += -Xcompiler -fPIC -c -x cu
+else ifeq ($(findstring hipcc,$(GPUCC)),hipcc)
+  GPUFLAGS += -fPIC -c -x hip
+endif
+
 # Set the build flags appropriate to each AVX choice (example: "make AVX=none")
 # [NB MGONGPU_PVW512 is needed because "-mprefer-vector-width=256" is not exposed in a macro]
 # [See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96476]
@@ -253,20 +260,20 @@ $(BUILDDIR)/%.o : %.cc *.h $(BUILDDIR)/.build.$(TAG)
 # Generic target and build rules: objects from CUDA compilation
 $(BUILDDIR)/%_cu.o : %.cc *.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c -x cu $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $< -o $@
 
 #-------------------------------------------------------------------------------
 
 cxx_objects=$(addprefix $(BUILDDIR)/, Parameters_sm.o read_slha.o)
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 cu_objects=$(addprefix $(BUILDDIR)/, Parameters_sm_cu.o)
 endif
 
 # Target (and build rules): common (src) library
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so : $(cxx_objects) $(cu_objects)
 	@if [ ! -d $(LIBDIR) ]; then echo "mkdir -p $(LIBDIR)"; mkdir -p $(LIBDIR); fi
-	$(NVCC) -shared -o $@ $(cxx_objects) $(cu_objects) $(LDFLAGS)
+	$(GPUCC) -shared -o $@ $(cxx_objects) $(cu_objects) $(LDFLAGS)
 else
 $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so : $(cxx_objects)
 	@if [ ! -d $(LIBDIR) ]; then echo "mkdir -p $(LIBDIR)"; mkdir -p $(LIBDIR); fi
diff --git a/epochX/cudacpp/gq_ttq.mad/src/mgOnGpuConfig.h b/epochX/cudacpp/gq_ttq.mad/src/mgOnGpuConfig.h
index 80032e528b..6bde4466d0 100644
--- a/epochX/cudacpp/gq_ttq.mad/src/mgOnGpuConfig.h
+++ b/epochX/cudacpp/gq_ttq.mad/src/mgOnGpuConfig.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jul 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUCONFIG_H
 #define MGONGPUCONFIG_H 1
@@ -10,12 +10,26 @@
 // There are two different code bases for standalone_cudacpp (without multichannel) and madevent+cudacpp (with multichannel)
 #define MGONGPU_SUPPORTS_MULTICHANNEL 1
 
+// Is this a GPU (CUDA, HIP) or CPU implementation?
+#ifdef __CUDACC__
+#define MGONGPUCPP_GPUIMPL cuda
+#elif defined __HIPCC__
+#define MGONGPUCPP_GPUIMPL hip
+#include "hip/hip_runtime.h" // needed for blockDim, blockIdx, threadIdx: better in mgOnGpuConfig.h than in GpuAbstraction.h
+#else
+#undef MGONGPUCPP_GPUIMPL
+#endif
+
 // ** NB1 Throughputs (e.g. 6.8E8) are events/sec for "./gcheck.exe -p 65536 128 12"
 // ** NB2 Baseline on b7g47n0004 fluctuates (probably depends on load on other VMs)
 
 // Choose if curand is supported for generating random numbers
+// For HIP, by default, do not use curand (common random numbers will be used instead)
 // For both CUDA and C++, by default, do not inline, but allow this macro to be set from outside with e.g. -DMGONGPU_HAS_NO_CURAND
-// (there exist CUDA installations, e.g. using the HPC package, which do not include curand - see PR #784)
+// (there exist CUDA installations, e.g. using the HPC package, which do not include curand - see PR #784 and #785)
+#if defined __HIPCC__
+#define MGONGPU_HAS_NO_CURAND 1
+#else
 //#ifdef __CUDACC__
 //#undef MGONGPU_HAS_NO_CURAND // default
 ////#define MGONGPU_HAS_NO_CURAND 1
@@ -23,6 +37,7 @@
 //#undef MGONGPU_HAS_NO_CURAND // default
 ////#define MGONGPU_HAS_NO_CURAND 1
 //#endif
+#endif
 
 // Choose floating point precision (for everything but color algebra #537)
 // If one of these macros has been set from outside with e.g. -DMGONGPU_FPTYPE_FLOAT, nothing happens (issue #167)
@@ -54,23 +69,31 @@
 //#undef MGONGPU_HARDCODE_PARAM // default
 ////#define MGONGPU_HARDCODE_PARAM 1
 
-// Complex type in c++: std::complex or cxsmpl (CHOOSE ONLY ONE)
-#ifndef __CUDACC__
-//#define MGONGPU_CPPCXTYPE_STDCOMPLEX 1 // ~8 percent slower on float, same on double (5.1E6/double, 9.4E6/float)
-#define MGONGPU_CPPCXTYPE_CXSMPL 1 // new default (5.1E6/double, 10.2E6/float)
-#endif
-
-// Complex type in cuda: thrust or cucomplex or cxsmpl (CHOOSE ONLY ONE)
+// Complex type in CUDA: thrust or cucomplex or cxsmpl (CHOOSE ONLY ONE)
+// (NB THIS IS MGONGPU_*CU*CXTYPE_xxx)
 #ifdef __CUDACC__
 #define MGONGPU_CUCXTYPE_THRUST 1 // default (~1.15E9/double, ~3.2E9/float)
 //#define MGONGPU_CUCXTYPE_CUCOMPLEX 1 // ~10 percent slower (1.03E9/double, ~2.8E9/float)
 //#define MGONGPU_CUCXTYPE_CXSMPL 1 // ~10 percent slower (1.00E9/double, ~2.9E9/float)
+
+// Complex type in HIP: cxsmpl (ONLY ONE OPTION POSSIBLE? #810)
+// (NB THIS IS MGONGPU_*HIP*CXTYPE_xxx)
+#elif defined __HIPCC__
+#define MGONGPU_HIPCXTYPE_CXSMPL 1 // default for HIP
+
+// Complex type in C++: std::complex or cxsmpl (CHOOSE ONLY ONE)
+// (NB THIS IS MGONGPU_*CPP*CXTYPE_xxx)
+#else
+//#define MGONGPU_CPPCXTYPE_STDCOMPLEX 1 // ~8 percent slower on float, same on double (5.1E6/double, 9.4E6/float)
+#define MGONGPU_CPPCXTYPE_CXSMPL 1 // new default (5.1E6/double, 10.2E6/float)
 #endif
 
-// Cuda nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
+// CUDA nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
 #ifdef __CUDACC__
-#undef MGONGPU_NSIGHT_DEBUG // default
+#undef MGONGPU_NSIGHT_DEBUG // default in CUDA
 //#define MGONGPU_NSIGHT_DEBUG 1
+#else
+#undef MGONGPU_NSIGHT_DEBUG // only option in HIP or C++
 #endif
 
 // SANITY CHECKS (floating point precision for everything but color algebra #537)
@@ -86,17 +109,21 @@
 #error You cannot use double precision for color algebra and single precision elsewhere
 #endif
 
-// SANITY CHECKS (c++ complex number implementation)
-#ifndef __CUDACC__
-#if defined MGONGPU_CPPCXTYPE_STDCOMPLEX and defined MGONGPU_CPPCXTYPE_CXSMPL
-#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CPPCXTYPE_STDCOMPLEX or MGONGPU_CPPCXTYPE_CXSMPL
+// SANITY CHECKS (CUDA complex number implementation)
+#ifdef __CUDACC__
+#if defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CUCOMPLEX
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX for CUDA
+#elif defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CXSMPL for CUDA
+#elif defined MGONGPU_CUCXTYPE_CUCOMPLEX and defined MGONGPU_CUCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE OF MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL for CUDA
 #endif
 #endif
 
-// SANITY CHECKS (cuda complex number implementation)
-#ifdef __CUDACC__
-#if defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CUCOMPLEX and defined MGONGPU_CUCXTYPE_CXSMPL
-#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL
+// SANITY CHECKS (C++ complex number implementation)
+#ifndef MGONGPUCPP_GPUIMPL
+#if defined MGONGPU_CPPCXTYPE_STDCOMPLEX and defined MGONGPU_CPPCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CPPCXTYPE_STDCOMPLEX or MGONGPU_CPPCXTYPE_CXSMPL for C++
 #endif
 #endif
 
@@ -134,7 +161,7 @@ namespace mgOnGpu
   // Alignment requirement for using reinterpret_cast with SIMD vectorized code
   // (using reinterpret_cast with non aligned memory may lead to segmentation faults!)
   // Only needed for C++ code but can be enforced also in NVCC builds of C++ code using CUDA>=11.2 and C++17 (#318, #319, #333)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   constexpr int cppAlign = 64; // alignment requirement for SIMD vectorization (64-byte i.e. 512-bit)
 #endif
 
@@ -145,7 +172,7 @@ using mgOnGpu::fptype;
 using mgOnGpu::fptype2;
 
 // C++ SIMD vectorization width (this will be used to set neppV)
-#ifdef __CUDACC__ // CUDA implementation has no SIMD
+#ifdef MGONGPUCPP_GPUIMPL // CUDA and HIP implementations have no SIMD
 #undef MGONGPU_CPPSIMD
 #elif defined __AVX512VL__ && defined MGONGPU_PVW512 // C++ "512z" AVX512 with 512 width (512-bit ie 64-byte): 8 (DOUBLE) or 16 (FLOAT)
 #ifdef MGONGPU_FPTYPE_DOUBLE
@@ -175,9 +202,9 @@ using mgOnGpu::fptype2;
 #undef MGONGPU_CPPSIMD
 #endif
 
-// Cuda nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
+// CUDA nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
 // Arguments (not used so far): text is __FUNCTION__, code is 0 (start) or 1 (end)
-#if defined __CUDACC__ && defined MGONGPU_NSIGHT_DEBUG /* clang-format off */
+#if defined __CUDA__ && defined MGONGPU_NSIGHT_DEBUG /* clang-format off */
 #define mgDebugDeclare() __shared__ float mgDebugCounter[mgOnGpu::ntpbMAX];
 #define mgDebugInitialise() { mgDebugCounter[threadIdx.x] = 0; }
 #define mgDebug( code, text ) { mgDebugCounter[threadIdx.x] += 1; }
@@ -189,8 +216,8 @@ using mgOnGpu::fptype2;
 #define mgDebugFinalise() { /*noop*/ }
 #endif /* clang-format on */
 
-// Define empty CUDA declaration specifiers for C++
-#ifndef __CUDACC__
+// Define empty CUDA/HIP declaration specifiers for C++
+#ifndef MGONGPUCPP_GPUIMPL
 #define __global__
 #define __host__
 #define __device__
diff --git a/epochX/cudacpp/gq_ttq.mad/src/mgOnGpuCxtypes.h b/epochX/cudacpp/gq_ttq.mad/src/mgOnGpuCxtypes.h
index ca9a9f00c0..7ede1dbfae 100644
--- a/epochX/cudacpp/gq_ttq.mad/src/mgOnGpuCxtypes.h
+++ b/epochX/cudacpp/gq_ttq.mad/src/mgOnGpuCxtypes.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022, based on earlier work by D. Smith) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUCXTYPES_H
 #define MGONGPUCXTYPES_H 1
@@ -19,7 +19,7 @@
 #include <complex>
 
 // Complex type in cuda: thrust or cucomplex or cxsmpl
-#ifdef __CUDACC__
+#ifdef __CUDACC__ // this must be __CUDAC__ (not MGONGPUCPP_GPUIMPL)
 #if defined MGONGPU_CUCXTYPE_THRUST
 #pragma clang diagnostic push
 #pragma clang diagnostic ignored "-Wtautological-compare" // for icpx2021/clang13 (https://stackoverflow.com/a/15864661)
@@ -30,8 +30,13 @@
 #elif not defined MGONGPU_CUCXTYPE_CXSMPL
 #error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL
 #endif
+// Complex type in HIP: cxsmpl
+#elif defined __HIPCC__
+#if not defined MGONGPU_HIPCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_HIPCXTYPE_CXSMPL
+#endif
 #else
-// Complex type in c++: std::complex or cxsmpl
+// Complex type in c++ or HIP: std::complex or cxsmpl
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
 #include <cmath>
 #elif not defined MGONGPU_CPPCXTYPE_CXSMPL
@@ -82,7 +87,7 @@ namespace mgOnGpu /* clang-format off */
 using mgOnGpu::cxsmpl;
 
 // Printout to stream for user defined types
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -92,7 +97,7 @@ namespace mg5amcCpu
   inline __host__ std::ostream&
   operator<<( std::ostream& out, const cxsmpl<FP>& c )
   {
-    out << std::complex( c.real(), c.imag() );
+    out << std::complex<FP>( c.real(), c.imag() );
     return out;
   }
 
@@ -215,14 +220,14 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
   // --- Type definitions (complex type: cxtype)
-#ifdef __CUDACC__ // cuda
+#ifdef __CUDACC__ // this must be __CUDAC__ (not MGONGPUCPP_GPUIMPL)
 #if defined MGONGPU_CUCXTYPE_THRUST
   typedef thrust::complex<fptype> cxtype;
 #elif defined MGONGPU_CUCXTYPE_CUCOMPLEX
@@ -255,13 +260,13 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
-#if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+#if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
 
   //------------------------------
   // CUDA or C++ - using cxsmpl
@@ -303,11 +308,11 @@ namespace mg5amcCpu
     return cxmake( c.real(), c.imag() );
   }
 
-#endif // #if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+#endif // #if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
 
   //==========================================================================
 
-#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_THRUST // cuda + thrust
+#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_THRUST // cuda + thrust (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // CUDA - using thrust::complex
@@ -347,7 +352,7 @@ namespace mg5amcCpu
 
   //==========================================================================
 
-#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_CUCOMPLEX // cuda + cucomplex
+#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_CUCOMPLEX // cuda + cucomplex (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // CUDA - using cuComplex
@@ -566,7 +571,7 @@ namespace mg5amcCpu
 
   //==========================================================================
 
-#if not defined __CUDACC__ and defined MGONGPU_CPPCXTYPE_STDCOMPLEX // c++ + stdcomplex
+#if not defined __CUDACC__ and defined MGONGPU_CPPCXTYPE_STDCOMPLEX // c++/hip + stdcomplex (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // C++ - using std::complex
@@ -633,7 +638,7 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gq_ttq.mad/src/mgOnGpuFptypes.h b/epochX/cudacpp/gq_ttq.mad/src/mgOnGpuFptypes.h
index 905c97d700..fa3a02664b 100644
--- a/epochX/cudacpp/gq_ttq.mad/src/mgOnGpuFptypes.h
+++ b/epochX/cudacpp/gq_ttq.mad/src/mgOnGpuFptypes.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUFPTYPES_H
 #define MGONGPUFPTYPES_H 1
@@ -12,7 +12,7 @@
 #include <cmath>
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL // cuda
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -20,7 +20,7 @@ namespace mg5amcCpu
 {
   //==========================================================================
 
-#ifdef __CUDACC__ // cuda
+#ifdef MGONGPUCPP_GPUIMPL // cuda
 
   //------------------------------
   // Floating point types - Cuda
@@ -64,11 +64,11 @@ namespace mg5amcCpu
 #endif
   }
 
-#endif // #ifdef __CUDACC__
+#endif // #ifdef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
   //------------------------------
   // Floating point types - C++
@@ -92,7 +92,7 @@ namespace mg5amcCpu
     return std::sqrt( f );
   }
 
-#endif // #ifndef __CUDACC__
+#endif // #ifndef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
diff --git a/epochX/cudacpp/gq_ttq.mad/src/mgOnGpuVectors.h b/epochX/cudacpp/gq_ttq.mad/src/mgOnGpuVectors.h
index e1299ba81e..cdae04326b 100644
--- a/epochX/cudacpp/gq_ttq.mad/src/mgOnGpuVectors.h
+++ b/epochX/cudacpp/gq_ttq.mad/src/mgOnGpuVectors.h
@@ -32,7 +32,7 @@
 #endif
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -131,7 +131,7 @@ namespace mg5amcCpu
 #endif
 #endif
 
-#else // i.e #ifndef MGONGPU_CPPSIMD (this includes #ifdef __CUDACC__)
+#else // i.e #ifndef MGONGPU_CPPSIMD (this includes #ifdef MGONGPUCPP_GPUIMPL)
 
   const int neppV = 1;
 
@@ -153,13 +153,13 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
   // Printout to stream for user defined types
 
@@ -805,11 +805,11 @@ namespace mg5amcCpu
 
 #endif // #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
 
-#endif // #ifndef __CUDACC__
+#endif // #ifndef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
   //------------------------------
   // Vector types - CUDA
@@ -853,12 +853,12 @@ namespace mg5amcCpu
     return mask;
   }
 
-#endif // #ifdef __CUDACC__
+#endif // #ifdef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
   // Scalar-or-vector types: scalar in CUDA, vector or scalar in C++
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   typedef bool bool_sv;
   typedef fptype fptype_sv;
   typedef fptype2 fptype2_sv;
@@ -879,7 +879,7 @@ namespace mg5amcCpu
 #endif
 
   // Scalar-or-vector zeros: scalar in CUDA, vector or scalar in C++
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   inline __host__ __device__ cxtype cxzero_sv(){ return cxtype( 0, 0 ); }
 #elif defined MGONGPU_CPPSIMD
   inline cxtype_v cxzero_sv() { return cxtype_v(); } // RRRR=0000 IIII=0000
diff --git a/epochX/cudacpp/gq_ttq.mad/src/rambo.h b/epochX/cudacpp/gq_ttq.mad/src/rambo.h
index e02ea52496..cd7e1008ea 100644
--- a/epochX/cudacpp/gq_ttq.mad/src/rambo.h
+++ b/epochX/cudacpp/gq_ttq.mad/src/rambo.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -18,7 +18,7 @@
 #include <iostream>
 
 // Simplified rambo version for 2 to N (with N>=2) processes with massless particles
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +83,7 @@ namespace mg5amcCpu
       static bool first = true;
       if( first )
       {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
         if constexpr( M_ACCESS::isOnDevice() ) // avoid
         {
           const int ievt0 = 0;
@@ -166,7 +166,7 @@ namespace mg5amcCpu
     wt = po2log;
     if( nparf != 2 ) wt = ( 2. * nparf - 4. ) * log( energy ) + z[nparf - 1];
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     // issue warnings if weight is too small or too large
     static int iwarn[5] = { 0, 0, 0, 0, 0 };
     if( wt < -180. )
diff --git a/epochX/cudacpp/gq_ttq.mad/src/read_slha.cc b/epochX/cudacpp/gq_ttq.mad/src/read_slha.cc
index 055b19a779..f8e46f2e66 100644
--- a/epochX/cudacpp/gq_ttq.mad/src/read_slha.cc
+++ b/epochX/cudacpp/gq_ttq.mad/src/read_slha.cc
@@ -11,7 +11,11 @@
 
 #include <algorithm>
 #include <cstdlib>
-#include <filesystem>
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
 #include <fstream>
 #include <iostream>
 
@@ -60,7 +64,14 @@ SLHAReader::read_slha_file( std::string file_name, bool verbose )
     {
       std::cout << "WARNING! Card file '" << file_name << "' does not exist:"
                 << " look for the file in directory $" << envpath << "='" << getenv( envpath ) << "'" << std::endl;
+      /*
+#ifdef __HIPCC__
+      const std::string file_name2 = std::experimental::filesystem::path( getenv( envpath ) ) / std::experimental::filesystem::path( file_name ).filename();
+#else
       const std::string file_name2 = std::filesystem::path( getenv( envpath ) ) / std::filesystem::path( file_name ).filename();
+#endif
+      */
+      const std::string file_name2 = std::string( getenv( envpath ) ) + "/" + file_name; // bypass std::filesystem #803
       param_card.open( file_name2.c_str(), std::ifstream::in );
       if( param_card.good() )
       {
diff --git a/epochX/cudacpp/gq_ttq.sa/CODEGEN_cudacpp_gq_ttq_log.txt b/epochX/cudacpp/gq_ttq.sa/CODEGEN_cudacpp_gq_ttq_log.txt
index f659f6bb8d..fe303ed372 100644
--- a/epochX/cudacpp/gq_ttq.sa/CODEGEN_cudacpp_gq_ttq_log.txt
+++ b/epochX/cudacpp/gq_ttq.sa/CODEGEN_cudacpp_gq_ttq_log.txt
@@ -52,7 +52,7 @@ Note that you can still compile and run aMC@NLO with the built-in PDFs
 
 Using default text editor "vi". Set another one in ./input/mg5_configuration.txt
 Using default eps viewer "evince". Set another one in ./input/mg5_configuration.txt
-No valid web browser found. Please set in ./input/mg5_configuration.txt
+Using default web browser "firefox". Set another one in ./input/mg5_configuration.txt
 import /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_cudacpp_gq_ttq.mg
 The import format was not given, so we guess it as command
 set stdout_level DEBUG
@@ -61,7 +61,7 @@ set zerowidth_tchannel F
 define q = u c d s u~ c~ d~ s~
 INFO: load particles 
 INFO: load vertices 
-DEBUG: model prefixing  takes 0.0054836273193359375 
+DEBUG: model prefixing  takes 0.0057065486907958984 
 INFO: Restrict model sm with file models/sm/restrict_default.dat . 
 DEBUG: Simplifying conditional expressions 
 DEBUG: remove interactions: u s w+ at order: QED=1 
@@ -170,7 +170,7 @@ INFO: Crossed process found for g u~ > t t~ u~, reuse diagrams.
 INFO: Crossed process found for g c~ > t t~ c~, reuse diagrams. 
 INFO: Crossed process found for g d~ > t t~ d~, reuse diagrams. 
 INFO: Crossed process found for g s~ > t t~ s~, reuse diagrams. 
-8 processes with 40 diagrams generated in 0.080 s
+8 processes with 40 diagrams generated in 0.077 s
 Total: 8 processes with 40 diagrams
 output standalone_cudacpp ../TMPOUT/CODEGEN_cudacpp_gq_ttq
 Load PLUGIN.CUDACPP_OUTPUT
@@ -211,7 +211,7 @@ Generated helas calls for 2 subprocesses (10 diagrams) in 0.031 s
 ALOHA: aloha starts to compute helicity amplitudes
 ALOHA: aloha creates FFV1 routines
 ALOHA: aloha creates VVV1 routines
-ALOHA: aloha creates 2 routines in  0.146 s
+ALOHA: aloha creates 2 routines in  0.141 s
 <class 'aloha.create_aloha.AbstractRoutine'> FFV1
 <class 'aloha.create_aloha.AbstractRoutine'> FFV1
 <class 'aloha.create_aloha.AbstractRoutine'> FFV1
@@ -227,7 +227,7 @@ INFO: Created files Parameters_sm.h and Parameters_sm.cc in directory
 INFO: /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_cudacpp_gq_ttq/src/. and /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_cudacpp_gq_ttq/src/. 
 quit
 
-real	0m0.709s
-user	0m0.586s
-sys	0m0.064s
-Code generation completed in 0 seconds
+real	0m0.652s
+user	0m0.583s
+sys	0m0.057s
+Code generation completed in 1 seconds
diff --git a/epochX/cudacpp/gq_ttq.sa/COPYRIGHT b/epochX/cudacpp/gq_ttq.sa/COPYRIGHT
index a134b5fef9..9036d9260a 100644
--- a/epochX/cudacpp/gq_ttq.sa/COPYRIGHT
+++ b/epochX/cudacpp/gq_ttq.sa/COPYRIGHT
@@ -15,6 +15,7 @@ The full development team currently includes the following authors :
   Stephan Hageboeck (CERN)
   Olivier Mattelaer (Universite Catholique de Louvain, original author)
   Stefan Roiser (CERN, original author)
+  Jorgen Teig (CERN)
   Andrea Valassi (CERN, original author)
   Zenny Wettersten (CERN)
 See https://github.com/madgraph5/madgraph4gpu for more details. For the full
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/Bridge.h b/epochX/cudacpp/gq_ttq.sa/SubProcesses/Bridge.h
index bf8b5e024d..f9ed70dfde 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/Bridge.h
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/Bridge.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Roiser (Nov 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef BRIDGE_H
 #define BRIDGE_H 1
@@ -14,16 +14,23 @@
 #include "MemoryAccessMomenta.h"  // for MemoryAccessMomenta::neppM
 #include "MemoryBuffers.h"        // for HostBufferMomenta, DeviceBufferMomenta etc
 
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
+
+#include <sys/stat.h> // bypass std::filesystem #803
+
 #include <algorithm>
 #include <cassert>
 #include <cmath>
 #include <cstring>
-#include <filesystem>
 #include <iostream>
 #include <memory>
 #include <type_traits>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +90,7 @@ namespace mg5amcCpu
     Bridge& operator=( const Bridge& ) = delete;
     Bridge& operator=( Bridge&& ) = delete;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     /**
      * Set the gpublocks and gputhreads for the gpusequence - throws if evnt != gpublocks*gputhreads
      * (this is needed for BridgeKernel tests rather than for actual production use in Fortran)
@@ -150,7 +157,7 @@ namespace mg5amcCpu
     unsigned int m_nevt; // number of events
     int m_nGoodHel;      // the number of good helicities (-1 initially when they have not yet been calculated)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     int m_gputhreads; // number of gpu threads (default set from number of events, can be modified)
     int m_gpublocks;  // number of gpu blocks (default set from number of events, can be modified)
     DeviceBuffer<FORTRANFPTYPE, sizePerEventMomenta> m_devMomentaF;
@@ -187,12 +194,12 @@ namespace mg5amcCpu
   // Forward declare transposition methods
   //
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
   template<typename Tin, typename Tout>
   __global__ void dev_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt );
 
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
 
   template<typename Tin, typename Tout>
   void hst_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt );
@@ -209,7 +216,7 @@ namespace mg5amcCpu
   Bridge<FORTRANFPTYPE>::Bridge( unsigned int nevtF, unsigned int nparF, unsigned int np4F )
     : m_nevt( nevtF )
     , m_nGoodHel( -1 )
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     , m_gputhreads( 256 )                  // default number of gpu threads
     , m_gpublocks( m_nevt / m_gputhreads ) // this ensures m_nevt <= m_gpublocks*m_gputhreads
     , m_devMomentaF( m_nevt )
@@ -233,7 +240,7 @@ namespace mg5amcCpu
   {
     if( nparF != CPPProcess::npar ) throw std::runtime_error( "Bridge constructor: npar mismatch" );
     if( np4F != CPPProcess::np4 ) throw std::runtime_error( "Bridge constructor: np4 mismatch" );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( ( m_nevt < s_gputhreadsmin ) || ( m_nevt % s_gputhreadsmin != 0 ) )
       throw std::runtime_error( "Bridge constructor: nevt should be a multiple of " + std::to_string( s_gputhreadsmin ) );
     while( m_nevt != m_gpublocks * m_gputhreads )
@@ -249,20 +256,28 @@ namespace mg5amcCpu
 #else
     std::cout << "WARNING! Instantiate host Bridge (nevt=" << m_nevt << ")" << std::endl;
     m_pmek.reset( new MatrixElementKernelHost( m_hstMomentaC, m_hstGs, m_hstRndHel, m_hstRndCol, m_hstMEs, m_hstSelHel, m_hstSelCol, m_nevt ) );
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
     // Create a process object, read param card and set parameters
     // FIXME: the process instance can happily go out of scope because it is only needed to read parameters?
     // FIXME: the CPPProcess should really be a singleton? what if fbridgecreate is called from several Fortran threads?
     CPPProcess process( /*verbose=*/false );
     std::string paramCard = "../../Cards/param_card.dat";
-    if( !std::filesystem::exists( paramCard ) )
-    {
-      paramCard = "../" + paramCard;
-    }
+    /*
+#ifdef __HIPCC__
+    if( !std::experimental::filesystem::exists( paramCard ) ) paramCard = "../" + paramCard;
+#else
+    if( !std::filesystem::exists( paramCard ) ) paramCard = "../" + paramCard;
+#endif
+    */
+    //struct stat dummybuffer; // bypass std::filesystem #803
+    //if( !( stat( paramCard.c_str(), &dummyBuffer ) == 0 ) ) paramCard = "../" + paramCard; //
+    auto fileExists = []( std::string& fileName )
+    { struct stat buffer; return stat( fileName.c_str(), &buffer ) == 0; };
+    if( !fileExists( paramCard ) ) paramCard = "../" + paramCard; // bypass std::filesystem #803
     process.initProc( paramCard );
   }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::set_gpugrid( const int gpublocks, const int gputhreads )
   {
@@ -276,7 +291,7 @@ namespace mg5amcCpu
   }
 #endif
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::gpu_sequence( const FORTRANFPTYPE* momenta,
                                             const FORTRANFPTYPE* gs,
@@ -291,14 +306,14 @@ namespace mg5amcCpu
     constexpr int neppM = MemoryAccessMomenta::neppM;
     if constexpr( neppM == 1 && std::is_same_v<FORTRANFPTYPE, fptype> )
     {
-      checkCuda( cudaMemcpy( m_devMomentaC.data(), momenta, m_devMomentaC.bytes(), cudaMemcpyHostToDevice ) );
+      gpuMemcpy( m_devMomentaC.data(), momenta, m_devMomentaC.bytes(), gpuMemcpyHostToDevice );
     }
     else
     {
-      checkCuda( cudaMemcpy( m_devMomentaF.data(), momenta, m_devMomentaF.bytes(), cudaMemcpyHostToDevice ) );
+      gpuMemcpy( m_devMomentaF.data(), momenta, m_devMomentaF.bytes(), gpuMemcpyHostToDevice );
       const int thrPerEvt = CPPProcess::npar * CPPProcess::np4; // AV: transpose alg does 1 element per thread (NOT 1 event per thread)
       //const int thrPerEvt = 1; // AV: try new alg with 1 event per thread... this seems slower
-      dev_transposeMomentaF2C<<<m_gpublocks * thrPerEvt, m_gputhreads>>>( m_devMomentaF.data(), m_devMomentaC.data(), m_nevt );
+      gpuLaunchKernel( dev_transposeMomentaF2C, m_gpublocks * thrPerEvt, m_gputhreads, m_devMomentaF.data(), m_devMomentaC.data(), m_nevt );
     }
     if constexpr( std::is_same_v<FORTRANFPTYPE, fptype> )
     {
@@ -341,7 +356,7 @@ namespace mg5amcCpu
   }
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::cpu_sequence( const FORTRANFPTYPE* momenta,
                                             const FORTRANFPTYPE* gs,
@@ -396,7 +411,7 @@ namespace mg5amcCpu
   // - C++ array: momenta[npagM][npar][np4][neppM] with nevt=npagM*neppM (AOSOA)
   //
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename Tin, typename Tout>
   __global__ void dev_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt )
   {
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/BridgeKernels.cc b/epochX/cudacpp/gq_ttq.sa/SubProcesses/BridgeKernels.cc
index d58066c9c1..eaf4037a24 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/BridgeKernels.cc
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/BridgeKernels.cc
@@ -1,17 +1,18 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "BridgeKernels.h"
 
+#include "GpuAbstraction.h"
 #include "MemoryAccessMomenta.h"
 
 #include <sstream>
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -45,7 +46,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu
 {
 
@@ -96,7 +97,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/BridgeKernels.h b/epochX/cudacpp/gq_ttq.sa/SubProcesses/BridgeKernels.h
index 15eb4bff4d..3efef8ce97 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/BridgeKernels.h
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/BridgeKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef BRIDGEKERNELS_H
 #define BRIDGEKERNELS_H 1
@@ -12,7 +12,7 @@
 #include "MatrixElementKernels.h"
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -49,7 +49,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A Bridge wrapper class encapsulating matrix element calculations on a CPU host
   class BridgeKernelHost final : public BridgeKernelBase
   {
@@ -89,7 +89,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A Bridge wrapper class encapsulating matrix element calculations on a GPU device
   class BridgeKernelDevice : public BridgeKernelBase
   {
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/CommonRandomNumberKernel.cc b/epochX/cudacpp/gq_ttq.sa/SubProcesses/CommonRandomNumberKernel.cc
index 985b39f576..010bc4cbd0 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/CommonRandomNumberKernel.cc
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/CommonRandomNumberKernel.cc
@@ -1,15 +1,16 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "CommonRandomNumbers.h"
+#include "GpuAbstraction.h"
 #include "MemoryBuffers.h"
 #include "RandomNumberKernels.h"
 
 #include <cassert>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/CrossSectionKernels.cc b/epochX/cudacpp/gq_ttq.sa/SubProcesses/CrossSectionKernels.cc
index 0b355a3c8d..c15b39844d 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/CrossSectionKernels.cc
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/CrossSectionKernels.cc
@@ -1,10 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "CrossSectionKernels.h"
 
+#include "GpuAbstraction.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessWeights.h"
 #include "MemoryBuffers.h"
@@ -77,7 +78,7 @@ debug_me_is_abnormal( const fptype& me, size_t ievtALL )
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -185,7 +186,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/CrossSectionKernels.h b/epochX/cudacpp/gq_ttq.sa/SubProcesses/CrossSectionKernels.h
index 7933ca4bbf..4d9659e04e 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/CrossSectionKernels.h
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/CrossSectionKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef CROSSSECTIONKERNELS_H
 #define CROSSSECTIONKERNELS_H 1
@@ -13,7 +13,7 @@
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -96,7 +96,7 @@ namespace mg5amcCpu
   //--------------------------------------------------------------------------
 
   /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating the calculation of event statistics on a GPU device
   class CrossSectionKernelDevice : public CrossSectionKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/CudaRuntime.h b/epochX/cudacpp/gq_ttq.sa/SubProcesses/CudaRuntime.h
deleted file mode 100644
index 64ce52f4b3..0000000000
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/CudaRuntime.h
+++ /dev/null
@@ -1,85 +0,0 @@
-// Copyright (C) 2020-2023 CERN and UCLouvain.
-// Licensed under the GNU Lesser General Public License (version 3 or later).
-// Created by: S. Roiser (Jul 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
-
-#ifndef MG5AMC_CUDARUNTIME_H
-#define MG5AMC_CUDARUNTIME_H 1
-
-// MG5AMC on GPU uses the CUDA runtime API, not the lower level CUDA driver API
-// See https://docs.nvidia.com/cuda/cuda-runtime-api/driver-vs-runtime-api.html#driver-vs-runtime-api
-
-#include <cassert>
-#include <iostream>
-
-//--------------------------------------------------------------------------
-
-// See https://stackoverflow.com/a/14038590
-#ifdef __CUDACC__ /* clang-format off */
-#define checkCuda( code ) { assertCuda( code, __FILE__, __LINE__ ); }
-inline void assertCuda( cudaError_t code, const char* file, int line, bool abort = true )
-{
-  if( code != cudaSuccess )
-  {
-    printf( "ERROR! assertCuda: '%s' (%d) in %s:%d\n", cudaGetErrorString( code ), code, file, line );
-    if( abort ) assert( code == cudaSuccess );
-  }
-}
-#endif /* clang-format on */
-
-//--------------------------------------------------------------------------
-
-#ifdef __CUDACC__
-namespace mg5amcGpu
-{
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  // *** FIXME! This will all need to be designed differently when going to multi-GPU nodes! ***
-  struct CudaRuntime final
-  {
-    CudaRuntime( const bool debug = true )
-      : m_debug( debug ) { setUp( m_debug ); }
-    ~CudaRuntime() { tearDown( m_debug ); }
-    CudaRuntime( const CudaRuntime& ) = delete;
-    CudaRuntime( CudaRuntime&& ) = delete;
-    CudaRuntime& operator=( const CudaRuntime& ) = delete;
-    CudaRuntime& operator=( CudaRuntime&& ) = delete;
-    bool m_debug;
-
-    // Set up CUDA application
-    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
-    // Calling cudaSetDevice on startup is useful to properly book-keep the time spent in CUDA initialization
-    static void setUp( const bool debug = true )
-    {
-      // ** NB: it is useful to call cudaSetDevice, or cudaFree, to properly book-keep the time spent in CUDA initialization
-      // ** NB: otherwise, the first CUDA operation (eg a cudaMemcpyToSymbol in CPPProcess ctor) appears to take much longer!
-      /*
-      // [We initially added cudaFree(0) to "ease profile analysis" only because it shows up as a big recognizable block!]
-      // No explicit initialization is needed: https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#initialization
-      // It is not clear what cudaFree(0) does at all: https://stackoverflow.com/questions/69967813/
-      if ( debug ) std::cout << "__CudaRuntime: calling cudaFree(0)" << std::endl;
-      checkCuda( cudaFree( 0 ) ); // SLOW!
-      */
-      // Replace cudaFree(0) by cudaSetDevice(0), even if it is not really needed either
-      // (but see https://developer.nvidia.com/blog/cuda-pro-tip-always-set-current-device-avoid-multithreading-bugs)
-      if( debug ) std::cout << "__CudaRuntime: calling cudaSetDevice(0)" << std::endl;
-      checkCuda( cudaSetDevice( 0 ) ); // SLOW!
-    }
-
-    // Tear down CUDA application (call cudaDeviceReset)
-    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
-    // Calling cudaDeviceReset on shutdown is only needed for checking memory leaks in cuda-memcheck
-    // See https://docs.nvidia.com/cuda/cuda-memcheck/index.html#leak-checking
-    static void tearDown( const bool debug = true )
-    {
-      if( debug ) std::cout << "__CudaRuntime: calling cudaDeviceReset()" << std::endl;
-      checkCuda( cudaDeviceReset() );
-    }
-  };
-
-}
-#endif
-
-//--------------------------------------------------------------------------
-
-#endif // MG5AMC_CUDARUNTIME_H
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/CurandRandomNumberKernel.cc b/epochX/cudacpp/gq_ttq.sa/SubProcesses/CurandRandomNumberKernel.cc
index eb56333b03..08a16f6f2c 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/CurandRandomNumberKernel.cc
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/CurandRandomNumberKernel.cc
@@ -1,9 +1,9 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "MemoryBuffers.h"
 #include "RandomNumberKernels.h"
 
@@ -22,7 +22,7 @@ inline void assertCurand( curandStatus_t code, const char *file, int line, bool
 }
 #endif /* clang-format on */
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -36,7 +36,7 @@ namespace mg5amcCpu
   {
     if( m_isOnDevice )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if( !m_rnarray.isOnDevice() )
         throw std::runtime_error( "CurandRandomNumberKernel on device with a host random number array" );
 #else
@@ -114,7 +114,7 @@ namespace mg5amcCpu
     /*
     printf( "\nCurandRandomNumberKernel::generateRnarray size = %d\n", (int)m_rnarray.size() );
     fptype* data = m_rnarray.data();
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( m_rnarray.isOnDevice() )
     {
       data = new fptype[m_rnarray.size()]();
@@ -123,7 +123,7 @@ namespace mg5amcCpu
 #endif
     for( int i = 0; i < ( (int)m_rnarray.size() / 4 ); i++ )
       printf( "[%4d] %f %f %f %f\n", i * 4, data[i * 4], data[i * 4 + 2], data[i * 4 + 2], data[i * 4 + 3] );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( m_rnarray.isOnDevice() ) delete[] data;
 #endif
     */
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/EventStatistics.h b/epochX/cudacpp/gq_ttq.sa/SubProcesses/EventStatistics.h
index 48b51e0a49..b425a5bade 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/EventStatistics.h
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/EventStatistics.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef EventStatistics_H
 #define EventStatistics_H 1
@@ -16,7 +16,7 @@
 #include <limits>
 #include <string>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/GpuAbstraction.h b/epochX/cudacpp/gq_ttq.sa/SubProcesses/GpuAbstraction.h
new file mode 100644
index 0000000000..9c467b1e04
--- /dev/null
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/GpuAbstraction.h
@@ -0,0 +1,69 @@
+// Copyright (C) 2020-2023 CERN and UCLouvain.
+// Licensed under the GNU Lesser General Public License (version 3 or later).
+// Created by: J. Teig (Jul 2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+
+#ifndef MG5AMC_GPUABSTRACTION_H
+#define MG5AMC_GPUABSTRACTION_H 1
+
+#include <cassert>
+
+//--------------------------------------------------------------------------
+
+#ifdef __CUDACC__
+
+#define gpuError_t cudaError_t
+#define gpuPeekAtLastError cudaPeekAtLastError
+#define gpuGetErrorString cudaGetErrorString
+#define gpuSuccess cudaSuccess
+
+#define gpuMallocHost( ptr, size ) checkGpu( cudaMallocHost( ptr, size ) )
+#define gpuMalloc( ptr, size ) checkGpu( cudaMalloc( ptr, size ) )
+
+#define gpuMemcpy( dstData, srcData, srcBytes, func ) checkGpu( cudaMemcpy( dstData, srcData, srcBytes, func ) )
+#define gpuMemcpyHostToDevice cudaMemcpyHostToDevice
+#define gpuMemcpyDeviceToHost cudaMemcpyDeviceToHost
+#define gpuMemcpyToSymbol( type1, type2, size ) checkGpu( cudaMemcpyToSymbol( type1, type2, size ) )
+
+#define gpuFree( ptr ) checkGpu( cudaFree( ptr ) )
+#define gpuFreeHost( ptr ) checkGpu( cudaFreeHost( ptr ) )
+
+#define gpuSetDevice cudaSetDevice
+#define gpuDeviceSynchronize cudaDeviceSynchronize
+#define gpuDeviceReset cudaDeviceReset
+
+#define gpuLaunchKernel( kernel, blocks, threads, ... ) kernel<<<blocks, threads>>>( __VA_ARGS__ )
+#define gpuLaunchKernelSharedMem( kernel, blocks, threads, sharedMem, ... ) kernel<<<blocks, threads, sharedMem>>>( __VA_ARGS__ )
+
+//--------------------------------------------------------------------------
+
+#elif defined __HIPCC__
+
+#define gpuError_t hipError_t
+#define gpuPeekAtLastError hipPeekAtLastError
+#define gpuGetErrorString hipGetErrorString
+#define gpuSuccess hipSuccess
+
+#define gpuMallocHost( ptr, size ) checkGpu( hipHostMalloc( ptr, size ) ) // HostMalloc better
+#define gpuMalloc( ptr, size ) checkGpu( hipMalloc( ptr, size ) )
+
+#define gpuMemcpy( dstData, srcData, srcBytes, func ) checkGpu( hipMemcpy( dstData, srcData, srcBytes, func ) )
+#define gpuMemcpyHostToDevice hipMemcpyHostToDevice
+#define gpuMemcpyDeviceToHost hipMemcpyDeviceToHost
+#define gpuMemcpyToSymbol( type1, type2, size ) checkGpu( hipMemcpyToSymbol( type1, type2, size ) )
+
+#define gpuFree( ptr ) checkGpu( hipFree( ptr ) )
+#define gpuFreeHost( ptr ) checkGpu( hipHostFree( ptr ) )
+
+#define gpuSetDevice hipSetDevice
+#define gpuDeviceSynchronize hipDeviceSynchronize
+#define gpuDeviceReset hipDeviceReset
+
+#define gpuLaunchKernel( kernel, blocks, threads, ... ) kernel<<<blocks, threads>>>( __VA_ARGS__ )
+#define gpuLaunchKernelSharedMem( kernel, blocks, threads, sharedMem, ... ) kernel<<<blocks, threads, sharedMem>>>( __VA_ARGS__ )
+
+//--------------------------------------------------------------------------
+
+#endif
+
+#endif // MG5AMC_GPUABSTRACTION_H
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/GpuRuntime.h b/epochX/cudacpp/gq_ttq.sa/SubProcesses/GpuRuntime.h
new file mode 100644
index 0000000000..93579ef08b
--- /dev/null
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/GpuRuntime.h
@@ -0,0 +1,85 @@
+// Copyright (C) 2020-2023 CERN and UCLouvain.
+// Licensed under the GNU Lesser General Public License (version 3 or later).
+// Created by: J. Teig (Jun 2023, based on earlier work by S. Roiser) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+
+#ifndef MG5AMC_GPURUNTIME_H
+#define MG5AMC_GPURUNTIME_H 1
+
+// MG5AMC on GPU uses the CUDA runtime API, not the lower level CUDA driver API
+// See https://docs.nvidia.com/cuda/cuda-runtime-api/driver-vs-runtime-api.html#driver-vs-runtime-api
+
+#include "GpuAbstraction.h"
+
+#include <iostream>
+
+//--------------------------------------------------------------------------
+
+// See https://stackoverflow.com/a/14038590
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
+#define checkGpu( code ) { assertGpu( code, __FILE__, __LINE__ ); }
+inline void assertGpu( gpuError_t code, const char* file, int line, bool abort = true )
+{
+  if( code != gpuSuccess )
+  {
+    printf( "ERROR! assertGpu: '%s' (%d) in %s:%d\n", gpuGetErrorString( code ), code, file, line );
+    if( abort ) assert( code == gpuSuccess );
+  }
+}
+#endif /* clang-format on */
+
+//--------------------------------------------------------------------------
+
+#ifdef MGONGPUCPP_GPUIMPL
+namespace mg5amcGpu
+{
+  // Instantiate a GpuRuntime at the beginnining of the application's main to
+  // invoke gpuSetDevice(0) in the constructor and book a gpuDeviceReset() call in the destructor
+  // *** FIXME! This will all need to be designed differently when going to multi-GPU nodes! ***
+  struct GpuRuntime final
+  {
+    GpuRuntime( const bool debug = true )
+      : m_debug( debug ) { setUp( m_debug ); }
+    ~GpuRuntime() { tearDown( m_debug ); }
+    GpuRuntime( const GpuRuntime& ) = delete;
+    GpuRuntime( GpuRuntime&& ) = delete;
+    GpuRuntime& operator=( const GpuRuntime& ) = delete;
+    GpuRuntime& operator=( GpuRuntime&& ) = delete;
+    bool m_debug;
+
+    // Set up CUDA application
+    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
+    // Calling cudaSetDevice on startup is useful to properly book-keep the time spent in CUDA initialization
+    static void setUp( const bool debug = true )
+    {
+      // ** NB: it is useful to call cudaSetDevice, or cudaFree, to properly book-keep the time spent in CUDA initialization
+      // ** NB: otherwise, the first CUDA operation (eg a cudaMemcpyToSymbol in CPPProcess ctor) appears to take much longer!
+      /*
+      // [We initially added cudaFree(0) to "ease profile analysis" only because it shows up as a big recognizable block!]
+      // No explicit initialization is needed: https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#initialization
+      // It is not clear what cudaFree(0) does at all: https://stackoverflow.com/questions/69967813/
+      if ( debug ) std::cout << "__CudaRuntime: calling cudaFree(0)" << std::endl;
+      checkCuda( cudaFree( 0 ) ); // SLOW!
+      */
+      // Replace cudaFree(0) by cudaSetDevice(0), even if it is not really needed either
+      // (but see https://developer.nvidia.com/blog/cuda-pro-tip-always-set-current-device-avoid-multithreading-bugs)
+      if( debug ) std::cout << "__GpuRuntime: calling GpuSetDevice(0)" << std::endl;
+      checkGpu( gpuSetDevice( 0 ) ); // SLOW!
+    }
+
+    // Tear down CUDA application (call cudaDeviceReset)
+    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
+    // Calling cudaDeviceReset on shutdown is only needed for checking memory leaks in cuda-memcheck
+    // See https://docs.nvidia.com/cuda/cuda-memcheck/index.html#leak-checking
+    static void tearDown( const bool debug = true )
+    {
+      if( debug ) std::cout << "__GpuRuntime: calling GpuDeviceReset()" << std::endl;
+      checkGpu( gpuDeviceReset() );
+    }
+  };
+}
+#endif
+
+//--------------------------------------------------------------------------
+
+#endif // MG5AMC_GPURUNTIME_H
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/MadgraphTest.h b/epochX/cudacpp/gq_ttq.sa/SubProcesses/MadgraphTest.h
index ef40624c88..6054185300 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/MadgraphTest.h
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/MadgraphTest.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Hageboeck (Dec 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MADGRAPHTEST_H_
 #define MADGRAPHTEST_H_ 1
@@ -14,7 +14,11 @@
 
 #include <array>
 #include <cmath>
-#include <filesystem>
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
 #include <fstream>
 #include <iomanip>
 #include <memory>
@@ -22,7 +26,7 @@
 #include <string>
 #include <vector>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using mg5amcGpu::CPPProcess;
 #else
 using mg5amcCpu::CPPProcess;
@@ -201,7 +205,7 @@ class MadgraphTest : public testing::TestWithParam<TestDriverBase*>
 
 // Since we link both the CPU-only and GPU tests into the same executable, we prevent
 // a multiply defined symbol by only compiling this in the non-CUDA phase:
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
 /// Compare momenta and matrix elements.
 /// This uses an implementation of TestDriverBase to run a madgraph workflow,
@@ -219,7 +223,14 @@ TEST_P( MadgraphTest, CompareMomentaAndME )
   const char* dumpEventsC = getenv( "CUDACPP_RUNTEST_DUMPEVENTS" );
   const bool dumpEvents = ( dumpEventsC != 0 ) && ( std::string( dumpEventsC ) != "" );
   const std::string refFileName = testDriver->getRefFileName();
+  /*
+#ifdef __HIPCC__
+  const std::string dumpFileName = std::experimental::filesystem::path( refFileName ).filename();
+#else
   const std::string dumpFileName = std::filesystem::path( refFileName ).filename();
+#endif
+  */
+  const std::string dumpFileName = refFileName; // bypass std::filesystem #803
   std::ofstream dumpFile;
   if( dumpEvents )
   {
@@ -307,6 +318,6 @@ TEST_P( MadgraphTest, CompareMomentaAndME )
   }
 }
 
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
 
 #endif /* MADGRAPHTEST_H_ */
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/MatrixElementKernels.cc b/epochX/cudacpp/gq_ttq.sa/SubProcesses/MatrixElementKernels.cc
index 74b5239ebf..81699dfea9 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/MatrixElementKernels.cc
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/MatrixElementKernels.cc
@@ -1,12 +1,12 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "MatrixElementKernels.h"
 
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h" // Includes the abstraction for Nvidia/AMD compilation
 #include "MemoryAccessMomenta.h"
 #include "MemoryBuffers.h"
 
@@ -14,7 +14,7 @@
 
 //============================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu
 {
 
@@ -150,7 +150,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
@@ -209,13 +209,13 @@ namespace mg5amcGpu
     PinnedHostBufferHelicityMask hstIsGoodHel( ncomb );
     DeviceBufferHelicityMask devIsGoodHel( ncomb );
     // ... 0d1. Compute good helicity mask on the device
-    computeDependentCouplings<<<m_gpublocks, m_gputhreads>>>( m_gs.data(), m_couplings.data() );
+    gpuLaunchKernel( computeDependentCouplings, m_gpublocks, m_gputhreads, m_gs.data(), m_couplings.data() );
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-    sigmaKin_getGoodHel<<<m_gpublocks, m_gputhreads>>>( m_momenta.data(), m_couplings.data(), m_matrixElements.data(), m_numerators.data(), m_denominators.data(), devIsGoodHel.data() );
+    gpuLaunchKernel( sigmaKin_getGoodHel, m_gpublocks, m_gputhreads, m_momenta.data(), m_couplings.data(), m_matrixElements.data(), m_numerators.data(), m_denominators.data(), devIsGoodHel.data() );
 #else
-    sigmaKin_getGoodHel<<<m_gpublocks, m_gputhreads>>>( m_momenta.data(), m_couplings.data(), m_matrixElements.data(), devIsGoodHel.data() );
+    gpuLaunchKernel( sigmaKin_getGoodHel, m_gpublocks, m_gputhreads, m_momenta.data(), m_couplings.data(), m_matrixElements.data(), devIsGoodHel.data() );
 #endif
-    checkCuda( cudaPeekAtLastError() );
+    checkGpu( gpuPeekAtLastError() );
     // ... 0d2. Copy back good helicity mask to the host
     copyHostFromDevice( hstIsGoodHel, devIsGoodHel );
     // ... 0d3. Copy back good helicity list to constant memory on the device
@@ -226,19 +226,19 @@ namespace mg5amcGpu
 
   void MatrixElementKernelDevice::computeMatrixElements( const unsigned int channelId )
   {
-    computeDependentCouplings<<<m_gpublocks, m_gputhreads>>>( m_gs.data(), m_couplings.data() );
+    gpuLaunchKernel( computeDependentCouplings, m_gpublocks, m_gputhreads, m_gs.data(), m_couplings.data() );
 #ifndef MGONGPU_NSIGHT_DEBUG
     constexpr unsigned int sharedMemSize = 0;
 #else
     constexpr unsigned int sharedMemSize = ntpbMAX * sizeof( float );
 #endif
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-    sigmaKin<<<m_gpublocks, m_gputhreads, sharedMemSize>>>( m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), channelId, m_numerators.data(), m_denominators.data(), m_selhel.data(), m_selcol.data() );
+    gpuLaunchKernelSharedMem( sigmaKin, m_gpublocks, m_gputhreads, sharedMemSize, m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), channelId, m_numerators.data(), m_denominators.data(), m_selhel.data(), m_selcol.data() );
 #else
-    sigmaKin<<<m_gpublocks, m_gputhreads, sharedMemSize>>>( m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), m_selhel.data(), m_selcol.data() );
+    gpuLaunchKernelSharedMem( sigmaKin, m_gpublocks, m_gputhreads, sharedMemSize, m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), m_selhel.data(), m_selcol.data() );
 #endif
-    checkCuda( cudaPeekAtLastError() );
-    checkCuda( cudaDeviceSynchronize() );
+    checkGpu( gpuPeekAtLastError() );
+    checkGpu( gpuDeviceSynchronize() );
   }
 
   //--------------------------------------------------------------------------
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/MatrixElementKernels.h b/epochX/cudacpp/gq_ttq.sa/SubProcesses/MatrixElementKernels.h
index 23e84757a2..72bd8f195b 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/MatrixElementKernels.h
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/MatrixElementKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MATRIXELEMENTKERNELS_H
 #define MATRIXELEMENTKERNELS_H 1
@@ -10,7 +10,7 @@
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -81,7 +81,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating matrix element calculations on a CPU host
   class MatrixElementKernelHost final : public MatrixElementKernelBase, public NumberOfEvents
   {
@@ -130,7 +130,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating matrix element calculations on a GPU device
   class MatrixElementKernelDevice : public MatrixElementKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessAmplitudes.h b/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessAmplitudes.h
index 573b3bbbc9..ffb76e93de 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessAmplitudes.h
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessAmplitudes.h
@@ -15,7 +15,7 @@
 #define MGONGPU_TRIVIAL_AMPLITUDES 1
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessCouplings.h b/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessCouplings.h
index 35a3af42e0..3afdf3e554 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessCouplings.h
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessCouplings.h
@@ -15,7 +15,7 @@
 #include "MemoryBuffers.h"       // for HostBufferCouplings::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessCouplingsFixed.h b/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessCouplingsFixed.h
index dc0d93afff..ffcdf4dbef 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessCouplingsFixed.h
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessCouplingsFixed.h
@@ -14,7 +14,7 @@
 //#include "MemoryAccessHelpers.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessDenominators.h b/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessDenominators.h
index 3bce635718..66f2d32a6b 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessDenominators.h
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessDenominators.h
@@ -10,7 +10,7 @@
 #include "MemoryAccessGs.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessGs.h b/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessGs.h
index 31311aa375..4c726b30f3 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessGs.h
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessGs.h
@@ -13,7 +13,7 @@
 #include "MemoryBuffers.h" // for HostBufferMatrixElements::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessHelpers.h b/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessHelpers.h
index c82a6c7635..db73e4e064 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessHelpers.h
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessHelpers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessHelpers_H
 #define MemoryAccessHelpers_H 1
@@ -105,7 +105,7 @@ class KernelAccessHelper : public MemoryAccessHelper<T>
     }
     else
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
       //printf( "kernelAccessRecord: ievt=%d threadId=%d\n", ievt, threadIdx.x );
       return T::ieventAccessRecord( buffer, ievt ); // NB fptype and fptype_sv coincide for CUDA
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessMatrixElements.h b/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessMatrixElements.h
index f32e6fea5b..3741011971 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessMatrixElements.h
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessMatrixElements.h
@@ -13,7 +13,7 @@
 #include "MemoryBuffers.h" // for HostBufferMatrixElements::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessMomenta.h b/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessMomenta.h
index 29266de32c..3be229d392 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessMomenta.h
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessMomenta.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessMomenta_H
 #define MemoryAccessMomenta_H 1
@@ -13,7 +13,7 @@
 #include "MemoryAccessVectors.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -30,7 +30,7 @@ namespace mg5amcCpu
 
     // Number of Events Per Page in the momenta AOSOA memory buffer layout
     // (these are all best kept as a compile-time constants: see issue #23)
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
     // -----------------------------------------------------------------------------------------------
     // --- GPUs: neppM is best set to a power of 2 times the number of fptype's in a 32-byte cacheline
     // --- This is relevant to ensure coalesced access to momenta in global memory
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessNumerators.h b/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessNumerators.h
index b152183b28..18991f4fa6 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessNumerators.h
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessNumerators.h
@@ -10,7 +10,7 @@
 #include "MemoryAccessGs.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessRandomNumbers.h b/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessRandomNumbers.h
index e2988d39f3..40cb089135 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessRandomNumbers.h
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessRandomNumbers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessRandomNumbers_H
 #define MemoryAccessRandomNumbers_H 1
@@ -11,7 +11,7 @@
 #include "CPPProcess.h"
 #include "MemoryAccessHelpers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using mg5amcGpu::CPPProcess;
 #else
 using mg5amcCpu::CPPProcess;
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessVectors.h b/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessVectors.h
index e9b197368e..08faccff0f 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessVectors.h
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessVectors.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessVectors_H
 #define MemoryAccessVectors_H 1
@@ -10,7 +10,7 @@
 
 #include "mgOnGpuVectors.h"
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu // this is only needed for CPU SIMD vectorization
 {
 
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessWavefunctions.h b/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessWavefunctions.h
index 5428aaf933..33bef4559e 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessWavefunctions.h
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryAccessWavefunctions.h
@@ -15,7 +15,7 @@
 #define MGONGPU_TRIVIAL_WAVEFUNCTIONS 1
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryBuffers.h b/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryBuffers.h
index 3093e6ed18..7756a71621 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryBuffers.h
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/MemoryBuffers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021, based on earlier work by S. Hageboeck) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryBuffers_H
 #define MemoryBuffers_H 1
@@ -11,12 +11,12 @@
 #include "mgOnGpuCxtypes.h"
 
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "Parameters_sm.h"
 
 #include <sstream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -87,7 +87,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   constexpr bool HostBufferALIGNED = false;   // ismisaligned=false
   constexpr bool HostBufferMISALIGNED = true; // ismisaligned=true
 
@@ -119,7 +119,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA pinned host buffer
   template<typename T>
   class PinnedHostBufferBase : public BufferBase<T>
@@ -128,18 +128,18 @@ namespace mg5amcCpu
     PinnedHostBufferBase( const size_t size )
       : BufferBase<T>( size, false )
     {
-      checkCuda( cudaMallocHost( &( this->m_data ), this->bytes() ) );
+      gpuMallocHost( &( this->m_data ), this->bytes() );
     }
     virtual ~PinnedHostBufferBase()
     {
-      checkCuda( cudaFreeHost( this->m_data ) );
+      gpuFreeHost( this->m_data );
     }
   };
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA device buffer
   template<typename T>
   class DeviceBufferBase : public BufferBase<T>
@@ -148,18 +148,18 @@ namespace mg5amcCpu
     DeviceBufferBase( const size_t size )
       : BufferBase<T>( size, true )
     {
-      checkCuda( cudaMalloc( &( this->m_data ), this->bytes() ) );
+      gpuMalloc( &( this->m_data ), this->bytes() );
     }
     virtual ~DeviceBufferBase()
     {
-      checkCuda( cudaFree( this->m_data ) );
+      gpuFree( this->m_data );
     }
   };
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for a given number of events
   template<typename T, size_t sizePerEvent, bool ismisaligned>
   class HostBuffer : public HostBufferBase<T, ismisaligned>, virtual private NumberOfEvents
@@ -175,7 +175,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA pinned host buffer for a given number of events
   template<typename T, size_t sizePerEvent>
   class PinnedHostBuffer : public PinnedHostBufferBase<T>, virtual private NumberOfEvents
@@ -191,7 +191,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA device buffer for a given number of events
   template<typename T, size_t sizePerEvent>
   class DeviceBuffer : public DeviceBufferBase<T>, virtual private NumberOfEvents
@@ -213,7 +213,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for momenta random numbers
   constexpr size_t sizePerEventRndNumMomenta = MemoryBuffers::np4 * MemoryBuffers::nparf;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for momenta random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumMomenta, HostBufferALIGNED> HostBufferRndNumMomenta;
 #else
@@ -232,7 +232,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer with ONE fptype per event
   constexpr size_t sizePerEventOneFp = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer with ONE fptype per event
   typedef HostBuffer<fptype, sizePerEventOneFp, HostBufferALIGNED> HostBufferOneFp;
 #else
@@ -257,7 +257,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for Gs
   constexpr size_t sizePerEventGs = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventGs, HostBufferALIGNED> HostBufferGs;
 #else
@@ -276,7 +276,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for numerators
   constexpr size_t sizePerEventNumerators = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventNumerators, HostBufferALIGNED> HostBufferNumerators;
 #else
@@ -296,7 +296,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for denominators
   constexpr size_t sizePerEventDenominators = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventDenominators, HostBufferALIGNED> HostBufferDenominators;
 #else
@@ -315,7 +315,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for random numbers
   constexpr size_t sizePerEventCouplings = MemoryBuffers::ndcoup * MemoryBuffers::nx2;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventCouplings, HostBufferALIGNED> HostBufferCouplings;
 #else
@@ -333,7 +333,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for momenta
   constexpr size_t sizePerEventMomenta = MemoryBuffers::np4 * MemoryBuffers::npar;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for momenta
   typedef HostBuffer<fptype, sizePerEventMomenta, HostBufferALIGNED> HostBufferMomenta;
   //typedef HostBuffer<fptype, sizePerEventMomenta, HostBufferMISALIGNED> HostBufferMomenta; // TEST MISALIGNMENT!
@@ -352,7 +352,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for sampling weights
   constexpr size_t sizePerEventWeights = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for sampling weights
   typedef HostBuffer<fptype, sizePerEventWeights, HostBufferALIGNED> HostBufferWeights;
 #else
@@ -370,7 +370,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for matrix elements
   constexpr size_t sizePerEventMatrixElements = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for matrix elements
   typedef HostBuffer<fptype, sizePerEventMatrixElements, HostBufferALIGNED> HostBufferMatrixElements;
 #else
@@ -385,7 +385,7 @@ namespace mg5amcCpu
   // A base class encapsulating a memory buffer for the helicity mask
   typedef BufferBase<bool> BufferHelicityMask;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for the helicity mask
   typedef HostBufferBase<bool, HostBufferALIGNED> HostBufferHelicityMask;
 #else
@@ -403,7 +403,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for wavefunctions
   constexpr size_t sizePerEventWavefunctions = MemoryBuffers::nw6 * MemoryBuffers::nx2;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for wavefunctions
   typedef HostBuffer<fptype, sizePerEventWavefunctions, HostBufferALIGNED> HostBufferWavefunctions;
 #else
@@ -421,7 +421,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for helicity random numbers
   constexpr size_t sizePerEventRndNumHelicity = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for helicity random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumHelicity, HostBufferALIGNED> HostBufferRndNumHelicity;
 #else
@@ -439,7 +439,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for color random numbers
   constexpr size_t sizePerEventRndNumColor = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for color random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumColor, HostBufferALIGNED> HostBufferRndNumColor;
 #else
@@ -457,7 +457,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for helicity selection
   constexpr size_t sizePerEventSelectedHelicity = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for helicity selection
   typedef HostBuffer<int, sizePerEventSelectedHelicity, HostBufferALIGNED> HostBufferSelectedHelicity;
 #else
@@ -475,7 +475,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for color selection
   constexpr size_t sizePerEventSelectedColor = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for color selection
   typedef HostBuffer<int, sizePerEventSelectedColor, HostBufferALIGNED> HostBufferSelectedColor;
 #else
@@ -487,7 +487,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<class Tdst, class Tsrc>
   void copyDeviceFromHost( Tdst& dst, const Tsrc& src ) // keep the same order of arguments as in memcpy
   {
@@ -504,13 +504,13 @@ namespace mg5amcCpu
       throw std::runtime_error( sstr.str() );
     }
     // NB (PR #45): cudaMemcpy involves an intermediate memcpy to pinned memory if host array is a not a pinned host array
-    checkCuda( cudaMemcpy( dst.data(), src.data(), src.bytes(), cudaMemcpyHostToDevice ) );
+    gpuMemcpy( dst.data(), src.data(), src.bytes(), gpuMemcpyHostToDevice );
   }
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<class Tdst, class Tsrc>
   void copyHostFromDevice( Tdst& dst, const Tsrc& src ) // keep the same order of arguments as in memcpy
   {
@@ -527,7 +527,7 @@ namespace mg5amcCpu
       throw std::runtime_error( sstr.str() );
     }
     // NB (PR #45): cudaMemcpy involves an intermediate memcpy to pinned memory if host array is a not a pinned host array
-    checkCuda( cudaMemcpy( dst.data(), src.data(), src.bytes(), cudaMemcpyDeviceToHost ) );
+    gpuMemcpy( dst.data(), src.data(), src.bytes(), gpuMemcpyDeviceToHost );
   }
 #endif
 
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gu_ttxu/CPPProcess.cc b/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gu_ttxu/CPPProcess.cc
index 90e90b3aa9..497f35fa66 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gu_ttxu/CPPProcess.cc
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gu_ttxu/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_sm.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -49,7 +48,7 @@
 // Process: g d > t t~ d WEIGHTED<=3 @1
 // Process: g s > t t~ s WEIGHTED<=3 @1
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +82,7 @@ namespace mg5amcCpu
   __device__ const fptype cIPD[2] = { (fptype)Parameters_sm::mdl_MT, (fptype)Parameters_sm::mdl_WT };
   __device__ const fptype* cIPC = nullptr; // unused as nicoup=0
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ fptype cIPD[2];
   __device__ __constant__ fptype* cIPC = nullptr; // unused as nicoup=0
 #else
@@ -93,7 +92,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -121,13 +120,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -154,7 +153,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -190,7 +189,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -203,8 +202,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -336,7 +337,7 @@ namespace mg5amcCpu
         { 4, 0, 12, 4 },
         { 0, 4, 4, 12 } }; // 2-D array[4][4]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -393,7 +394,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -452,7 +453,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -515,8 +516,8 @@ namespace mg5amcCpu
       { 1, -1, 1, 1, 1 },
       { 1, -1, 1, -1, -1 },
       { 1, -1, 1, -1, 1 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -557,9 +558,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     const fptype tIPD[2] = { (fptype)m_pars->mdl_MT, (fptype)m_pars->mdl_WT };
     //const cxtype tIPC[0] = { ... }; // nicoup=0
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) ) );
-    //checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ) ); // nicoup=0
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) );
+    //gpuMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
 #else
     memcpy( cIPD, tIPD, 2 * sizeof( fptype ) );
     //memcpy( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
@@ -596,7 +597,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -625,6 +626,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -638,6 +643,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -661,12 +668,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -687,7 +694,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -813,9 +820,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -839,7 +846,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -859,7 +866,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 96 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -873,9 +880,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -903,7 +913,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -1113,7 +1123,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gu_ttxu/CPPProcess.h b/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gu_ttxu/CPPProcess.h
index bf037c6c28..ce22572055 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gu_ttxu/CPPProcess.h
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gu_ttxu/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -110,7 +110,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -123,7 +123,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -153,7 +153,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gu_ttxu/CudaRuntime.h b/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gu_ttxu/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gu_ttxu/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gu_ttxu/GpuAbstraction.h b/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gu_ttxu/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gu_ttxu/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gu_ttxu/GpuRuntime.h b/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gu_ttxu/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gu_ttxu/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gu_ttxu/check_sa.cc b/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gu_ttxu/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gu_ttxu/check_sa.cc
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gu_ttxu/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gux_ttxux/CPPProcess.cc b/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gux_ttxux/CPPProcess.cc
index 76c9403933..61269d2eac 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gux_ttxux/CPPProcess.cc
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gux_ttxux/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_sm.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -49,7 +48,7 @@
 // Process: g d~ > t t~ d~ WEIGHTED<=3 @1
 // Process: g s~ > t t~ s~ WEIGHTED<=3 @1
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +82,7 @@ namespace mg5amcCpu
   __device__ const fptype cIPD[2] = { (fptype)Parameters_sm::mdl_MT, (fptype)Parameters_sm::mdl_WT };
   __device__ const fptype* cIPC = nullptr; // unused as nicoup=0
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ fptype cIPD[2];
   __device__ __constant__ fptype* cIPC = nullptr; // unused as nicoup=0
 #else
@@ -93,7 +92,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -121,13 +120,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -154,7 +153,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -190,7 +189,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -203,8 +202,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -336,7 +337,7 @@ namespace mg5amcCpu
         { 4, 0, 12, 4 },
         { 0, 4, 4, 12 } }; // 2-D array[4][4]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -393,7 +394,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -452,7 +453,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -515,8 +516,8 @@ namespace mg5amcCpu
       { 1, 1, 1, 1, -1 },
       { 1, 1, 1, -1, 1 },
       { 1, 1, 1, -1, -1 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -557,9 +558,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     const fptype tIPD[2] = { (fptype)m_pars->mdl_MT, (fptype)m_pars->mdl_WT };
     //const cxtype tIPC[0] = { ... }; // nicoup=0
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) ) );
-    //checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ) ); // nicoup=0
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) );
+    //gpuMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
 #else
     memcpy( cIPD, tIPD, 2 * sizeof( fptype ) );
     //memcpy( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
@@ -596,7 +597,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -625,6 +626,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -638,6 +643,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -661,12 +668,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -687,7 +694,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -813,9 +820,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -839,7 +846,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -859,7 +866,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 96 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -873,9 +880,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -903,7 +913,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -1113,7 +1123,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gux_ttxux/CPPProcess.h b/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gux_ttxux/CPPProcess.h
index 0f49f5247b..46c4347506 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gux_ttxux/CPPProcess.h
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gux_ttxux/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -110,7 +110,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -123,7 +123,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -153,7 +153,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gux_ttxux/CudaRuntime.h b/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gux_ttxux/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gux_ttxux/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gux_ttxux/GpuAbstraction.h b/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gux_ttxux/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gux_ttxux/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gux_ttxux/GpuRuntime.h b/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gux_ttxux/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gux_ttxux/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gux_ttxux/check_sa.cc b/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gux_ttxux/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gux_ttxux/check_sa.cc
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/P1_Sigma_sm_gux_ttxux/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/RamboSamplingKernels.cc b/epochX/cudacpp/gq_ttq.sa/SubProcesses/RamboSamplingKernels.cc
index da68aa9255..79abbcc4f8 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/RamboSamplingKernels.cc
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/RamboSamplingKernels.cc
@@ -1,11 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "RamboSamplingKernels.h"
 
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "MemoryAccessMomenta.h"
 #include "MemoryAccessRandomNumbers.h"
 #include "MemoryAccessWeights.h"
@@ -14,7 +14,7 @@
 
 #include <sstream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -92,7 +92,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingKernelDevice::RamboSamplingKernelDevice( const fptype energy,               // input: energy
                                                         const BufferRndNumMomenta& rndmom, // input: random numbers in [0,1]
                                                         BufferMomenta& momenta,            // output: momenta
@@ -135,7 +135,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   getMomentaInitialDevice( const fptype energy,
                            fptype* momenta )
@@ -147,17 +147,17 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   void
   RamboSamplingKernelDevice::getMomentaInitial()
   {
-    getMomentaInitialDevice<<<m_gpublocks, m_gputhreads>>>( m_energy, m_momenta.data() );
+    gpuLaunchKernel( getMomentaInitialDevice, m_gpublocks, m_gputhreads, m_energy, m_momenta.data() );
   }
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   getMomentaFinalDevice( const fptype energy,
                          const fptype* rndmom,
@@ -171,11 +171,11 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   void
   RamboSamplingKernelDevice::getMomentaFinal()
   {
-    getMomentaFinalDevice<<<m_gpublocks, m_gputhreads>>>( m_energy, m_rndmom.data(), m_momenta.data(), m_weights.data() );
+    gpuLaunchKernel( getMomentaFinalDevice, m_gpublocks, m_gputhreads, m_energy, m_rndmom.data(), m_momenta.data(), m_weights.data() );
   }
 #endif
 
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/RamboSamplingKernels.h b/epochX/cudacpp/gq_ttq.sa/SubProcesses/RamboSamplingKernels.h
index 184089efd7..7c214cd74b 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/RamboSamplingKernels.h
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/RamboSamplingKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef RAMBOSAMPLINGKERNELS_H
 #define RAMBOSAMPLINGKERNELS_H 1
@@ -10,7 +10,7 @@
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -93,7 +93,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating RAMBO phase space sampling on a GPU device
   class RamboSamplingKernelDevice final : public SamplingKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/RandomNumberKernels.h b/epochX/cudacpp/gq_ttq.sa/SubProcesses/RandomNumberKernels.h
index 188a72c2c9..21d63beeac 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/RandomNumberKernels.h
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/RandomNumberKernels.h
@@ -1,14 +1,14 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef RANDOMNUMBERKERNELS_H
 #define RANDOMNUMBERKERNELS_H 1
 
 #include "mgOnGpuConfig.h"
 
-// NB This must come AFTER mgOnGpuConfig.h which contains our definition of __global__ when __CUDACC__ is not defined
+// NB This must come AFTER mgOnGpuConfig.h which contains our definition of __global__ when MGONGPUCPP_GPUIMPL is not defined
 #ifndef MGONGPU_HAS_NO_CURAND
 //#include "curand.h"
 struct curandGenerator_st; // forward definition from curand.h
@@ -16,7 +16,7 @@ struct curandGenerator_st; // forward definition from curand.h
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/cudacpp.mk b/epochX/cudacpp/gq_ttq.sa/SubProcesses/cudacpp.mk
index 509307506b..df74dfc284 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/cudacpp.mk
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/cudacpp.mk
@@ -1,7 +1,7 @@
 # Copyright (C) 2020-2023 CERN and UCLouvain.
 # Licensed under the GNU Lesser General Public License (version 3 or later).
 # Created by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-# Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+# Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #=== Determine the name of this makefile (https://ftp.gnu.org/old-gnu/Manuals/make-3.80/html_node/make_17.html)
 #=== NB: use ':=' to ensure that the value of CUDACPP_MAKEFILE is not modified further down after including make_opts
@@ -42,10 +42,10 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure common compiler flags for C++ and CUDA
+#=== Configure common compiler flags for C++ and CUDA/HIP
 
 INCFLAGS = -I.
-OPTFLAGS = -O3 # this ends up in CUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
+OPTFLAGS = -O3 # this ends up in GPUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
 
 # Dependency on src directory
 MG5AMC_COMMONLIB = mg5amc_common
@@ -121,24 +121,45 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure the CUDA compiler
+#=== Configure the GPU compiler (CUDA or HIP)
+
+# FIXME! (AV 24.01.2024)
+# In the current implementation (without separate builds for C++ and CUDA/HIP), we first check for cudacc and hipcc in CUDA_HOME and HIP_HOME.
+# If CUDA_HOME or HIP_HOME are not set, try to determine them from the path to cudacc and hipcc.
+# While convoluted, this is currently necessary to allow disabling CUDA/HIP builds by setting CUDA_HOME or HIP_HOME to invalid paths.
+# This will (probably?) be fixed when separate C++ and CUDA/HIP builds are implemented (PR #775).
 
 # If CXX is not a single word (example "clang++ --gcc-toolchain...") then disable CUDA builds (issue #505)
-# This is because it is impossible to pass this to "CUFLAGS += -ccbin <host-compiler>" below
+# This is because it is impossible to pass this to "GPUFLAGS += -ccbin <host-compiler>" below
 ifneq ($(words $(subst ccache ,,$(CXX))),1) # allow at most "CXX=ccache <host-compiler>" from outside
   $(warning CUDA builds are not supported for multi-word CXX "$(CXX)")
   override CUDA_HOME=disabled
 endif
 
-# If CUDA_HOME is not set, try to set it from the location of nvcc
+# If CUDA_HOME is not set, try to set it from the path to nvcc
 ifndef CUDA_HOME
-  CUDA_HOME = $(patsubst %bin/nvcc,%,$(shell which nvcc 2>/dev/null))
+  CUDA_HOME = $(patsubst %/bin/nvcc,%,$(shell which nvcc 2>/dev/null))
   $(warning CUDA_HOME was not set: using "$(CUDA_HOME)")
 endif
 
-# Set NVCC as $(CUDA_HOME)/bin/nvcc if it exists
+# If HIP_HOME is not set, try to set it from the path to hipcc
+ifndef HIP_HOME
+  HIP_HOME = $(patsubst %/bin/hipcc,%,$(shell which hipcc 2>/dev/null))
+  $(warning HIP_HOME was not set: using "$(HIP_HOME)")
+endif
+
+# FIXME! (AV 24.01.2024)
+# In the current implementation (without separate builds for C++ and CUDA/HIP),
+# builds are performed for HIP only if CUDA is not found in the path.
+# If both CUDA and HIP are installed, HIP builds can be triggered by unsetting CUDA_HOME.
+# This will be fixed when separate C++ and CUDA/HIP builds are implemented (PR #775).
+
+#--- Option 1: CUDA exists -> use CUDA
+
+# Set GPUCC as $(CUDA_HOME)/bin/nvcc if it exists
 ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
-  NVCC = $(CUDA_HOME)/bin/nvcc
+
+  GPUCC = $(CUDA_HOME)/bin/nvcc
   USE_NVTX ?=-DUSE_NVTX
   # See https://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/index.html
   # See https://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/
@@ -148,7 +169,7 @@ ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
   # Examples: use 60 for P100 (Piz Daint), 80 for A100 (Juwels Booster, NVidia raplab/Curiosity).
   MADGRAPH_CUDA_ARCHITECTURE ?= 70
   ###CUARCHFLAGS = -gencode arch=compute_$(MADGRAPH_CUDA_ARCHITECTURE),code=compute_$(MADGRAPH_CUDA_ARCHITECTURE) -gencode arch=compute_$(MADGRAPH_CUDA_ARCHITECTURE),code=sm_$(MADGRAPH_CUDA_ARCHITECTURE) # Older implementation (AV): go back to this one for multi-GPU support #533
-  ###CUARCHFLAGS = --gpu-architecture=compute_$(MADGRAPH_CUDA_ARCHITECTURE) --gpu-code=sm_$(MADGRAPH_CUDA_ARCHITECTURE),compute_$(MADGRAPH_CUDA_ARCHITECTURE) # Newer implementation (SH): cannot use this as-is for multi-GPU support #533
+  ###CUARCHFLAGS = --gpu-architecture=compute_$(MADGRAPH_CUDA_ARCHITECTURE) --gpu-code=sm_$(MADGRAPH_CUDA_ARCHITECTURE),compute_$(MADGRAPH_CUDA_ARCHITECTURE)  # Newer implementation (SH): cannot use this as-is for multi-GPU support #533
   comma:=,
   CUARCHFLAGS = $(foreach arch,$(subst $(comma), ,$(MADGRAPH_CUDA_ARCHITECTURE)),-gencode arch=compute_$(arch),code=compute_$(arch) -gencode arch=compute_$(arch),code=sm_$(arch))
   CUINC = -I$(CUDA_HOME)/include/
@@ -158,41 +179,79 @@ ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
     CURANDLIBFLAGS = -L$(CUDA_HOME)/lib64/ -lcurand # NB: -lcuda is not needed here!
   endif
   CUOPTFLAGS = -lineinfo
-  CUFLAGS = $(foreach opt, $(OPTFLAGS), -Xcompiler $(opt)) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
-  ###CUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
-  ###NVCC_VERSION = $(shell $(NVCC) --version | grep 'Cuda compilation tools' | cut -d' ' -f5 | cut -d, -f1)
-  CUFLAGS += -std=c++17 # need CUDA >= 11.2 (see #333): this is enforced in mgOnGpuConfig.h
+  ###GPUFLAGS = $(OPTFLAGS) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
+  GPUFLAGS = $(foreach opt, $(OPTFLAGS), -Xcompiler $(opt)) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
+  ###GPUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
+  ###GPUCC_VERSION = $(shell $(GPUCC) --version | grep 'Cuda compilation tools' | cut -d' ' -f5 | cut -d, -f1)
+  GPUFLAGS += -std=c++17 # need CUDA >= 11.2 (see #333): this is enforced in mgOnGpuConfig.h
   # Without -maxrregcount: baseline throughput: 6.5E8 (16384 32 12) up to 7.3E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 160 # improves throughput: 6.9E8 (16384 32 12) up to 7.7E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 128 # improves throughput: 7.3E8 (16384 32 12) up to 7.6E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 96 # degrades throughput: 4.1E8 (16384 32 12) up to 4.5E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 64 # degrades throughput: 1.7E8 (16384 32 12) flat at 1.7E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 160 # improves throughput: 6.9E8 (16384 32 12) up to 7.7E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 128 # improves throughput: 7.3E8 (16384 32 12) up to 7.6E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 96 # degrades throughput: 4.1E8 (16384 32 12) up to 4.5E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 64 # degrades throughput: 1.7E8 (16384 32 12) flat at 1.7E8 (65536 128 12)
+  CUBUILDRULEFLAGS = -Xcompiler -fPIC -c
+  CCBUILDRULEFLAGS = -Xcompiler -fPIC -c -x cu
+  CUDATESTFLAGS = -lcuda
+
+  # Set the host C++ compiler for GPUCC via "-ccbin <host-compiler>"
+  # (NB issue #505: this must be a single word, "clang++ --gcc-toolchain..." is not supported)
+  GPUFLAGS += -ccbin $(shell which $(subst ccache ,,$(CXX)))
+
+  # Allow newer (unsupported) C++ compilers with older versions of CUDA if ALLOW_UNSUPPORTED_COMPILER_IN_CUDA is set (#504)
+  ifneq ($(origin ALLOW_UNSUPPORTED_COMPILER_IN_CUDA),undefined)
+  GPUFLAGS += -allow-unsupported-compiler
+  endif
+
 else ifneq ($(origin REQUIRE_CUDA),undefined)
+
   # If REQUIRE_CUDA is set but no cuda is found, stop here (e.g. for CI tests on GPU #443)
-  $(error No cuda installation found (set CUDA_HOME or make nvcc visible in PATH))
+  $(error No cuda installation found (set CUDA_HOME or make GPUCC visible in PATH))
+
+#--- Option 2: CUDA does not exist, HIP exists -> use HIP
+
+# Set GPUCC as $(HIP_HOME)/bin/hipcc if it exists
+else ifneq ($(wildcard $(HIP_HOME)/bin/hipcc),)
+
+  GPUCC = $(HIP_HOME)/bin/hipcc
+  #USE_NVTX ?=-DUSE_NVTX # should maybe find something equivalent to this in HIP?
+  HIPARCHFLAGS = -target x86_64-linux-gnu --offload-arch=gfx90a
+  HIPINC = -I$(HIP_HOME)/include/
+  # Note: -DHIP_FAST_MATH is equivalent to -use_fast_math in HIP 
+  # (but only for single precision line 208: https://rocm-developer-tools.github.io/HIP/hcc__detail_2math__functions_8h_source.html)
+  # Note: CUOPTFLAGS should not be used for HIP, it had been added here but was then removed (#808)
+  GPUFLAGS = $(OPTFLAGS) $(INCFLAGS) $(HIPINC) $(HIPARCHFLAGS) -DHIP_FAST_MATH -DHIP_PLATFORM=amd -fPIC
+  ###GPUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
+  GPUFLAGS += -std=c++17
+  ###GPUFLAGS+= --maxrregcount 255 # (AV: is this option valid on HIP and meaningful on AMD GPUs?)
+  CUBUILDRULEFLAGS = -fPIC -c
+  CCBUILDRULEFLAGS = -fPIC -c -x hip
+
+else ifneq ($(origin REQUIRE_HIP),undefined)
+
+  # If REQUIRE_HIP is set but no HIP is found, stop here (e.g. for CI tests on GPU #443)
+  $(error No hip installation found (set HIP_HOME or make GPUCC visible in PATH))
+
+#--- Option 3: CUDA does not exist, HIP does not exist -> switch off both CUDA and HIP
+
 else
-  # No cuda. Switch cuda compilation off and go to common random numbers in C++
+
+  # No cudacc and no hipcc: switch CUDA and HIP compilation off and go to common random numbers in C++
   $(warning CUDA_HOME is not set or is invalid: export CUDA_HOME to compile with cuda)
-  override NVCC=
+  $(warning HIP_HOME is not set or is invalid: export HIP_HOME to compile with hip)
+  override GPUCC=
   override USE_NVTX=
   override CUINC=
   override CURANDLIBFLAGS=
-endif
-export NVCC
-export CUFLAGS
 
-# Set the host C++ compiler for nvcc via "-ccbin <host-compiler>"
-# (NB issue #505: this must be a single word, "clang++ --gcc-toolchain..." is not supported)
-CUFLAGS += -ccbin $(shell which $(subst ccache ,,$(CXX)))
-
-# Allow newer (unsupported) C++ compilers with older versions of CUDA if ALLOW_UNSUPPORTED_COMPILER_IN_CUDA is set (#504)
-ifneq ($(origin ALLOW_UNSUPPORTED_COMPILER_IN_CUDA),undefined)
-CUFLAGS += -allow-unsupported-compiler
 endif
 
+# Export GPUCC (so that it can also be used in cudacpp_src.mk?)
+export GPUCC
+export GPUFLAGS
+
 #-------------------------------------------------------------------------------
 
-#=== Configure ccache for C++ and CUDA builds
+#=== Configure ccache for C++ and CUDA/HIP builds
 
 # Enable ccache if USECCACHE=1
 ifeq ($(USECCACHE)$(shell echo $(CXX) | grep ccache),1)
@@ -201,15 +260,15 @@ endif
 #ifeq ($(USECCACHE)$(shell echo $(AR) | grep ccache),1)
 #  override AR:=ccache $(AR)
 #endif
-ifneq ($(NVCC),)
-  ifeq ($(USECCACHE)$(shell echo $(NVCC) | grep ccache),1)
-    override NVCC:=ccache $(NVCC)
+ifneq ($(GPUCC),)
+  ifeq ($(USECCACHE)$(shell echo $(GPUCC) | grep ccache),1)
+    override GPUCC:=ccache $(GPUCC)
   endif
 endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure PowerPC-specific compiler flags for C++ and CUDA
+#=== Configure PowerPC-specific compiler flags for C++ and CUDA/HIP
 
 # PowerPC-specific CXX compiler flags (being reviewed)
 ifeq ($(UNAME_P),ppc64le)
@@ -225,9 +284,9 @@ else
   ######CXXFLAGS+= -fno-semantic-interposition # no benefit (neither alone, nor combined with -flto)
 endif
 
-# PowerPC-specific CUDA compiler flags (to be reviewed!)
+# PowerPC-specific CUDA/HIP compiler flags (to be reviewed!)
 ifeq ($(UNAME_P),ppc64le)
-  CUFLAGS+= -Xcompiler -mno-float128
+  GPUFLAGS+= -Xcompiler -mno-float128
 endif
 
 #-------------------------------------------------------------------------------
@@ -235,9 +294,11 @@ endif
 #=== Configure defaults and check if user-defined choices exist for OMPFLAGS, AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
 
 # Set the default OMPFLAGS choice
-ifneq ($(shell $(CXX) --version | egrep '^Intel'),)
+ifneq ($(findstring hipcc,$(GPUCC)),)
+override OMPFLAGS = # disable OpenMP MT when using hipcc #802
+else ifneq ($(shell $(CXX) --version | egrep '^Intel'),)
 override OMPFLAGS = -fopenmp
-###override OMPFLAGS = # disable OpenMP MT on Intel (was ok without nvcc but not ok with nvcc before #578)
+###override OMPFLAGS = # disable OpenMP MT on Intel (was ok without GPUCC but not ok with GPUCC before #578)
 else ifneq ($(shell $(CXX) --version | egrep '^(clang)'),)
 override OMPFLAGS = -fopenmp
 ###override OMPFLAGS = # disable OpenMP MT on clang (was not ok without or with nvcc before #578)
@@ -293,7 +354,10 @@ endif
 
 # Set the default RNDGEN (random number generator) choice
 ifeq ($(RNDGEN),)
-  ifeq ($(NVCC),)
+  ifeq ($(GPUCC),)
+    override RNDGEN = hasNoCurand
+  # Edgecase for HIP compilation
+  else ifeq ($(findstring hipcc,$(GPUCC)),hipcc)
     override RNDGEN = hasNoCurand
   else ifeq ($(RNDGEN),)
     override RNDGEN = hasCurand
@@ -310,7 +374,7 @@ export OMPFLAGS
 
 #-------------------------------------------------------------------------------
 
-#=== Set the CUDA/C++ compiler flags appropriate to user-defined choices of AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
+#=== Set the CUDA/HIP/C++ compiler flags appropriate to user-defined choices of AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
 
 # Set the build flags appropriate to OMPFLAGS
 $(info OMPFLAGS=$(OMPFLAGS))
@@ -368,13 +432,13 @@ CXXFLAGS+= $(AVXFLAGS)
 $(info FPTYPE=$(FPTYPE))
 ifeq ($(FPTYPE),d)
   CXXFLAGS += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
-  CUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
+  GPUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
 else ifeq ($(FPTYPE),f)
   CXXFLAGS += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
-  CUFLAGS  += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
+  GPUFLAGS  += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
 else ifeq ($(FPTYPE),m)
   CXXFLAGS += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
-  CUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
+  GPUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
 else
   $(error Unknown FPTYPE='$(FPTYPE)': only 'd', 'f' and 'm' are supported)
 endif
@@ -383,7 +447,7 @@ endif
 $(info HELINL=$(HELINL))
 ifeq ($(HELINL),1)
   CXXFLAGS += -DMGONGPU_INLINE_HELAMPS
-  CUFLAGS  += -DMGONGPU_INLINE_HELAMPS
+  GPUFLAGS  += -DMGONGPU_INLINE_HELAMPS
 else ifneq ($(HELINL),0)
   $(error Unknown HELINL='$(HELINL)': only '0' and '1' are supported)
 endif
@@ -392,7 +456,7 @@ endif
 $(info HRDCOD=$(HRDCOD))
 ifeq ($(HRDCOD),1)
   CXXFLAGS += -DMGONGPU_HARDCODE_PARAM
-  CUFLAGS  += -DMGONGPU_HARDCODE_PARAM
+  GPUFLAGS  += -DMGONGPU_HARDCODE_PARAM
 else ifneq ($(HRDCOD),0)
   $(error Unknown HRDCOD='$(HRDCOD)': only '0' and '1' are supported)
 endif
@@ -444,11 +508,11 @@ ifeq ($(UNAME_S),Darwin)
   override CULIBFLAGSRPATH2 =
 else
   # RPATH to cuda/cpp libs when linking executables
-  override CXXLIBFLAGSRPATH = -Wl,-rpath,$(LIBDIRRPATH)
-  override CULIBFLAGSRPATH = -Xlinker -rpath,$(LIBDIRRPATH)
+  override CXXLIBFLAGSRPATH = -Wl,-rpath=$(LIBDIRRPATH)
+  override CULIBFLAGSRPATH = -Xlinker -rpath=$(LIBDIRRPATH)
   # RPATH to common lib when linking cuda/cpp libs
-  override CXXLIBFLAGSRPATH2 = -Wl,-rpath,'$$ORIGIN'
-  override CULIBFLAGSRPATH2 = -Xlinker -rpath,'$$ORIGIN'
+  override CXXLIBFLAGSRPATH2 = -Wl,-rpath='$$ORIGIN'
+  override CULIBFLAGSRPATH2 = -Xlinker -rpath='$$ORIGIN'
 endif
 
 # Setting LD_LIBRARY_PATH or DYLD_LIBRARY_PATH in the RUNTIME is no longer necessary (neither on Linux nor on Mac)
@@ -461,7 +525,7 @@ override RUNTIME =
 cxx_main=$(BUILDDIR)/check.exe
 fcxx_main=$(BUILDDIR)/fcheck.exe
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 cu_main=$(BUILDDIR)/gcheck.exe
 fcu_main=$(BUILDDIR)/fgcheck.exe
 else
@@ -492,14 +556,15 @@ $(BUILDDIR)/.build.$(TAG):
 	@touch $(BUILDDIR)/.build.$(TAG)
 
 # Generic target and build rules: objects from CUDA compilation
-ifneq ($(NVCC),)
+# NB: CCBUILDRULEFLAGS includes "-x cu" for nvcc and "-x hip" for hipcc (#810)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/%.o : %.cu *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $(CUBUILDRULEFLAGS) $< -o $@
 
 $(BUILDDIR)/%_cu.o : %.cc *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c -x cu $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $(CCBUILDRULEFLAGS) $< -o $@
 endif
 
 # Generic target and build rules: objects from C++ compilation
@@ -509,11 +574,14 @@ $(BUILDDIR)/%.o : %.cc *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	$(CXX) $(CPPFLAGS) $(CXXFLAGS) -fPIC -c $< -o $@
 
 # Apply special build flags only to CrossSectionKernel.cc and gCrossSectionKernel.cu (no fast math, see #117 and #516)
+# Added edgecase for HIP compilation
 ifeq ($(shell $(CXX) --version | grep ^nvc++),)
 $(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS := $(filter-out -ffast-math,$(CXXFLAGS))
 $(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS += -fno-fast-math
-ifneq ($(NVCC),)
-$(BUILDDIR)/gCrossSectionKernels.o: CUFLAGS += -Xcompiler -fno-fast-math
+ifeq ($(findstring nvcc,$(GPUCC)),nvcc)
+  $(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -Xcompiler -fno-fast-math
+else
+  $(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -fno-fast-math
 endif
 endif
 
@@ -530,10 +598,10 @@ ifeq ($(RNDGEN),hasCurand)
 $(BUILDDIR)/CurandRandomNumberKernel.o: CXXFLAGS += $(CUINC)
 endif
 
-# Avoid "warning: builtin __has_trivial_... is deprecated; use __is_trivially_... instead" in nvcc with icx2023 (#592)
+# Avoid "warning: builtin __has_trivial_... is deprecated; use __is_trivially_... instead" in GPUCC with icx2023 (#592)
 ifneq ($(shell $(CXX) --version | egrep '^(Intel)'),)
-ifneq ($(NVCC),)
-CUFLAGS += -Xcompiler -Wno-deprecated-builtins
+ifneq ($(GPUCC),)
+GPUFLAGS += -Wno-deprecated-builtins
 endif
 endif
 
@@ -541,8 +609,8 @@ endif
 # This patch does remove the warning, but I prefer to keep it disabled for the moment...
 ###ifneq ($(shell $(CXX) --version | egrep '^(clang|Apple clang|Intel)'),)
 ###$(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS += -Wno-overriding-t-option
-###ifneq ($(NVCC),)
-###$(BUILDDIR)/gCrossSectionKernels.o: CUFLAGS += -Xcompiler -Wno-overriding-t-option
+###ifneq ($(GPUCC),)
+###$(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -Xcompiler -Wno-overriding-t-option
 ###endif
 ###endif
 
@@ -569,7 +637,7 @@ MG5AMC_CXXLIB = mg5amc_$(processid_short)_cpp
 cxx_objects_lib=$(BUILDDIR)/CPPProcess.o $(BUILDDIR)/MatrixElementKernels.o $(BUILDDIR)/BridgeKernels.o $(BUILDDIR)/CrossSectionKernels.o
 cxx_objects_exe=$(BUILDDIR)/CommonRandomNumberKernel.o $(BUILDDIR)/RamboSamplingKernels.o
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 MG5AMC_CULIB = mg5amc_$(processid_short)_cuda
 cu_objects_lib=$(BUILDDIR)/gCPPProcess.o $(BUILDDIR)/gMatrixElementKernels.o $(BUILDDIR)/gBridgeKernels.o $(BUILDDIR)/gCrossSectionKernels.o
 cu_objects_exe=$(BUILDDIR)/gCommonRandomNumberKernel.o $(BUILDDIR)/gRamboSamplingKernels.o
@@ -581,11 +649,17 @@ $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: cxx_objects_lib += $(BUILDDIR)/fbridge.o
 $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib)
 	$(CXX) -shared -o $@ $(cxx_objects_lib) $(CXXLIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: $(BUILDDIR)/fbridge_cu.o
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: cu_objects_lib += $(BUILDDIR)/fbridge_cu.o
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cu_objects_lib)
-	$(NVCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+# Bypass std::filesystem completely to ease portability on LUMI #803
+#ifneq ($(findstring hipcc,$(GPUCC)),)
+#	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB) -lstdc++fs
+#else
+#	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+#endif
 endif
 
 #-------------------------------------------------------------------------------
@@ -602,16 +676,16 @@ $(cxx_main): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PAT
 $(cxx_main): $(BUILDDIR)/check_sa.o $(LIBDIR)/lib$(MG5AMC_CXXLIB).so $(cxx_objects_exe) $(BUILDDIR)/CurandRandomNumberKernel.o
 	$(CXX) -o $@ $(BUILDDIR)/check_sa.o $(OMPFLAGS) -ldl -pthread $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe) $(BUILDDIR)/CurandRandomNumberKernel.o $(CURANDLIBFLAGS)
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(cu_main): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(cu_main): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(cu_main): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(cu_main): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 else ifneq ($(shell $(CXX) --version | grep ^nvc++),) # support nvc++ #531
 $(cu_main): LIBFLAGS += -L$(patsubst %bin/nvc++,%lib,$(subst ccache ,,$(CXX))) -lnvhpcatm -lnvcpumath -lnvc
 endif
 $(cu_main): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(cu_main): $(BUILDDIR)/gcheck_sa.o $(LIBDIR)/lib$(MG5AMC_CULIB).so $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o
-	$(NVCC) -o $@ $(BUILDDIR)/gcheck_sa.o $(CUARCHFLAGS) $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o $(CURANDLIBFLAGS)
+	$(GPUCC) -o $@ $(BUILDDIR)/gcheck_sa.o $(CUARCHFLAGS) $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o $(CURANDLIBFLAGS)
 endif
 
 #-------------------------------------------------------------------------------
@@ -635,19 +709,27 @@ $(fcxx_main): LIBFLAGS += -L$(shell dirname $(shell $(FC) --print-file-name libg
 endif
 $(fcxx_main): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(fcxx_main): $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler.o $(LIBDIR)/lib$(MG5AMC_CXXLIB).so $(cxx_objects_exe)
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(BUILDDIR)/fcheck_sa.o $(OMPFLAGS) $(BUILDDIR)/fsampler.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe) -lstdc++
+else
 	$(CXX) -o $@ $(BUILDDIR)/fcheck_sa.o $(OMPFLAGS) $(BUILDDIR)/fsampler.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe)
+endif
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(fcu_main): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(fcu_main): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(fcu_main): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(fcu_main): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 endif
 ifeq ($(UNAME_S),Darwin)
 $(fcu_main): LIBFLAGS += -L$(shell dirname $(shell $(FC) --print-file-name libgfortran.dylib)) # add path to libgfortran on Mac #375
 endif
 $(fcu_main): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(fcu_main): $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBDIR)/lib$(MG5AMC_CULIB).so $(cu_objects_exe)
-	$(NVCC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe)
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) -lstdc++ -L$(shell dirname $(shell $(GPUCC) -print-prog-name=clang))/../../lib -lamdhip64
+else
+	$(GPUCC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe)
+endif
 endif
 
 #-------------------------------------------------------------------------------
@@ -659,7 +741,7 @@ $(BUILDDIR)/testxxx.o: testxxx_cc_ref.txt
 $(testmain): $(BUILDDIR)/testxxx.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/testxxx.o # Comment out this line to skip the C++ test of xxx functions
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/testxxx_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/testxxx_cu.o: INCFLAGS += $(GTESTINC)
 $(BUILDDIR)/testxxx_cu.o: testxxx_cc_ref.txt
@@ -672,7 +754,7 @@ $(BUILDDIR)/testmisc.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/testmisc.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/testmisc.o # Comment out this line to skip the C++ miscellaneous tests
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/testmisc_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/testmisc_cu.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/testmisc_cu.o
@@ -684,12 +766,12 @@ $(BUILDDIR)/runTest.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/runTest.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/runTest.o
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/runTest_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/runTest_cu.o: INCFLAGS += $(GTESTINC)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(testmain): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(testmain): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(testmain): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(testmain): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 else ifneq ($(shell $(CXX) --version | grep ^nvc++),) # support nvc++ #531
 $(testmain): LIBFLAGS += -L$(patsubst %bin/nvc++,%lib,$(subst ccache ,,$(CXX))) -lnvhpcatm -lnvcpumath -lnvc
 endif
@@ -713,14 +795,23 @@ $(testmain): LIBFLAGS += -lgomp
 endif
 endif
 
-ifeq ($(NVCC),) # link only runTest.o
+# Bypass std::filesystem completely to ease portability on LUMI #803
+#ifneq ($(findstring hipcc,$(GPUCC)),)
+#$(testmain): LIBFLAGS += -lstdc++fs
+#endif
+
+ifeq ($(GPUCC),) # link only runTest.o
 $(testmain): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(testmain): $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib) $(cxx_objects_exe) $(GTESTLIBS)
 	$(CXX) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) -ldl -pthread $(LIBFLAGS)
 else # link both runTest.o and runTest_cu.o
 $(testmain): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(testmain): $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) $(GTESTLIBS)
-	$(NVCC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) -lcuda
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) $(CUDATESTFLAGS) -lstdc++ -lpthread  -L$(shell dirname $(shell $(GPUCC) -print-prog-name=clang))/../../lib -lamdhip64
+else
+	$(GPUCC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) $(CUDATESTFLAGS)
+endif
 endif
 
 # Use target gtestlibs to build only googletest
@@ -829,9 +920,9 @@ ifeq ($(USECCACHE),1)
 	ccache --version | head -1
 endif
 	@echo ""
-	@echo NVCC=$(NVCC)
-ifneq ($(NVCC),)
-	$(NVCC) --version
+	@echo GPUCC=$(GPUCC)
+ifneq ($(GPUCC),)
+	$(GPUCC) --version
 endif
 	@echo ""
 	@echo CXX=$(CXX)
@@ -850,7 +941,7 @@ endif
 
 # Target: check (run the C++ test executable)
 # [NB THIS IS WHAT IS USED IN THE GITHUB CI!]
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 check: runTest cmpFcheck cmpFGcheck
 else
 check: runTest cmpFcheck
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/fbridge.cc b/epochX/cudacpp/gq_ttq.sa/SubProcesses/fbridge.cc
index 2d2b36d560..22ce3f5115 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/fbridge.cc
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/fbridge.cc
@@ -1,11 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Roiser (Oct 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "Bridge.h"
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 
 extern "C"
 {
@@ -22,7 +22,7 @@ extern "C"
    * Using the same Fortran MadEvent code, linking to the hetrerogeneous library would allow access to both CPU and GPU implementations.
    * The specific heterogeneous configuration (how many GPUs, how many threads on each CPU, etc) could be loaded in CUDA/C++ from a data file.
    */
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -46,8 +46,8 @@ extern "C"
    */
   void fbridgecreate_( CppObjectInFortran** ppbridge, const int* pnevtF, const int* pnparF, const int* pnp4F )
   {
-#ifdef __CUDACC__
-    CudaRuntime::setUp();
+#ifdef MGONGPUCPP_GPUIMPL
+    GpuRuntime::setUp();
 #endif
     // (NB: CPPProcess::initProc no longer needs to be executed here because it is called in the Bridge constructor)
     // FIXME: disable OMP in Bridge when called from Fortran
@@ -65,8 +65,8 @@ extern "C"
     Bridge<FORTRANFPTYPE>* pbridge = dynamic_cast<Bridge<FORTRANFPTYPE>*>( *ppbridge );
     if( pbridge == 0 ) throw std::runtime_error( "fbridgedelete_: invalid Bridge address" );
     delete pbridge;
-#ifdef __CUDACC__
-    CudaRuntime::tearDown();
+#ifdef MGONGPUCPP_GPUIMPL
+    GpuRuntime::tearDown();
 #endif
   }
 
@@ -96,7 +96,7 @@ extern "C"
   {
     Bridge<FORTRANFPTYPE>* pbridge = dynamic_cast<Bridge<FORTRANFPTYPE>*>( *ppbridge );
     if( pbridge == 0 ) throw std::runtime_error( "fbridgesequence_: invalid Bridge address" );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Use the device/GPU implementation in the CUDA library
     // (there is also a host implementation in this library)
     pbridge->gpu_sequence( momenta, gs, rndhel, rndcol, *pchannelId, mes, selhel, selcol );
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/fsampler.cc b/epochX/cudacpp/gq_ttq.sa/SubProcesses/fsampler.cc
index 2fb445372d..3743934f41 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/fsampler.cc
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/fsampler.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Feb 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "mgOnGpuConfig.h"
 
@@ -13,7 +13,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -40,7 +40,7 @@ namespace mg5amcCpu
   private:
     const int m_nevt; // The number of events in each iteration
     int m_iiter;      // The iteration counter (for random number seeding)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     HostBufferRndNumMomenta m_hstRndmom; // Memory buffers for random numbers
     HostBufferMomenta m_hstMomenta;      // Memory buffers for momenta
     HostBufferWeights m_hstWeights;      // Memory buffers for sampling weights
@@ -105,7 +105,7 @@ namespace mg5amcCpu
 
 extern "C"
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/runTest.cc b/epochX/cudacpp/gq_ttq.sa/SubProcesses/runTest.cc
index d4a760a71b..de327f2321 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/runTest.cc
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/runTest.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Hageboeck (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*xxx to run only testxxx.cc tests
 //----------------------------------------------------------------------------
@@ -18,7 +18,7 @@
 #include "RandomNumberKernels.h"
 #include "epoch_process_id.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using namespace mg5amcGpu;
 #else
 using namespace mg5amcCpu;
@@ -35,7 +35,7 @@ struct CUDA_CPU_TestBase : public TestDriverBase
     : TestDriverBase( npar, refFileName ) {}
 };
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 struct CPUTest : public CUDA_CPU_TestBase
 {
   // Struct data members (process, and memory structures for random numbers, momenta, matrix elements and weights on host and device)
@@ -119,7 +119,7 @@ struct CPUTest : public CUDA_CPU_TestBase
 };
 #endif
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 struct CUDATest : public CUDA_CPU_TestBase
 {
   // Reset the device when our test goes out of scope. Note that this should happen after
@@ -128,7 +128,7 @@ struct CUDATest : public CUDA_CPU_TestBase
   {
     ~DeviceReset()
     {
-      checkCuda( cudaDeviceReset() ); // this is needed by cuda-memcheck --leak-check full
+      checkGpu( gpuDeviceReset() ); // this is needed by cuda-memcheck --leak-check full
     }
   } deviceResetter;
 
@@ -256,7 +256,7 @@ INSTANTIATE_TEST_SUITE_P( prefix, \
                           test_suite_name, \
                           testing::Values( new CUDATest( MG_EPOCH_REFERENCE_FILE_NAME ) ) );
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 MG_INSTANTIATE_TEST_SUITE_GPU( XTESTID_GPU( MG_EPOCH_PROCESS_ID ), MadgraphTest );
 #else
 MG_INSTANTIATE_TEST_SUITE_CPU( XTESTID_CPU( MG_EPOCH_PROCESS_ID ), MadgraphTest );
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/testmisc.cc b/epochX/cudacpp/gq_ttq.sa/SubProcesses/testmisc.cc
index 895d6eeb56..ba9e59a8a3 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/testmisc.cc
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/testmisc.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*misc to run only testmisc.cc tests
 //----------------------------------------------------------------------------
@@ -17,7 +17,7 @@
 #include <sstream>
 #include <typeinfo>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #define TESTID( s ) s##_GPU_MISC
 #else
 #define TESTID( s ) s##_CPU_MISC
@@ -26,7 +26,7 @@
 #define XTESTID( s ) TESTID( s )
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -59,7 +59,7 @@ namespace mg5amcCpu
 
 TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testmisc )
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gq_ttq.sa/SubProcesses/testxxx.cc b/epochX/cudacpp/gq_ttq.sa/SubProcesses/testxxx.cc
index 3361fe5aa9..e5167de00c 100644
--- a/epochX/cudacpp/gq_ttq.sa/SubProcesses/testxxx.cc
+++ b/epochX/cudacpp/gq_ttq.sa/SubProcesses/testxxx.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Apr 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*xxx to run only testxxx.cc tests
 //----------------------------------------------------------------------------
@@ -24,7 +24,7 @@
 #include <iomanip>
 #include <iostream>
 #include <vector>
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #define TESTID( s ) s##_GPU_XXX
 #else
 #define TESTID( s ) s##_CPU_XXX
@@ -32,7 +32,7 @@
 
 #define XTESTID( s ) TESTID( s )
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -42,7 +42,7 @@ namespace mg5amcCpu
   int FPEhandlerIevt = -1;
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU): '" << FPEhandlerMessage << "' ievt=" << FPEhandlerIevt << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU neppV=" << neppV << "): '" << FPEhandlerMessage << "' ievt=" << FPEhandlerIevt << std::endl;
@@ -53,7 +53,7 @@ namespace mg5amcCpu
 
 TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -77,7 +77,7 @@ TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
   assert( nevt % neppM == 0 ); // nevt must be a multiple of neppM
   assert( nevt % neppV == 0 ); // nevt must be a multiple of neppV
   // Fill in the input momenta
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   mg5amcGpu::PinnedHostBufferMomenta hstMomenta( nevt ); // AOSOA[npagM][npar=4][np4=4][neppM]
 #else
   mg5amcCpu::HostBufferMomenta hstMomenta( nevt ); // AOSOA[npagM][npar=4][np4=4][neppM]
@@ -322,7 +322,7 @@ TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
   {
     for( int ievt = 0; ievt < nevt; ievt++ )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       using namespace mg5amcGpu;
 #else
       using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gq_ttq.sa/src/HelAmps_sm.h b/epochX/cudacpp/gq_ttq.sa/src/HelAmps_sm.h
index cd4e6de668..45000c7246 100644
--- a/epochX/cudacpp/gq_ttq.sa/src/HelAmps_sm.h
+++ b/epochX/cudacpp/gq_ttq.sa/src/HelAmps_sm.h
@@ -5,7 +5,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -28,7 +28,7 @@
 //#include <iomanip>
 //#include <iostream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gq_ttq.sa/src/Parameters_sm.cc b/epochX/cudacpp/gq_ttq.sa/src/Parameters_sm.cc
index c06dcbb252..8b92ea0bd6 100644
--- a/epochX/cudacpp/gq_ttq.sa/src/Parameters_sm.cc
+++ b/epochX/cudacpp/gq_ttq.sa/src/Parameters_sm.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -17,7 +17,7 @@
 #include <iomanip>
 #include <iostream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using namespace mg5amcGpu;
 #else
 using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/gq_ttq.sa/src/Parameters_sm.h b/epochX/cudacpp/gq_ttq.sa/src/Parameters_sm.h
index a6eb185434..a3615ec77a 100644
--- a/epochX/cudacpp/gq_ttq.sa/src/Parameters_sm.h
+++ b/epochX/cudacpp/gq_ttq.sa/src/Parameters_sm.h
@@ -27,7 +27,7 @@
 #include "read_slha.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -93,7 +93,7 @@ namespace mg5amcCpu
 #include <limits>
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -217,7 +217,7 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -236,7 +236,7 @@ namespace mg5amcCpu
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wunused-variable"  // e.g. <<warning: unused variable ‘mdl_G__exp__2’ [-Wunused-variable]>>
 #pragma GCC diagnostic ignored "-Wunused-parameter" // e.g. <<warning: unused parameter ‘G’ [-Wunused-parameter]>>
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #pragma nv_diagnostic push
 #pragma nv_diag_suppress 177 // e.g. <<warning #177-D: variable "mdl_G__exp__2" was declared but never referenced>>
 #endif
@@ -263,7 +263,7 @@ namespace mg5amcCpu
       // End SM implementation - no special handling of vectors of floats as in EFT (#439)
       return out;
     }
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #pragma GCC diagnostic pop
 #pragma nv_diagnostic pop
 #endif
diff --git a/epochX/cudacpp/gq_ttq.sa/src/cudacpp_src.mk b/epochX/cudacpp/gq_ttq.sa/src/cudacpp_src.mk
index d4cc628aec..b2b9da5288 100644
--- a/epochX/cudacpp/gq_ttq.sa/src/cudacpp_src.mk
+++ b/epochX/cudacpp/gq_ttq.sa/src/cudacpp_src.mk
@@ -19,7 +19,7 @@ SHELL := /bin/bash
 #=== Configure common compiler flags for CUDA and C++
 
 INCFLAGS = -I.
-OPTFLAGS = -O3 # this ends up in CUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
+OPTFLAGS = -O3 # this ends up in GPUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
 
 #-------------------------------------------------------------------------------
 
@@ -45,13 +45,13 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure the CUDA compiler (note: NVCC is already exported including ccache)
+#=== Configure the CUDA compiler (note: GPUCC is already exported including ccache)
 
-###$(info NVCC=$(NVCC))
+###$(info GPUCC=$(GPUCC))
 
 #-------------------------------------------------------------------------------
 
-#=== Configure ccache for C++ builds (note: NVCC is already exported including ccache)
+#=== Configure ccache for C++ builds (note: GPUCC is already exported including ccache)
 
 # Enable ccache if USECCACHE=1
 ifeq ($(USECCACHE)$(shell echo $(CXX) | grep ccache),1)
@@ -92,6 +92,13 @@ endif
 ###$(info OMPFLAGS=$(OMPFLAGS))
 CXXFLAGS += $(OMPFLAGS)
 
+# Add correct flags for nvcc (-x cu) and hipcc (-x hip) for GPU code (see #810)
+ifeq ($(findstring nvcc,$(GPUCC)),nvcc)
+  GPUFLAGS += -Xcompiler -fPIC -c -x cu
+else ifeq ($(findstring hipcc,$(GPUCC)),hipcc)
+  GPUFLAGS += -fPIC -c -x hip
+endif
+
 # Set the build flags appropriate to each AVX choice (example: "make AVX=none")
 # [NB MGONGPU_PVW512 is needed because "-mprefer-vector-width=256" is not exposed in a macro]
 # [See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96476]
@@ -253,20 +260,20 @@ $(BUILDDIR)/%.o : %.cc *.h $(BUILDDIR)/.build.$(TAG)
 # Generic target and build rules: objects from CUDA compilation
 $(BUILDDIR)/%_cu.o : %.cc *.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c -x cu $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $< -o $@
 
 #-------------------------------------------------------------------------------
 
 cxx_objects=$(addprefix $(BUILDDIR)/, Parameters_sm.o read_slha.o)
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 cu_objects=$(addprefix $(BUILDDIR)/, Parameters_sm_cu.o)
 endif
 
 # Target (and build rules): common (src) library
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so : $(cxx_objects) $(cu_objects)
 	@if [ ! -d $(LIBDIR) ]; then echo "mkdir -p $(LIBDIR)"; mkdir -p $(LIBDIR); fi
-	$(NVCC) -shared -o $@ $(cxx_objects) $(cu_objects) $(LDFLAGS)
+	$(GPUCC) -shared -o $@ $(cxx_objects) $(cu_objects) $(LDFLAGS)
 else
 $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so : $(cxx_objects)
 	@if [ ! -d $(LIBDIR) ]; then echo "mkdir -p $(LIBDIR)"; mkdir -p $(LIBDIR); fi
diff --git a/epochX/cudacpp/gq_ttq.sa/src/mgOnGpuConfig.h b/epochX/cudacpp/gq_ttq.sa/src/mgOnGpuConfig.h
index b247654dcf..475749ca7c 100644
--- a/epochX/cudacpp/gq_ttq.sa/src/mgOnGpuConfig.h
+++ b/epochX/cudacpp/gq_ttq.sa/src/mgOnGpuConfig.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jul 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUCONFIG_H
 #define MGONGPUCONFIG_H 1
@@ -10,12 +10,26 @@
 // There are two different code bases for standalone_cudacpp (without multichannel) and madevent+cudacpp (with multichannel)
 #undef MGONGPU_SUPPORTS_MULTICHANNEL
 
+// Is this a GPU (CUDA, HIP) or CPU implementation?
+#ifdef __CUDACC__
+#define MGONGPUCPP_GPUIMPL cuda
+#elif defined __HIPCC__
+#define MGONGPUCPP_GPUIMPL hip
+#include "hip/hip_runtime.h" // needed for blockDim, blockIdx, threadIdx: better in mgOnGpuConfig.h than in GpuAbstraction.h
+#else
+#undef MGONGPUCPP_GPUIMPL
+#endif
+
 // ** NB1 Throughputs (e.g. 6.8E8) are events/sec for "./gcheck.exe -p 65536 128 12"
 // ** NB2 Baseline on b7g47n0004 fluctuates (probably depends on load on other VMs)
 
 // Choose if curand is supported for generating random numbers
+// For HIP, by default, do not use curand (common random numbers will be used instead)
 // For both CUDA and C++, by default, do not inline, but allow this macro to be set from outside with e.g. -DMGONGPU_HAS_NO_CURAND
-// (there exist CUDA installations, e.g. using the HPC package, which do not include curand - see PR #784)
+// (there exist CUDA installations, e.g. using the HPC package, which do not include curand - see PR #784 and #785)
+#if defined __HIPCC__
+#define MGONGPU_HAS_NO_CURAND 1
+#else
 //#ifdef __CUDACC__
 //#undef MGONGPU_HAS_NO_CURAND // default
 ////#define MGONGPU_HAS_NO_CURAND 1
@@ -23,6 +37,7 @@
 //#undef MGONGPU_HAS_NO_CURAND // default
 ////#define MGONGPU_HAS_NO_CURAND 1
 //#endif
+#endif
 
 // Choose floating point precision (for everything but color algebra #537)
 // If one of these macros has been set from outside with e.g. -DMGONGPU_FPTYPE_FLOAT, nothing happens (issue #167)
@@ -54,23 +69,31 @@
 //#undef MGONGPU_HARDCODE_PARAM // default
 ////#define MGONGPU_HARDCODE_PARAM 1
 
-// Complex type in c++: std::complex or cxsmpl (CHOOSE ONLY ONE)
-#ifndef __CUDACC__
-//#define MGONGPU_CPPCXTYPE_STDCOMPLEX 1 // ~8 percent slower on float, same on double (5.1E6/double, 9.4E6/float)
-#define MGONGPU_CPPCXTYPE_CXSMPL 1 // new default (5.1E6/double, 10.2E6/float)
-#endif
-
-// Complex type in cuda: thrust or cucomplex or cxsmpl (CHOOSE ONLY ONE)
+// Complex type in CUDA: thrust or cucomplex or cxsmpl (CHOOSE ONLY ONE)
+// (NB THIS IS MGONGPU_*CU*CXTYPE_xxx)
 #ifdef __CUDACC__
 #define MGONGPU_CUCXTYPE_THRUST 1 // default (~1.15E9/double, ~3.2E9/float)
 //#define MGONGPU_CUCXTYPE_CUCOMPLEX 1 // ~10 percent slower (1.03E9/double, ~2.8E9/float)
 //#define MGONGPU_CUCXTYPE_CXSMPL 1 // ~10 percent slower (1.00E9/double, ~2.9E9/float)
+
+// Complex type in HIP: cxsmpl (ONLY ONE OPTION POSSIBLE? #810)
+// (NB THIS IS MGONGPU_*HIP*CXTYPE_xxx)
+#elif defined __HIPCC__
+#define MGONGPU_HIPCXTYPE_CXSMPL 1 // default for HIP
+
+// Complex type in C++: std::complex or cxsmpl (CHOOSE ONLY ONE)
+// (NB THIS IS MGONGPU_*CPP*CXTYPE_xxx)
+#else
+//#define MGONGPU_CPPCXTYPE_STDCOMPLEX 1 // ~8 percent slower on float, same on double (5.1E6/double, 9.4E6/float)
+#define MGONGPU_CPPCXTYPE_CXSMPL 1 // new default (5.1E6/double, 10.2E6/float)
 #endif
 
-// Cuda nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
+// CUDA nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
 #ifdef __CUDACC__
-#undef MGONGPU_NSIGHT_DEBUG // default
+#undef MGONGPU_NSIGHT_DEBUG // default in CUDA
 //#define MGONGPU_NSIGHT_DEBUG 1
+#else
+#undef MGONGPU_NSIGHT_DEBUG // only option in HIP or C++
 #endif
 
 // SANITY CHECKS (floating point precision for everything but color algebra #537)
@@ -86,17 +109,21 @@
 #error You cannot use double precision for color algebra and single precision elsewhere
 #endif
 
-// SANITY CHECKS (c++ complex number implementation)
-#ifndef __CUDACC__
-#if defined MGONGPU_CPPCXTYPE_STDCOMPLEX and defined MGONGPU_CPPCXTYPE_CXSMPL
-#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CPPCXTYPE_STDCOMPLEX or MGONGPU_CPPCXTYPE_CXSMPL
+// SANITY CHECKS (CUDA complex number implementation)
+#ifdef __CUDACC__
+#if defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CUCOMPLEX
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX for CUDA
+#elif defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CXSMPL for CUDA
+#elif defined MGONGPU_CUCXTYPE_CUCOMPLEX and defined MGONGPU_CUCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE OF MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL for CUDA
 #endif
 #endif
 
-// SANITY CHECKS (cuda complex number implementation)
-#ifdef __CUDACC__
-#if defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CUCOMPLEX and defined MGONGPU_CUCXTYPE_CXSMPL
-#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL
+// SANITY CHECKS (C++ complex number implementation)
+#ifndef MGONGPUCPP_GPUIMPL
+#if defined MGONGPU_CPPCXTYPE_STDCOMPLEX and defined MGONGPU_CPPCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CPPCXTYPE_STDCOMPLEX or MGONGPU_CPPCXTYPE_CXSMPL for C++
 #endif
 #endif
 
@@ -134,7 +161,7 @@ namespace mgOnGpu
   // Alignment requirement for using reinterpret_cast with SIMD vectorized code
   // (using reinterpret_cast with non aligned memory may lead to segmentation faults!)
   // Only needed for C++ code but can be enforced also in NVCC builds of C++ code using CUDA>=11.2 and C++17 (#318, #319, #333)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   constexpr int cppAlign = 64; // alignment requirement for SIMD vectorization (64-byte i.e. 512-bit)
 #endif
 
@@ -145,7 +172,7 @@ using mgOnGpu::fptype;
 using mgOnGpu::fptype2;
 
 // C++ SIMD vectorization width (this will be used to set neppV)
-#ifdef __CUDACC__ // CUDA implementation has no SIMD
+#ifdef MGONGPUCPP_GPUIMPL // CUDA and HIP implementations have no SIMD
 #undef MGONGPU_CPPSIMD
 #elif defined __AVX512VL__ && defined MGONGPU_PVW512 // C++ "512z" AVX512 with 512 width (512-bit ie 64-byte): 8 (DOUBLE) or 16 (FLOAT)
 #ifdef MGONGPU_FPTYPE_DOUBLE
@@ -175,9 +202,9 @@ using mgOnGpu::fptype2;
 #undef MGONGPU_CPPSIMD
 #endif
 
-// Cuda nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
+// CUDA nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
 // Arguments (not used so far): text is __FUNCTION__, code is 0 (start) or 1 (end)
-#if defined __CUDACC__ && defined MGONGPU_NSIGHT_DEBUG /* clang-format off */
+#if defined __CUDA__ && defined MGONGPU_NSIGHT_DEBUG /* clang-format off */
 #define mgDebugDeclare() __shared__ float mgDebugCounter[mgOnGpu::ntpbMAX];
 #define mgDebugInitialise() { mgDebugCounter[threadIdx.x] = 0; }
 #define mgDebug( code, text ) { mgDebugCounter[threadIdx.x] += 1; }
@@ -189,8 +216,8 @@ using mgOnGpu::fptype2;
 #define mgDebugFinalise() { /*noop*/ }
 #endif /* clang-format on */
 
-// Define empty CUDA declaration specifiers for C++
-#ifndef __CUDACC__
+// Define empty CUDA/HIP declaration specifiers for C++
+#ifndef MGONGPUCPP_GPUIMPL
 #define __global__
 #define __host__
 #define __device__
diff --git a/epochX/cudacpp/gq_ttq.sa/src/mgOnGpuCxtypes.h b/epochX/cudacpp/gq_ttq.sa/src/mgOnGpuCxtypes.h
index ca9a9f00c0..7ede1dbfae 100644
--- a/epochX/cudacpp/gq_ttq.sa/src/mgOnGpuCxtypes.h
+++ b/epochX/cudacpp/gq_ttq.sa/src/mgOnGpuCxtypes.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022, based on earlier work by D. Smith) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUCXTYPES_H
 #define MGONGPUCXTYPES_H 1
@@ -19,7 +19,7 @@
 #include <complex>
 
 // Complex type in cuda: thrust or cucomplex or cxsmpl
-#ifdef __CUDACC__
+#ifdef __CUDACC__ // this must be __CUDAC__ (not MGONGPUCPP_GPUIMPL)
 #if defined MGONGPU_CUCXTYPE_THRUST
 #pragma clang diagnostic push
 #pragma clang diagnostic ignored "-Wtautological-compare" // for icpx2021/clang13 (https://stackoverflow.com/a/15864661)
@@ -30,8 +30,13 @@
 #elif not defined MGONGPU_CUCXTYPE_CXSMPL
 #error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL
 #endif
+// Complex type in HIP: cxsmpl
+#elif defined __HIPCC__
+#if not defined MGONGPU_HIPCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_HIPCXTYPE_CXSMPL
+#endif
 #else
-// Complex type in c++: std::complex or cxsmpl
+// Complex type in c++ or HIP: std::complex or cxsmpl
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
 #include <cmath>
 #elif not defined MGONGPU_CPPCXTYPE_CXSMPL
@@ -82,7 +87,7 @@ namespace mgOnGpu /* clang-format off */
 using mgOnGpu::cxsmpl;
 
 // Printout to stream for user defined types
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -92,7 +97,7 @@ namespace mg5amcCpu
   inline __host__ std::ostream&
   operator<<( std::ostream& out, const cxsmpl<FP>& c )
   {
-    out << std::complex( c.real(), c.imag() );
+    out << std::complex<FP>( c.real(), c.imag() );
     return out;
   }
 
@@ -215,14 +220,14 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
   // --- Type definitions (complex type: cxtype)
-#ifdef __CUDACC__ // cuda
+#ifdef __CUDACC__ // this must be __CUDAC__ (not MGONGPUCPP_GPUIMPL)
 #if defined MGONGPU_CUCXTYPE_THRUST
   typedef thrust::complex<fptype> cxtype;
 #elif defined MGONGPU_CUCXTYPE_CUCOMPLEX
@@ -255,13 +260,13 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
-#if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+#if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
 
   //------------------------------
   // CUDA or C++ - using cxsmpl
@@ -303,11 +308,11 @@ namespace mg5amcCpu
     return cxmake( c.real(), c.imag() );
   }
 
-#endif // #if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+#endif // #if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
 
   //==========================================================================
 
-#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_THRUST // cuda + thrust
+#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_THRUST // cuda + thrust (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // CUDA - using thrust::complex
@@ -347,7 +352,7 @@ namespace mg5amcCpu
 
   //==========================================================================
 
-#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_CUCOMPLEX // cuda + cucomplex
+#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_CUCOMPLEX // cuda + cucomplex (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // CUDA - using cuComplex
@@ -566,7 +571,7 @@ namespace mg5amcCpu
 
   //==========================================================================
 
-#if not defined __CUDACC__ and defined MGONGPU_CPPCXTYPE_STDCOMPLEX // c++ + stdcomplex
+#if not defined __CUDACC__ and defined MGONGPU_CPPCXTYPE_STDCOMPLEX // c++/hip + stdcomplex (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // C++ - using std::complex
@@ -633,7 +638,7 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/gq_ttq.sa/src/mgOnGpuFptypes.h b/epochX/cudacpp/gq_ttq.sa/src/mgOnGpuFptypes.h
index 905c97d700..fa3a02664b 100644
--- a/epochX/cudacpp/gq_ttq.sa/src/mgOnGpuFptypes.h
+++ b/epochX/cudacpp/gq_ttq.sa/src/mgOnGpuFptypes.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUFPTYPES_H
 #define MGONGPUFPTYPES_H 1
@@ -12,7 +12,7 @@
 #include <cmath>
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL // cuda
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -20,7 +20,7 @@ namespace mg5amcCpu
 {
   //==========================================================================
 
-#ifdef __CUDACC__ // cuda
+#ifdef MGONGPUCPP_GPUIMPL // cuda
 
   //------------------------------
   // Floating point types - Cuda
@@ -64,11 +64,11 @@ namespace mg5amcCpu
 #endif
   }
 
-#endif // #ifdef __CUDACC__
+#endif // #ifdef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
   //------------------------------
   // Floating point types - C++
@@ -92,7 +92,7 @@ namespace mg5amcCpu
     return std::sqrt( f );
   }
 
-#endif // #ifndef __CUDACC__
+#endif // #ifndef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
diff --git a/epochX/cudacpp/gq_ttq.sa/src/mgOnGpuVectors.h b/epochX/cudacpp/gq_ttq.sa/src/mgOnGpuVectors.h
index e1299ba81e..cdae04326b 100644
--- a/epochX/cudacpp/gq_ttq.sa/src/mgOnGpuVectors.h
+++ b/epochX/cudacpp/gq_ttq.sa/src/mgOnGpuVectors.h
@@ -32,7 +32,7 @@
 #endif
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -131,7 +131,7 @@ namespace mg5amcCpu
 #endif
 #endif
 
-#else // i.e #ifndef MGONGPU_CPPSIMD (this includes #ifdef __CUDACC__)
+#else // i.e #ifndef MGONGPU_CPPSIMD (this includes #ifdef MGONGPUCPP_GPUIMPL)
 
   const int neppV = 1;
 
@@ -153,13 +153,13 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
   // Printout to stream for user defined types
 
@@ -805,11 +805,11 @@ namespace mg5amcCpu
 
 #endif // #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
 
-#endif // #ifndef __CUDACC__
+#endif // #ifndef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
   //------------------------------
   // Vector types - CUDA
@@ -853,12 +853,12 @@ namespace mg5amcCpu
     return mask;
   }
 
-#endif // #ifdef __CUDACC__
+#endif // #ifdef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
   // Scalar-or-vector types: scalar in CUDA, vector or scalar in C++
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   typedef bool bool_sv;
   typedef fptype fptype_sv;
   typedef fptype2 fptype2_sv;
@@ -879,7 +879,7 @@ namespace mg5amcCpu
 #endif
 
   // Scalar-or-vector zeros: scalar in CUDA, vector or scalar in C++
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   inline __host__ __device__ cxtype cxzero_sv(){ return cxtype( 0, 0 ); }
 #elif defined MGONGPU_CPPSIMD
   inline cxtype_v cxzero_sv() { return cxtype_v(); } // RRRR=0000 IIII=0000
diff --git a/epochX/cudacpp/gq_ttq.sa/src/rambo.h b/epochX/cudacpp/gq_ttq.sa/src/rambo.h
index e02ea52496..cd7e1008ea 100644
--- a/epochX/cudacpp/gq_ttq.sa/src/rambo.h
+++ b/epochX/cudacpp/gq_ttq.sa/src/rambo.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -18,7 +18,7 @@
 #include <iostream>
 
 // Simplified rambo version for 2 to N (with N>=2) processes with massless particles
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +83,7 @@ namespace mg5amcCpu
       static bool first = true;
       if( first )
       {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
         if constexpr( M_ACCESS::isOnDevice() ) // avoid
         {
           const int ievt0 = 0;
@@ -166,7 +166,7 @@ namespace mg5amcCpu
     wt = po2log;
     if( nparf != 2 ) wt = ( 2. * nparf - 4. ) * log( energy ) + z[nparf - 1];
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     // issue warnings if weight is too small or too large
     static int iwarn[5] = { 0, 0, 0, 0, 0 };
     if( wt < -180. )
diff --git a/epochX/cudacpp/gq_ttq.sa/src/read_slha.cc b/epochX/cudacpp/gq_ttq.sa/src/read_slha.cc
index 055b19a779..f8e46f2e66 100644
--- a/epochX/cudacpp/gq_ttq.sa/src/read_slha.cc
+++ b/epochX/cudacpp/gq_ttq.sa/src/read_slha.cc
@@ -11,7 +11,11 @@
 
 #include <algorithm>
 #include <cstdlib>
-#include <filesystem>
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
 #include <fstream>
 #include <iostream>
 
@@ -60,7 +64,14 @@ SLHAReader::read_slha_file( std::string file_name, bool verbose )
     {
       std::cout << "WARNING! Card file '" << file_name << "' does not exist:"
                 << " look for the file in directory $" << envpath << "='" << getenv( envpath ) << "'" << std::endl;
+      /*
+#ifdef __HIPCC__
+      const std::string file_name2 = std::experimental::filesystem::path( getenv( envpath ) ) / std::experimental::filesystem::path( file_name ).filename();
+#else
       const std::string file_name2 = std::filesystem::path( getenv( envpath ) ) / std::filesystem::path( file_name ).filename();
+#endif
+      */
+      const std::string file_name2 = std::string( getenv( envpath ) ) + "/" + file_name; // bypass std::filesystem #803
       param_card.open( file_name2.c_str(), std::ifstream::in );
       if( param_card.good() )
       {
diff --git a/epochX/cudacpp/heft_gg_h.sa/CODEGEN_cudacpp_heft_gg_h_log.txt b/epochX/cudacpp/heft_gg_h.sa/CODEGEN_cudacpp_heft_gg_h_log.txt
index 800492306f..1054438636 100644
--- a/epochX/cudacpp/heft_gg_h.sa/CODEGEN_cudacpp_heft_gg_h_log.txt
+++ b/epochX/cudacpp/heft_gg_h.sa/CODEGEN_cudacpp_heft_gg_h_log.txt
@@ -52,7 +52,7 @@ Note that you can still compile and run aMC@NLO with the built-in PDFs
 
 Using default text editor "vi". Set another one in ./input/mg5_configuration.txt
 Using default eps viewer "evince". Set another one in ./input/mg5_configuration.txt
-No valid web browser found. Please set in ./input/mg5_configuration.txt
+Using default web browser "firefox". Set another one in ./input/mg5_configuration.txt
 import /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_cudacpp_heft_gg_h.mg
 The import format was not given, so we guess it as command
 set stdout_level DEBUG
@@ -62,6 +62,12 @@ set auto_convert_model T
 save options auto_convert_model
 save configuration file to /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/mg5amcnlo/input/mg5_configuration.txt
 import model heft
+INFO: reload from .py file 
+INFO: load particles 
+INFO: load vertices 
+WARNING: coupling GC_13=-(complex(0,1)*GH) has direct dependence in aS but has QCD order set to 0. Automatic computation of scale uncertainty can be wrong for such model. 
+WARNING: coupling GC_16=(complex(0,1)*Gphi)/8. has direct dependence in aS but has QCD order set to 0. Automatic computation of scale uncertainty can be wrong for such model. 
+DEBUG: model prefixing  takes 0.005749940872192383 
 INFO: Restrict model heft with file models/heft/restrict_default.dat . 
 DEBUG: Simplifying conditional expressions 
 DEBUG: remove interactions: s u w+ at order: QED=1 
@@ -153,7 +159,7 @@ Generated helas calls for 1 subprocesses (1 diagrams) in 0.002 s
 DEBUG:  Entering PLUGIN_ProcessExporter.convert_model (create the model) [output.py at line 202] 
 ALOHA: aloha starts to compute helicity amplitudes
 ALOHA: aloha creates VVS3 routines
-ALOHA: aloha creates 1 routines in  0.062 s
+ALOHA: aloha creates 1 routines in  0.069 s
 <class 'aloha.create_aloha.AbstractRoutine'> VVS3
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_cudacpp_heft_gg_h/src/./HelAmps_heft.h
 INFO: Created file HelAmps_heft.h in directory /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_cudacpp_heft_gg_h/src/. 
@@ -165,7 +171,7 @@ INFO: Created files Parameters_heft.h and Parameters_heft.cc in directory
 INFO: /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_cudacpp_heft_gg_h/src/. and /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_cudacpp_heft_gg_h/src/. 
 quit
 
-real	0m0.471s
-user	0m0.367s
-sys	0m0.052s
-Code generation completed in 0 seconds
+real	0m0.616s
+user	0m0.387s
+sys	0m0.049s
+Code generation completed in 1 seconds
diff --git a/epochX/cudacpp/heft_gg_h.sa/COPYRIGHT b/epochX/cudacpp/heft_gg_h.sa/COPYRIGHT
index a134b5fef9..9036d9260a 100644
--- a/epochX/cudacpp/heft_gg_h.sa/COPYRIGHT
+++ b/epochX/cudacpp/heft_gg_h.sa/COPYRIGHT
@@ -15,6 +15,7 @@ The full development team currently includes the following authors :
   Stephan Hageboeck (CERN)
   Olivier Mattelaer (Universite Catholique de Louvain, original author)
   Stefan Roiser (CERN, original author)
+  Jorgen Teig (CERN)
   Andrea Valassi (CERN, original author)
   Zenny Wettersten (CERN)
 See https://github.com/madgraph5/madgraph4gpu for more details. For the full
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/Bridge.h b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/Bridge.h
index bf8b5e024d..f9ed70dfde 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/Bridge.h
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/Bridge.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Roiser (Nov 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef BRIDGE_H
 #define BRIDGE_H 1
@@ -14,16 +14,23 @@
 #include "MemoryAccessMomenta.h"  // for MemoryAccessMomenta::neppM
 #include "MemoryBuffers.h"        // for HostBufferMomenta, DeviceBufferMomenta etc
 
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
+
+#include <sys/stat.h> // bypass std::filesystem #803
+
 #include <algorithm>
 #include <cassert>
 #include <cmath>
 #include <cstring>
-#include <filesystem>
 #include <iostream>
 #include <memory>
 #include <type_traits>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +90,7 @@ namespace mg5amcCpu
     Bridge& operator=( const Bridge& ) = delete;
     Bridge& operator=( Bridge&& ) = delete;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     /**
      * Set the gpublocks and gputhreads for the gpusequence - throws if evnt != gpublocks*gputhreads
      * (this is needed for BridgeKernel tests rather than for actual production use in Fortran)
@@ -150,7 +157,7 @@ namespace mg5amcCpu
     unsigned int m_nevt; // number of events
     int m_nGoodHel;      // the number of good helicities (-1 initially when they have not yet been calculated)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     int m_gputhreads; // number of gpu threads (default set from number of events, can be modified)
     int m_gpublocks;  // number of gpu blocks (default set from number of events, can be modified)
     DeviceBuffer<FORTRANFPTYPE, sizePerEventMomenta> m_devMomentaF;
@@ -187,12 +194,12 @@ namespace mg5amcCpu
   // Forward declare transposition methods
   //
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
   template<typename Tin, typename Tout>
   __global__ void dev_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt );
 
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
 
   template<typename Tin, typename Tout>
   void hst_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt );
@@ -209,7 +216,7 @@ namespace mg5amcCpu
   Bridge<FORTRANFPTYPE>::Bridge( unsigned int nevtF, unsigned int nparF, unsigned int np4F )
     : m_nevt( nevtF )
     , m_nGoodHel( -1 )
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     , m_gputhreads( 256 )                  // default number of gpu threads
     , m_gpublocks( m_nevt / m_gputhreads ) // this ensures m_nevt <= m_gpublocks*m_gputhreads
     , m_devMomentaF( m_nevt )
@@ -233,7 +240,7 @@ namespace mg5amcCpu
   {
     if( nparF != CPPProcess::npar ) throw std::runtime_error( "Bridge constructor: npar mismatch" );
     if( np4F != CPPProcess::np4 ) throw std::runtime_error( "Bridge constructor: np4 mismatch" );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( ( m_nevt < s_gputhreadsmin ) || ( m_nevt % s_gputhreadsmin != 0 ) )
       throw std::runtime_error( "Bridge constructor: nevt should be a multiple of " + std::to_string( s_gputhreadsmin ) );
     while( m_nevt != m_gpublocks * m_gputhreads )
@@ -249,20 +256,28 @@ namespace mg5amcCpu
 #else
     std::cout << "WARNING! Instantiate host Bridge (nevt=" << m_nevt << ")" << std::endl;
     m_pmek.reset( new MatrixElementKernelHost( m_hstMomentaC, m_hstGs, m_hstRndHel, m_hstRndCol, m_hstMEs, m_hstSelHel, m_hstSelCol, m_nevt ) );
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
     // Create a process object, read param card and set parameters
     // FIXME: the process instance can happily go out of scope because it is only needed to read parameters?
     // FIXME: the CPPProcess should really be a singleton? what if fbridgecreate is called from several Fortran threads?
     CPPProcess process( /*verbose=*/false );
     std::string paramCard = "../../Cards/param_card.dat";
-    if( !std::filesystem::exists( paramCard ) )
-    {
-      paramCard = "../" + paramCard;
-    }
+    /*
+#ifdef __HIPCC__
+    if( !std::experimental::filesystem::exists( paramCard ) ) paramCard = "../" + paramCard;
+#else
+    if( !std::filesystem::exists( paramCard ) ) paramCard = "../" + paramCard;
+#endif
+    */
+    //struct stat dummybuffer; // bypass std::filesystem #803
+    //if( !( stat( paramCard.c_str(), &dummyBuffer ) == 0 ) ) paramCard = "../" + paramCard; //
+    auto fileExists = []( std::string& fileName )
+    { struct stat buffer; return stat( fileName.c_str(), &buffer ) == 0; };
+    if( !fileExists( paramCard ) ) paramCard = "../" + paramCard; // bypass std::filesystem #803
     process.initProc( paramCard );
   }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::set_gpugrid( const int gpublocks, const int gputhreads )
   {
@@ -276,7 +291,7 @@ namespace mg5amcCpu
   }
 #endif
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::gpu_sequence( const FORTRANFPTYPE* momenta,
                                             const FORTRANFPTYPE* gs,
@@ -291,14 +306,14 @@ namespace mg5amcCpu
     constexpr int neppM = MemoryAccessMomenta::neppM;
     if constexpr( neppM == 1 && std::is_same_v<FORTRANFPTYPE, fptype> )
     {
-      checkCuda( cudaMemcpy( m_devMomentaC.data(), momenta, m_devMomentaC.bytes(), cudaMemcpyHostToDevice ) );
+      gpuMemcpy( m_devMomentaC.data(), momenta, m_devMomentaC.bytes(), gpuMemcpyHostToDevice );
     }
     else
     {
-      checkCuda( cudaMemcpy( m_devMomentaF.data(), momenta, m_devMomentaF.bytes(), cudaMemcpyHostToDevice ) );
+      gpuMemcpy( m_devMomentaF.data(), momenta, m_devMomentaF.bytes(), gpuMemcpyHostToDevice );
       const int thrPerEvt = CPPProcess::npar * CPPProcess::np4; // AV: transpose alg does 1 element per thread (NOT 1 event per thread)
       //const int thrPerEvt = 1; // AV: try new alg with 1 event per thread... this seems slower
-      dev_transposeMomentaF2C<<<m_gpublocks * thrPerEvt, m_gputhreads>>>( m_devMomentaF.data(), m_devMomentaC.data(), m_nevt );
+      gpuLaunchKernel( dev_transposeMomentaF2C, m_gpublocks * thrPerEvt, m_gputhreads, m_devMomentaF.data(), m_devMomentaC.data(), m_nevt );
     }
     if constexpr( std::is_same_v<FORTRANFPTYPE, fptype> )
     {
@@ -341,7 +356,7 @@ namespace mg5amcCpu
   }
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::cpu_sequence( const FORTRANFPTYPE* momenta,
                                             const FORTRANFPTYPE* gs,
@@ -396,7 +411,7 @@ namespace mg5amcCpu
   // - C++ array: momenta[npagM][npar][np4][neppM] with nevt=npagM*neppM (AOSOA)
   //
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename Tin, typename Tout>
   __global__ void dev_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt )
   {
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/BridgeKernels.cc b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/BridgeKernels.cc
index d58066c9c1..eaf4037a24 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/BridgeKernels.cc
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/BridgeKernels.cc
@@ -1,17 +1,18 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "BridgeKernels.h"
 
+#include "GpuAbstraction.h"
 #include "MemoryAccessMomenta.h"
 
 #include <sstream>
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -45,7 +46,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu
 {
 
@@ -96,7 +97,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/BridgeKernels.h b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/BridgeKernels.h
index 15eb4bff4d..3efef8ce97 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/BridgeKernels.h
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/BridgeKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef BRIDGEKERNELS_H
 #define BRIDGEKERNELS_H 1
@@ -12,7 +12,7 @@
 #include "MatrixElementKernels.h"
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -49,7 +49,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A Bridge wrapper class encapsulating matrix element calculations on a CPU host
   class BridgeKernelHost final : public BridgeKernelBase
   {
@@ -89,7 +89,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A Bridge wrapper class encapsulating matrix element calculations on a GPU device
   class BridgeKernelDevice : public BridgeKernelBase
   {
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/CommonRandomNumberKernel.cc b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/CommonRandomNumberKernel.cc
index 985b39f576..010bc4cbd0 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/CommonRandomNumberKernel.cc
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/CommonRandomNumberKernel.cc
@@ -1,15 +1,16 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "CommonRandomNumbers.h"
+#include "GpuAbstraction.h"
 #include "MemoryBuffers.h"
 #include "RandomNumberKernels.h"
 
 #include <cassert>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/CrossSectionKernels.cc b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/CrossSectionKernels.cc
index 0b355a3c8d..c15b39844d 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/CrossSectionKernels.cc
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/CrossSectionKernels.cc
@@ -1,10 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "CrossSectionKernels.h"
 
+#include "GpuAbstraction.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessWeights.h"
 #include "MemoryBuffers.h"
@@ -77,7 +78,7 @@ debug_me_is_abnormal( const fptype& me, size_t ievtALL )
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -185,7 +186,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/CrossSectionKernels.h b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/CrossSectionKernels.h
index 7933ca4bbf..4d9659e04e 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/CrossSectionKernels.h
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/CrossSectionKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef CROSSSECTIONKERNELS_H
 #define CROSSSECTIONKERNELS_H 1
@@ -13,7 +13,7 @@
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -96,7 +96,7 @@ namespace mg5amcCpu
   //--------------------------------------------------------------------------
 
   /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating the calculation of event statistics on a GPU device
   class CrossSectionKernelDevice : public CrossSectionKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/CudaRuntime.h b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/CudaRuntime.h
deleted file mode 100644
index 64ce52f4b3..0000000000
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/CudaRuntime.h
+++ /dev/null
@@ -1,85 +0,0 @@
-// Copyright (C) 2020-2023 CERN and UCLouvain.
-// Licensed under the GNU Lesser General Public License (version 3 or later).
-// Created by: S. Roiser (Jul 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
-
-#ifndef MG5AMC_CUDARUNTIME_H
-#define MG5AMC_CUDARUNTIME_H 1
-
-// MG5AMC on GPU uses the CUDA runtime API, not the lower level CUDA driver API
-// See https://docs.nvidia.com/cuda/cuda-runtime-api/driver-vs-runtime-api.html#driver-vs-runtime-api
-
-#include <cassert>
-#include <iostream>
-
-//--------------------------------------------------------------------------
-
-// See https://stackoverflow.com/a/14038590
-#ifdef __CUDACC__ /* clang-format off */
-#define checkCuda( code ) { assertCuda( code, __FILE__, __LINE__ ); }
-inline void assertCuda( cudaError_t code, const char* file, int line, bool abort = true )
-{
-  if( code != cudaSuccess )
-  {
-    printf( "ERROR! assertCuda: '%s' (%d) in %s:%d\n", cudaGetErrorString( code ), code, file, line );
-    if( abort ) assert( code == cudaSuccess );
-  }
-}
-#endif /* clang-format on */
-
-//--------------------------------------------------------------------------
-
-#ifdef __CUDACC__
-namespace mg5amcGpu
-{
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  // *** FIXME! This will all need to be designed differently when going to multi-GPU nodes! ***
-  struct CudaRuntime final
-  {
-    CudaRuntime( const bool debug = true )
-      : m_debug( debug ) { setUp( m_debug ); }
-    ~CudaRuntime() { tearDown( m_debug ); }
-    CudaRuntime( const CudaRuntime& ) = delete;
-    CudaRuntime( CudaRuntime&& ) = delete;
-    CudaRuntime& operator=( const CudaRuntime& ) = delete;
-    CudaRuntime& operator=( CudaRuntime&& ) = delete;
-    bool m_debug;
-
-    // Set up CUDA application
-    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
-    // Calling cudaSetDevice on startup is useful to properly book-keep the time spent in CUDA initialization
-    static void setUp( const bool debug = true )
-    {
-      // ** NB: it is useful to call cudaSetDevice, or cudaFree, to properly book-keep the time spent in CUDA initialization
-      // ** NB: otherwise, the first CUDA operation (eg a cudaMemcpyToSymbol in CPPProcess ctor) appears to take much longer!
-      /*
-      // [We initially added cudaFree(0) to "ease profile analysis" only because it shows up as a big recognizable block!]
-      // No explicit initialization is needed: https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#initialization
-      // It is not clear what cudaFree(0) does at all: https://stackoverflow.com/questions/69967813/
-      if ( debug ) std::cout << "__CudaRuntime: calling cudaFree(0)" << std::endl;
-      checkCuda( cudaFree( 0 ) ); // SLOW!
-      */
-      // Replace cudaFree(0) by cudaSetDevice(0), even if it is not really needed either
-      // (but see https://developer.nvidia.com/blog/cuda-pro-tip-always-set-current-device-avoid-multithreading-bugs)
-      if( debug ) std::cout << "__CudaRuntime: calling cudaSetDevice(0)" << std::endl;
-      checkCuda( cudaSetDevice( 0 ) ); // SLOW!
-    }
-
-    // Tear down CUDA application (call cudaDeviceReset)
-    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
-    // Calling cudaDeviceReset on shutdown is only needed for checking memory leaks in cuda-memcheck
-    // See https://docs.nvidia.com/cuda/cuda-memcheck/index.html#leak-checking
-    static void tearDown( const bool debug = true )
-    {
-      if( debug ) std::cout << "__CudaRuntime: calling cudaDeviceReset()" << std::endl;
-      checkCuda( cudaDeviceReset() );
-    }
-  };
-
-}
-#endif
-
-//--------------------------------------------------------------------------
-
-#endif // MG5AMC_CUDARUNTIME_H
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/CurandRandomNumberKernel.cc b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/CurandRandomNumberKernel.cc
index eb56333b03..08a16f6f2c 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/CurandRandomNumberKernel.cc
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/CurandRandomNumberKernel.cc
@@ -1,9 +1,9 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "MemoryBuffers.h"
 #include "RandomNumberKernels.h"
 
@@ -22,7 +22,7 @@ inline void assertCurand( curandStatus_t code, const char *file, int line, bool
 }
 #endif /* clang-format on */
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -36,7 +36,7 @@ namespace mg5amcCpu
   {
     if( m_isOnDevice )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if( !m_rnarray.isOnDevice() )
         throw std::runtime_error( "CurandRandomNumberKernel on device with a host random number array" );
 #else
@@ -114,7 +114,7 @@ namespace mg5amcCpu
     /*
     printf( "\nCurandRandomNumberKernel::generateRnarray size = %d\n", (int)m_rnarray.size() );
     fptype* data = m_rnarray.data();
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( m_rnarray.isOnDevice() )
     {
       data = new fptype[m_rnarray.size()]();
@@ -123,7 +123,7 @@ namespace mg5amcCpu
 #endif
     for( int i = 0; i < ( (int)m_rnarray.size() / 4 ); i++ )
       printf( "[%4d] %f %f %f %f\n", i * 4, data[i * 4], data[i * 4 + 2], data[i * 4 + 2], data[i * 4 + 3] );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( m_rnarray.isOnDevice() ) delete[] data;
 #endif
     */
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/EventStatistics.h b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/EventStatistics.h
index 48b51e0a49..b425a5bade 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/EventStatistics.h
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/EventStatistics.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef EventStatistics_H
 #define EventStatistics_H 1
@@ -16,7 +16,7 @@
 #include <limits>
 #include <string>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/GpuAbstraction.h b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/GpuAbstraction.h
new file mode 100644
index 0000000000..9c467b1e04
--- /dev/null
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/GpuAbstraction.h
@@ -0,0 +1,69 @@
+// Copyright (C) 2020-2023 CERN and UCLouvain.
+// Licensed under the GNU Lesser General Public License (version 3 or later).
+// Created by: J. Teig (Jul 2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+
+#ifndef MG5AMC_GPUABSTRACTION_H
+#define MG5AMC_GPUABSTRACTION_H 1
+
+#include <cassert>
+
+//--------------------------------------------------------------------------
+
+#ifdef __CUDACC__
+
+#define gpuError_t cudaError_t
+#define gpuPeekAtLastError cudaPeekAtLastError
+#define gpuGetErrorString cudaGetErrorString
+#define gpuSuccess cudaSuccess
+
+#define gpuMallocHost( ptr, size ) checkGpu( cudaMallocHost( ptr, size ) )
+#define gpuMalloc( ptr, size ) checkGpu( cudaMalloc( ptr, size ) )
+
+#define gpuMemcpy( dstData, srcData, srcBytes, func ) checkGpu( cudaMemcpy( dstData, srcData, srcBytes, func ) )
+#define gpuMemcpyHostToDevice cudaMemcpyHostToDevice
+#define gpuMemcpyDeviceToHost cudaMemcpyDeviceToHost
+#define gpuMemcpyToSymbol( type1, type2, size ) checkGpu( cudaMemcpyToSymbol( type1, type2, size ) )
+
+#define gpuFree( ptr ) checkGpu( cudaFree( ptr ) )
+#define gpuFreeHost( ptr ) checkGpu( cudaFreeHost( ptr ) )
+
+#define gpuSetDevice cudaSetDevice
+#define gpuDeviceSynchronize cudaDeviceSynchronize
+#define gpuDeviceReset cudaDeviceReset
+
+#define gpuLaunchKernel( kernel, blocks, threads, ... ) kernel<<<blocks, threads>>>( __VA_ARGS__ )
+#define gpuLaunchKernelSharedMem( kernel, blocks, threads, sharedMem, ... ) kernel<<<blocks, threads, sharedMem>>>( __VA_ARGS__ )
+
+//--------------------------------------------------------------------------
+
+#elif defined __HIPCC__
+
+#define gpuError_t hipError_t
+#define gpuPeekAtLastError hipPeekAtLastError
+#define gpuGetErrorString hipGetErrorString
+#define gpuSuccess hipSuccess
+
+#define gpuMallocHost( ptr, size ) checkGpu( hipHostMalloc( ptr, size ) ) // HostMalloc better
+#define gpuMalloc( ptr, size ) checkGpu( hipMalloc( ptr, size ) )
+
+#define gpuMemcpy( dstData, srcData, srcBytes, func ) checkGpu( hipMemcpy( dstData, srcData, srcBytes, func ) )
+#define gpuMemcpyHostToDevice hipMemcpyHostToDevice
+#define gpuMemcpyDeviceToHost hipMemcpyDeviceToHost
+#define gpuMemcpyToSymbol( type1, type2, size ) checkGpu( hipMemcpyToSymbol( type1, type2, size ) )
+
+#define gpuFree( ptr ) checkGpu( hipFree( ptr ) )
+#define gpuFreeHost( ptr ) checkGpu( hipHostFree( ptr ) )
+
+#define gpuSetDevice hipSetDevice
+#define gpuDeviceSynchronize hipDeviceSynchronize
+#define gpuDeviceReset hipDeviceReset
+
+#define gpuLaunchKernel( kernel, blocks, threads, ... ) kernel<<<blocks, threads>>>( __VA_ARGS__ )
+#define gpuLaunchKernelSharedMem( kernel, blocks, threads, sharedMem, ... ) kernel<<<blocks, threads, sharedMem>>>( __VA_ARGS__ )
+
+//--------------------------------------------------------------------------
+
+#endif
+
+#endif // MG5AMC_GPUABSTRACTION_H
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/GpuRuntime.h b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/GpuRuntime.h
new file mode 100644
index 0000000000..93579ef08b
--- /dev/null
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/GpuRuntime.h
@@ -0,0 +1,85 @@
+// Copyright (C) 2020-2023 CERN and UCLouvain.
+// Licensed under the GNU Lesser General Public License (version 3 or later).
+// Created by: J. Teig (Jun 2023, based on earlier work by S. Roiser) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+
+#ifndef MG5AMC_GPURUNTIME_H
+#define MG5AMC_GPURUNTIME_H 1
+
+// MG5AMC on GPU uses the CUDA runtime API, not the lower level CUDA driver API
+// See https://docs.nvidia.com/cuda/cuda-runtime-api/driver-vs-runtime-api.html#driver-vs-runtime-api
+
+#include "GpuAbstraction.h"
+
+#include <iostream>
+
+//--------------------------------------------------------------------------
+
+// See https://stackoverflow.com/a/14038590
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
+#define checkGpu( code ) { assertGpu( code, __FILE__, __LINE__ ); }
+inline void assertGpu( gpuError_t code, const char* file, int line, bool abort = true )
+{
+  if( code != gpuSuccess )
+  {
+    printf( "ERROR! assertGpu: '%s' (%d) in %s:%d\n", gpuGetErrorString( code ), code, file, line );
+    if( abort ) assert( code == gpuSuccess );
+  }
+}
+#endif /* clang-format on */
+
+//--------------------------------------------------------------------------
+
+#ifdef MGONGPUCPP_GPUIMPL
+namespace mg5amcGpu
+{
+  // Instantiate a GpuRuntime at the beginnining of the application's main to
+  // invoke gpuSetDevice(0) in the constructor and book a gpuDeviceReset() call in the destructor
+  // *** FIXME! This will all need to be designed differently when going to multi-GPU nodes! ***
+  struct GpuRuntime final
+  {
+    GpuRuntime( const bool debug = true )
+      : m_debug( debug ) { setUp( m_debug ); }
+    ~GpuRuntime() { tearDown( m_debug ); }
+    GpuRuntime( const GpuRuntime& ) = delete;
+    GpuRuntime( GpuRuntime&& ) = delete;
+    GpuRuntime& operator=( const GpuRuntime& ) = delete;
+    GpuRuntime& operator=( GpuRuntime&& ) = delete;
+    bool m_debug;
+
+    // Set up CUDA application
+    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
+    // Calling cudaSetDevice on startup is useful to properly book-keep the time spent in CUDA initialization
+    static void setUp( const bool debug = true )
+    {
+      // ** NB: it is useful to call cudaSetDevice, or cudaFree, to properly book-keep the time spent in CUDA initialization
+      // ** NB: otherwise, the first CUDA operation (eg a cudaMemcpyToSymbol in CPPProcess ctor) appears to take much longer!
+      /*
+      // [We initially added cudaFree(0) to "ease profile analysis" only because it shows up as a big recognizable block!]
+      // No explicit initialization is needed: https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#initialization
+      // It is not clear what cudaFree(0) does at all: https://stackoverflow.com/questions/69967813/
+      if ( debug ) std::cout << "__CudaRuntime: calling cudaFree(0)" << std::endl;
+      checkCuda( cudaFree( 0 ) ); // SLOW!
+      */
+      // Replace cudaFree(0) by cudaSetDevice(0), even if it is not really needed either
+      // (but see https://developer.nvidia.com/blog/cuda-pro-tip-always-set-current-device-avoid-multithreading-bugs)
+      if( debug ) std::cout << "__GpuRuntime: calling GpuSetDevice(0)" << std::endl;
+      checkGpu( gpuSetDevice( 0 ) ); // SLOW!
+    }
+
+    // Tear down CUDA application (call cudaDeviceReset)
+    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
+    // Calling cudaDeviceReset on shutdown is only needed for checking memory leaks in cuda-memcheck
+    // See https://docs.nvidia.com/cuda/cuda-memcheck/index.html#leak-checking
+    static void tearDown( const bool debug = true )
+    {
+      if( debug ) std::cout << "__GpuRuntime: calling GpuDeviceReset()" << std::endl;
+      checkGpu( gpuDeviceReset() );
+    }
+  };
+}
+#endif
+
+//--------------------------------------------------------------------------
+
+#endif // MG5AMC_GPURUNTIME_H
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MadgraphTest.h b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MadgraphTest.h
index ef40624c88..6054185300 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MadgraphTest.h
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MadgraphTest.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Hageboeck (Dec 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MADGRAPHTEST_H_
 #define MADGRAPHTEST_H_ 1
@@ -14,7 +14,11 @@
 
 #include <array>
 #include <cmath>
-#include <filesystem>
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
 #include <fstream>
 #include <iomanip>
 #include <memory>
@@ -22,7 +26,7 @@
 #include <string>
 #include <vector>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using mg5amcGpu::CPPProcess;
 #else
 using mg5amcCpu::CPPProcess;
@@ -201,7 +205,7 @@ class MadgraphTest : public testing::TestWithParam<TestDriverBase*>
 
 // Since we link both the CPU-only and GPU tests into the same executable, we prevent
 // a multiply defined symbol by only compiling this in the non-CUDA phase:
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
 /// Compare momenta and matrix elements.
 /// This uses an implementation of TestDriverBase to run a madgraph workflow,
@@ -219,7 +223,14 @@ TEST_P( MadgraphTest, CompareMomentaAndME )
   const char* dumpEventsC = getenv( "CUDACPP_RUNTEST_DUMPEVENTS" );
   const bool dumpEvents = ( dumpEventsC != 0 ) && ( std::string( dumpEventsC ) != "" );
   const std::string refFileName = testDriver->getRefFileName();
+  /*
+#ifdef __HIPCC__
+  const std::string dumpFileName = std::experimental::filesystem::path( refFileName ).filename();
+#else
   const std::string dumpFileName = std::filesystem::path( refFileName ).filename();
+#endif
+  */
+  const std::string dumpFileName = refFileName; // bypass std::filesystem #803
   std::ofstream dumpFile;
   if( dumpEvents )
   {
@@ -307,6 +318,6 @@ TEST_P( MadgraphTest, CompareMomentaAndME )
   }
 }
 
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
 
 #endif /* MADGRAPHTEST_H_ */
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MatrixElementKernels.cc b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MatrixElementKernels.cc
index 74b5239ebf..81699dfea9 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MatrixElementKernels.cc
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MatrixElementKernels.cc
@@ -1,12 +1,12 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "MatrixElementKernels.h"
 
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h" // Includes the abstraction for Nvidia/AMD compilation
 #include "MemoryAccessMomenta.h"
 #include "MemoryBuffers.h"
 
@@ -14,7 +14,7 @@
 
 //============================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu
 {
 
@@ -150,7 +150,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
@@ -209,13 +209,13 @@ namespace mg5amcGpu
     PinnedHostBufferHelicityMask hstIsGoodHel( ncomb );
     DeviceBufferHelicityMask devIsGoodHel( ncomb );
     // ... 0d1. Compute good helicity mask on the device
-    computeDependentCouplings<<<m_gpublocks, m_gputhreads>>>( m_gs.data(), m_couplings.data() );
+    gpuLaunchKernel( computeDependentCouplings, m_gpublocks, m_gputhreads, m_gs.data(), m_couplings.data() );
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-    sigmaKin_getGoodHel<<<m_gpublocks, m_gputhreads>>>( m_momenta.data(), m_couplings.data(), m_matrixElements.data(), m_numerators.data(), m_denominators.data(), devIsGoodHel.data() );
+    gpuLaunchKernel( sigmaKin_getGoodHel, m_gpublocks, m_gputhreads, m_momenta.data(), m_couplings.data(), m_matrixElements.data(), m_numerators.data(), m_denominators.data(), devIsGoodHel.data() );
 #else
-    sigmaKin_getGoodHel<<<m_gpublocks, m_gputhreads>>>( m_momenta.data(), m_couplings.data(), m_matrixElements.data(), devIsGoodHel.data() );
+    gpuLaunchKernel( sigmaKin_getGoodHel, m_gpublocks, m_gputhreads, m_momenta.data(), m_couplings.data(), m_matrixElements.data(), devIsGoodHel.data() );
 #endif
-    checkCuda( cudaPeekAtLastError() );
+    checkGpu( gpuPeekAtLastError() );
     // ... 0d2. Copy back good helicity mask to the host
     copyHostFromDevice( hstIsGoodHel, devIsGoodHel );
     // ... 0d3. Copy back good helicity list to constant memory on the device
@@ -226,19 +226,19 @@ namespace mg5amcGpu
 
   void MatrixElementKernelDevice::computeMatrixElements( const unsigned int channelId )
   {
-    computeDependentCouplings<<<m_gpublocks, m_gputhreads>>>( m_gs.data(), m_couplings.data() );
+    gpuLaunchKernel( computeDependentCouplings, m_gpublocks, m_gputhreads, m_gs.data(), m_couplings.data() );
 #ifndef MGONGPU_NSIGHT_DEBUG
     constexpr unsigned int sharedMemSize = 0;
 #else
     constexpr unsigned int sharedMemSize = ntpbMAX * sizeof( float );
 #endif
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-    sigmaKin<<<m_gpublocks, m_gputhreads, sharedMemSize>>>( m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), channelId, m_numerators.data(), m_denominators.data(), m_selhel.data(), m_selcol.data() );
+    gpuLaunchKernelSharedMem( sigmaKin, m_gpublocks, m_gputhreads, sharedMemSize, m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), channelId, m_numerators.data(), m_denominators.data(), m_selhel.data(), m_selcol.data() );
 #else
-    sigmaKin<<<m_gpublocks, m_gputhreads, sharedMemSize>>>( m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), m_selhel.data(), m_selcol.data() );
+    gpuLaunchKernelSharedMem( sigmaKin, m_gpublocks, m_gputhreads, sharedMemSize, m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), m_selhel.data(), m_selcol.data() );
 #endif
-    checkCuda( cudaPeekAtLastError() );
-    checkCuda( cudaDeviceSynchronize() );
+    checkGpu( gpuPeekAtLastError() );
+    checkGpu( gpuDeviceSynchronize() );
   }
 
   //--------------------------------------------------------------------------
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MatrixElementKernels.h b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MatrixElementKernels.h
index 23e84757a2..72bd8f195b 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MatrixElementKernels.h
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MatrixElementKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MATRIXELEMENTKERNELS_H
 #define MATRIXELEMENTKERNELS_H 1
@@ -10,7 +10,7 @@
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -81,7 +81,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating matrix element calculations on a CPU host
   class MatrixElementKernelHost final : public MatrixElementKernelBase, public NumberOfEvents
   {
@@ -130,7 +130,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating matrix element calculations on a GPU device
   class MatrixElementKernelDevice : public MatrixElementKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessAmplitudes.h b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessAmplitudes.h
index 573b3bbbc9..ffb76e93de 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessAmplitudes.h
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessAmplitudes.h
@@ -15,7 +15,7 @@
 #define MGONGPU_TRIVIAL_AMPLITUDES 1
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessCouplings.h b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessCouplings.h
index d65c9d6e04..85c3c9ed1c 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessCouplings.h
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessCouplings.h
@@ -15,7 +15,7 @@
 #include "MemoryBuffers.h"       // for HostBufferCouplings::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessCouplingsFixed.h b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessCouplingsFixed.h
index dc0d93afff..ffcdf4dbef 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessCouplingsFixed.h
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessCouplingsFixed.h
@@ -14,7 +14,7 @@
 //#include "MemoryAccessHelpers.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessDenominators.h b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessDenominators.h
index 3bce635718..66f2d32a6b 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessDenominators.h
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessDenominators.h
@@ -10,7 +10,7 @@
 #include "MemoryAccessGs.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessGs.h b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessGs.h
index 31311aa375..4c726b30f3 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessGs.h
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessGs.h
@@ -13,7 +13,7 @@
 #include "MemoryBuffers.h" // for HostBufferMatrixElements::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessHelpers.h b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessHelpers.h
index c82a6c7635..db73e4e064 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessHelpers.h
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessHelpers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessHelpers_H
 #define MemoryAccessHelpers_H 1
@@ -105,7 +105,7 @@ class KernelAccessHelper : public MemoryAccessHelper<T>
     }
     else
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
       //printf( "kernelAccessRecord: ievt=%d threadId=%d\n", ievt, threadIdx.x );
       return T::ieventAccessRecord( buffer, ievt ); // NB fptype and fptype_sv coincide for CUDA
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessMatrixElements.h b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessMatrixElements.h
index f32e6fea5b..3741011971 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessMatrixElements.h
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessMatrixElements.h
@@ -13,7 +13,7 @@
 #include "MemoryBuffers.h" // for HostBufferMatrixElements::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessMomenta.h b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessMomenta.h
index 29266de32c..3be229d392 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessMomenta.h
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessMomenta.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessMomenta_H
 #define MemoryAccessMomenta_H 1
@@ -13,7 +13,7 @@
 #include "MemoryAccessVectors.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -30,7 +30,7 @@ namespace mg5amcCpu
 
     // Number of Events Per Page in the momenta AOSOA memory buffer layout
     // (these are all best kept as a compile-time constants: see issue #23)
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
     // -----------------------------------------------------------------------------------------------
     // --- GPUs: neppM is best set to a power of 2 times the number of fptype's in a 32-byte cacheline
     // --- This is relevant to ensure coalesced access to momenta in global memory
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessNumerators.h b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessNumerators.h
index b152183b28..18991f4fa6 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessNumerators.h
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessNumerators.h
@@ -10,7 +10,7 @@
 #include "MemoryAccessGs.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessRandomNumbers.h b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessRandomNumbers.h
index e2988d39f3..40cb089135 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessRandomNumbers.h
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessRandomNumbers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessRandomNumbers_H
 #define MemoryAccessRandomNumbers_H 1
@@ -11,7 +11,7 @@
 #include "CPPProcess.h"
 #include "MemoryAccessHelpers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using mg5amcGpu::CPPProcess;
 #else
 using mg5amcCpu::CPPProcess;
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessVectors.h b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessVectors.h
index e9b197368e..08faccff0f 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessVectors.h
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessVectors.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessVectors_H
 #define MemoryAccessVectors_H 1
@@ -10,7 +10,7 @@
 
 #include "mgOnGpuVectors.h"
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu // this is only needed for CPU SIMD vectorization
 {
 
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessWavefunctions.h b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessWavefunctions.h
index 5428aaf933..33bef4559e 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessWavefunctions.h
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryAccessWavefunctions.h
@@ -15,7 +15,7 @@
 #define MGONGPU_TRIVIAL_WAVEFUNCTIONS 1
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryBuffers.h b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryBuffers.h
index 8109470148..78004e66cc 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryBuffers.h
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/MemoryBuffers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021, based on earlier work by S. Hageboeck) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryBuffers_H
 #define MemoryBuffers_H 1
@@ -11,12 +11,12 @@
 #include "mgOnGpuCxtypes.h"
 
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "Parameters_heft.h"
 
 #include <sstream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -87,7 +87,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   constexpr bool HostBufferALIGNED = false;   // ismisaligned=false
   constexpr bool HostBufferMISALIGNED = true; // ismisaligned=true
 
@@ -119,7 +119,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA pinned host buffer
   template<typename T>
   class PinnedHostBufferBase : public BufferBase<T>
@@ -128,18 +128,18 @@ namespace mg5amcCpu
     PinnedHostBufferBase( const size_t size )
       : BufferBase<T>( size, false )
     {
-      checkCuda( cudaMallocHost( &( this->m_data ), this->bytes() ) );
+      gpuMallocHost( &( this->m_data ), this->bytes() );
     }
     virtual ~PinnedHostBufferBase()
     {
-      checkCuda( cudaFreeHost( this->m_data ) );
+      gpuFreeHost( this->m_data );
     }
   };
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA device buffer
   template<typename T>
   class DeviceBufferBase : public BufferBase<T>
@@ -148,18 +148,18 @@ namespace mg5amcCpu
     DeviceBufferBase( const size_t size )
       : BufferBase<T>( size, true )
     {
-      checkCuda( cudaMalloc( &( this->m_data ), this->bytes() ) );
+      gpuMalloc( &( this->m_data ), this->bytes() );
     }
     virtual ~DeviceBufferBase()
     {
-      checkCuda( cudaFree( this->m_data ) );
+      gpuFree( this->m_data );
     }
   };
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for a given number of events
   template<typename T, size_t sizePerEvent, bool ismisaligned>
   class HostBuffer : public HostBufferBase<T, ismisaligned>, virtual private NumberOfEvents
@@ -175,7 +175,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA pinned host buffer for a given number of events
   template<typename T, size_t sizePerEvent>
   class PinnedHostBuffer : public PinnedHostBufferBase<T>, virtual private NumberOfEvents
@@ -191,7 +191,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA device buffer for a given number of events
   template<typename T, size_t sizePerEvent>
   class DeviceBuffer : public DeviceBufferBase<T>, virtual private NumberOfEvents
@@ -213,7 +213,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for momenta random numbers
   constexpr size_t sizePerEventRndNumMomenta = MemoryBuffers::np4 * MemoryBuffers::nparf;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for momenta random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumMomenta, HostBufferALIGNED> HostBufferRndNumMomenta;
 #else
@@ -232,7 +232,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer with ONE fptype per event
   constexpr size_t sizePerEventOneFp = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer with ONE fptype per event
   typedef HostBuffer<fptype, sizePerEventOneFp, HostBufferALIGNED> HostBufferOneFp;
 #else
@@ -257,7 +257,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for Gs
   constexpr size_t sizePerEventGs = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventGs, HostBufferALIGNED> HostBufferGs;
 #else
@@ -276,7 +276,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for numerators
   constexpr size_t sizePerEventNumerators = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventNumerators, HostBufferALIGNED> HostBufferNumerators;
 #else
@@ -296,7 +296,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for denominators
   constexpr size_t sizePerEventDenominators = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventDenominators, HostBufferALIGNED> HostBufferDenominators;
 #else
@@ -315,7 +315,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for random numbers
   constexpr size_t sizePerEventCouplings = MemoryBuffers::ndcoup * MemoryBuffers::nx2;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventCouplings, HostBufferALIGNED> HostBufferCouplings;
 #else
@@ -333,7 +333,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for momenta
   constexpr size_t sizePerEventMomenta = MemoryBuffers::np4 * MemoryBuffers::npar;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for momenta
   typedef HostBuffer<fptype, sizePerEventMomenta, HostBufferALIGNED> HostBufferMomenta;
   //typedef HostBuffer<fptype, sizePerEventMomenta, HostBufferMISALIGNED> HostBufferMomenta; // TEST MISALIGNMENT!
@@ -352,7 +352,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for sampling weights
   constexpr size_t sizePerEventWeights = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for sampling weights
   typedef HostBuffer<fptype, sizePerEventWeights, HostBufferALIGNED> HostBufferWeights;
 #else
@@ -370,7 +370,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for matrix elements
   constexpr size_t sizePerEventMatrixElements = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for matrix elements
   typedef HostBuffer<fptype, sizePerEventMatrixElements, HostBufferALIGNED> HostBufferMatrixElements;
 #else
@@ -385,7 +385,7 @@ namespace mg5amcCpu
   // A base class encapsulating a memory buffer for the helicity mask
   typedef BufferBase<bool> BufferHelicityMask;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for the helicity mask
   typedef HostBufferBase<bool, HostBufferALIGNED> HostBufferHelicityMask;
 #else
@@ -403,7 +403,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for wavefunctions
   constexpr size_t sizePerEventWavefunctions = MemoryBuffers::nw6 * MemoryBuffers::nx2;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for wavefunctions
   typedef HostBuffer<fptype, sizePerEventWavefunctions, HostBufferALIGNED> HostBufferWavefunctions;
 #else
@@ -421,7 +421,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for helicity random numbers
   constexpr size_t sizePerEventRndNumHelicity = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for helicity random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumHelicity, HostBufferALIGNED> HostBufferRndNumHelicity;
 #else
@@ -439,7 +439,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for color random numbers
   constexpr size_t sizePerEventRndNumColor = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for color random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumColor, HostBufferALIGNED> HostBufferRndNumColor;
 #else
@@ -457,7 +457,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for helicity selection
   constexpr size_t sizePerEventSelectedHelicity = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for helicity selection
   typedef HostBuffer<int, sizePerEventSelectedHelicity, HostBufferALIGNED> HostBufferSelectedHelicity;
 #else
@@ -475,7 +475,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for color selection
   constexpr size_t sizePerEventSelectedColor = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for color selection
   typedef HostBuffer<int, sizePerEventSelectedColor, HostBufferALIGNED> HostBufferSelectedColor;
 #else
@@ -487,7 +487,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<class Tdst, class Tsrc>
   void copyDeviceFromHost( Tdst& dst, const Tsrc& src ) // keep the same order of arguments as in memcpy
   {
@@ -504,13 +504,13 @@ namespace mg5amcCpu
       throw std::runtime_error( sstr.str() );
     }
     // NB (PR #45): cudaMemcpy involves an intermediate memcpy to pinned memory if host array is a not a pinned host array
-    checkCuda( cudaMemcpy( dst.data(), src.data(), src.bytes(), cudaMemcpyHostToDevice ) );
+    gpuMemcpy( dst.data(), src.data(), src.bytes(), gpuMemcpyHostToDevice );
   }
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<class Tdst, class Tsrc>
   void copyHostFromDevice( Tdst& dst, const Tsrc& src ) // keep the same order of arguments as in memcpy
   {
@@ -527,7 +527,7 @@ namespace mg5amcCpu
       throw std::runtime_error( sstr.str() );
     }
     // NB (PR #45): cudaMemcpy involves an intermediate memcpy to pinned memory if host array is a not a pinned host array
-    checkCuda( cudaMemcpy( dst.data(), src.data(), src.bytes(), cudaMemcpyDeviceToHost ) );
+    gpuMemcpy( dst.data(), src.data(), src.bytes(), gpuMemcpyDeviceToHost );
   }
 #endif
 
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/P1_Sigma_heft_gg_h/CPPProcess.cc b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/P1_Sigma_heft_gg_h/CPPProcess.cc
index 526bd7d296..624791e8b3 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/P1_Sigma_heft_gg_h/CPPProcess.cc
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/P1_Sigma_heft_gg_h/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_heft.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -46,7 +45,7 @@
 // Class member functions for calculating the matrix elements for
 // Process: g g > h HIG<=1 HIW<=1 WEIGHTED<=2 @1
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -80,7 +79,7 @@ namespace mg5amcCpu
   //__device__ const fptype* cIPD = nullptr; // unused as nparam=0
   __device__ const fptype* cIPC = nullptr; // unused as nicoup=0
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   //__device__ __constant__ fptype* cIPD = nullptr; // unused as nparam=0
   __device__ __constant__ fptype* cIPC = nullptr; // unused as nicoup=0
 #else
@@ -90,7 +89,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -118,13 +117,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -151,7 +150,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -187,7 +186,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -200,8 +199,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -268,7 +269,7 @@ namespace mg5amcCpu
       // [NB do keep 'static' for these constexpr arrays, see issue #283]
       static constexpr fptype2 cf[ncolor][ncolor] = { { 2 } }; // 2-D array[1][1]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -325,7 +326,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -384,7 +385,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -419,8 +420,8 @@ namespace mg5amcCpu
       { -1, 1, 0 },
       { 1, -1, 0 },
       { 1, 1, 0 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -459,9 +460,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     //const fptype tIPD[0] = { ... }; // nparam=0
     //const cxtype tIPC[0] = { ... }; // nicoup=0
-#ifdef __CUDACC__
-    //checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 0 * sizeof( fptype ) ) ); // nparam=0
-    //checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ) ); // nicoup=0
+#ifdef MGONGPUCPP_GPUIMPL
+    //gpuMemcpyToSymbol( cIPD, tIPD, 0 * sizeof( fptype ) ); // nparam=0
+    //gpuMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
 #else
     //memcpy( cIPD, tIPD, 0 * sizeof( fptype ) ); // nparam=0
     //memcpy( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
@@ -495,7 +496,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -524,6 +525,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -537,6 +542,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -560,12 +567,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -586,7 +593,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -712,9 +719,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -738,7 +745,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -758,7 +765,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 256 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -772,9 +779,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -802,7 +812,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -1012,7 +1022,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/P1_Sigma_heft_gg_h/CPPProcess.h b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/P1_Sigma_heft_gg_h/CPPProcess.h
index dbc5aa0e4e..e1caef360b 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/P1_Sigma_heft_gg_h/CPPProcess.h
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/P1_Sigma_heft_gg_h/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -107,7 +107,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -120,7 +120,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -150,7 +150,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/P1_Sigma_heft_gg_h/CudaRuntime.h b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/P1_Sigma_heft_gg_h/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/P1_Sigma_heft_gg_h/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/P1_Sigma_heft_gg_h/GpuAbstraction.h b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/P1_Sigma_heft_gg_h/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/P1_Sigma_heft_gg_h/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/P1_Sigma_heft_gg_h/GpuRuntime.h b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/P1_Sigma_heft_gg_h/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/P1_Sigma_heft_gg_h/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/P1_Sigma_heft_gg_h/check_sa.cc b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/P1_Sigma_heft_gg_h/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/P1_Sigma_heft_gg_h/check_sa.cc
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/P1_Sigma_heft_gg_h/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/RamboSamplingKernels.cc b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/RamboSamplingKernels.cc
index da68aa9255..79abbcc4f8 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/RamboSamplingKernels.cc
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/RamboSamplingKernels.cc
@@ -1,11 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "RamboSamplingKernels.h"
 
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "MemoryAccessMomenta.h"
 #include "MemoryAccessRandomNumbers.h"
 #include "MemoryAccessWeights.h"
@@ -14,7 +14,7 @@
 
 #include <sstream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -92,7 +92,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingKernelDevice::RamboSamplingKernelDevice( const fptype energy,               // input: energy
                                                         const BufferRndNumMomenta& rndmom, // input: random numbers in [0,1]
                                                         BufferMomenta& momenta,            // output: momenta
@@ -135,7 +135,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   getMomentaInitialDevice( const fptype energy,
                            fptype* momenta )
@@ -147,17 +147,17 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   void
   RamboSamplingKernelDevice::getMomentaInitial()
   {
-    getMomentaInitialDevice<<<m_gpublocks, m_gputhreads>>>( m_energy, m_momenta.data() );
+    gpuLaunchKernel( getMomentaInitialDevice, m_gpublocks, m_gputhreads, m_energy, m_momenta.data() );
   }
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   getMomentaFinalDevice( const fptype energy,
                          const fptype* rndmom,
@@ -171,11 +171,11 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   void
   RamboSamplingKernelDevice::getMomentaFinal()
   {
-    getMomentaFinalDevice<<<m_gpublocks, m_gputhreads>>>( m_energy, m_rndmom.data(), m_momenta.data(), m_weights.data() );
+    gpuLaunchKernel( getMomentaFinalDevice, m_gpublocks, m_gputhreads, m_energy, m_rndmom.data(), m_momenta.data(), m_weights.data() );
   }
 #endif
 
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/RamboSamplingKernels.h b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/RamboSamplingKernels.h
index 184089efd7..7c214cd74b 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/RamboSamplingKernels.h
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/RamboSamplingKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef RAMBOSAMPLINGKERNELS_H
 #define RAMBOSAMPLINGKERNELS_H 1
@@ -10,7 +10,7 @@
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -93,7 +93,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating RAMBO phase space sampling on a GPU device
   class RamboSamplingKernelDevice final : public SamplingKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/RandomNumberKernels.h b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/RandomNumberKernels.h
index 188a72c2c9..21d63beeac 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/RandomNumberKernels.h
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/RandomNumberKernels.h
@@ -1,14 +1,14 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef RANDOMNUMBERKERNELS_H
 #define RANDOMNUMBERKERNELS_H 1
 
 #include "mgOnGpuConfig.h"
 
-// NB This must come AFTER mgOnGpuConfig.h which contains our definition of __global__ when __CUDACC__ is not defined
+// NB This must come AFTER mgOnGpuConfig.h which contains our definition of __global__ when MGONGPUCPP_GPUIMPL is not defined
 #ifndef MGONGPU_HAS_NO_CURAND
 //#include "curand.h"
 struct curandGenerator_st; // forward definition from curand.h
@@ -16,7 +16,7 @@ struct curandGenerator_st; // forward definition from curand.h
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/cudacpp.mk b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/cudacpp.mk
index 509307506b..df74dfc284 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/cudacpp.mk
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/cudacpp.mk
@@ -1,7 +1,7 @@
 # Copyright (C) 2020-2023 CERN and UCLouvain.
 # Licensed under the GNU Lesser General Public License (version 3 or later).
 # Created by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-# Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+# Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #=== Determine the name of this makefile (https://ftp.gnu.org/old-gnu/Manuals/make-3.80/html_node/make_17.html)
 #=== NB: use ':=' to ensure that the value of CUDACPP_MAKEFILE is not modified further down after including make_opts
@@ -42,10 +42,10 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure common compiler flags for C++ and CUDA
+#=== Configure common compiler flags for C++ and CUDA/HIP
 
 INCFLAGS = -I.
-OPTFLAGS = -O3 # this ends up in CUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
+OPTFLAGS = -O3 # this ends up in GPUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
 
 # Dependency on src directory
 MG5AMC_COMMONLIB = mg5amc_common
@@ -121,24 +121,45 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure the CUDA compiler
+#=== Configure the GPU compiler (CUDA or HIP)
+
+# FIXME! (AV 24.01.2024)
+# In the current implementation (without separate builds for C++ and CUDA/HIP), we first check for cudacc and hipcc in CUDA_HOME and HIP_HOME.
+# If CUDA_HOME or HIP_HOME are not set, try to determine them from the path to cudacc and hipcc.
+# While convoluted, this is currently necessary to allow disabling CUDA/HIP builds by setting CUDA_HOME or HIP_HOME to invalid paths.
+# This will (probably?) be fixed when separate C++ and CUDA/HIP builds are implemented (PR #775).
 
 # If CXX is not a single word (example "clang++ --gcc-toolchain...") then disable CUDA builds (issue #505)
-# This is because it is impossible to pass this to "CUFLAGS += -ccbin <host-compiler>" below
+# This is because it is impossible to pass this to "GPUFLAGS += -ccbin <host-compiler>" below
 ifneq ($(words $(subst ccache ,,$(CXX))),1) # allow at most "CXX=ccache <host-compiler>" from outside
   $(warning CUDA builds are not supported for multi-word CXX "$(CXX)")
   override CUDA_HOME=disabled
 endif
 
-# If CUDA_HOME is not set, try to set it from the location of nvcc
+# If CUDA_HOME is not set, try to set it from the path to nvcc
 ifndef CUDA_HOME
-  CUDA_HOME = $(patsubst %bin/nvcc,%,$(shell which nvcc 2>/dev/null))
+  CUDA_HOME = $(patsubst %/bin/nvcc,%,$(shell which nvcc 2>/dev/null))
   $(warning CUDA_HOME was not set: using "$(CUDA_HOME)")
 endif
 
-# Set NVCC as $(CUDA_HOME)/bin/nvcc if it exists
+# If HIP_HOME is not set, try to set it from the path to hipcc
+ifndef HIP_HOME
+  HIP_HOME = $(patsubst %/bin/hipcc,%,$(shell which hipcc 2>/dev/null))
+  $(warning HIP_HOME was not set: using "$(HIP_HOME)")
+endif
+
+# FIXME! (AV 24.01.2024)
+# In the current implementation (without separate builds for C++ and CUDA/HIP),
+# builds are performed for HIP only if CUDA is not found in the path.
+# If both CUDA and HIP are installed, HIP builds can be triggered by unsetting CUDA_HOME.
+# This will be fixed when separate C++ and CUDA/HIP builds are implemented (PR #775).
+
+#--- Option 1: CUDA exists -> use CUDA
+
+# Set GPUCC as $(CUDA_HOME)/bin/nvcc if it exists
 ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
-  NVCC = $(CUDA_HOME)/bin/nvcc
+
+  GPUCC = $(CUDA_HOME)/bin/nvcc
   USE_NVTX ?=-DUSE_NVTX
   # See https://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/index.html
   # See https://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/
@@ -148,7 +169,7 @@ ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
   # Examples: use 60 for P100 (Piz Daint), 80 for A100 (Juwels Booster, NVidia raplab/Curiosity).
   MADGRAPH_CUDA_ARCHITECTURE ?= 70
   ###CUARCHFLAGS = -gencode arch=compute_$(MADGRAPH_CUDA_ARCHITECTURE),code=compute_$(MADGRAPH_CUDA_ARCHITECTURE) -gencode arch=compute_$(MADGRAPH_CUDA_ARCHITECTURE),code=sm_$(MADGRAPH_CUDA_ARCHITECTURE) # Older implementation (AV): go back to this one for multi-GPU support #533
-  ###CUARCHFLAGS = --gpu-architecture=compute_$(MADGRAPH_CUDA_ARCHITECTURE) --gpu-code=sm_$(MADGRAPH_CUDA_ARCHITECTURE),compute_$(MADGRAPH_CUDA_ARCHITECTURE) # Newer implementation (SH): cannot use this as-is for multi-GPU support #533
+  ###CUARCHFLAGS = --gpu-architecture=compute_$(MADGRAPH_CUDA_ARCHITECTURE) --gpu-code=sm_$(MADGRAPH_CUDA_ARCHITECTURE),compute_$(MADGRAPH_CUDA_ARCHITECTURE)  # Newer implementation (SH): cannot use this as-is for multi-GPU support #533
   comma:=,
   CUARCHFLAGS = $(foreach arch,$(subst $(comma), ,$(MADGRAPH_CUDA_ARCHITECTURE)),-gencode arch=compute_$(arch),code=compute_$(arch) -gencode arch=compute_$(arch),code=sm_$(arch))
   CUINC = -I$(CUDA_HOME)/include/
@@ -158,41 +179,79 @@ ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
     CURANDLIBFLAGS = -L$(CUDA_HOME)/lib64/ -lcurand # NB: -lcuda is not needed here!
   endif
   CUOPTFLAGS = -lineinfo
-  CUFLAGS = $(foreach opt, $(OPTFLAGS), -Xcompiler $(opt)) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
-  ###CUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
-  ###NVCC_VERSION = $(shell $(NVCC) --version | grep 'Cuda compilation tools' | cut -d' ' -f5 | cut -d, -f1)
-  CUFLAGS += -std=c++17 # need CUDA >= 11.2 (see #333): this is enforced in mgOnGpuConfig.h
+  ###GPUFLAGS = $(OPTFLAGS) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
+  GPUFLAGS = $(foreach opt, $(OPTFLAGS), -Xcompiler $(opt)) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
+  ###GPUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
+  ###GPUCC_VERSION = $(shell $(GPUCC) --version | grep 'Cuda compilation tools' | cut -d' ' -f5 | cut -d, -f1)
+  GPUFLAGS += -std=c++17 # need CUDA >= 11.2 (see #333): this is enforced in mgOnGpuConfig.h
   # Without -maxrregcount: baseline throughput: 6.5E8 (16384 32 12) up to 7.3E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 160 # improves throughput: 6.9E8 (16384 32 12) up to 7.7E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 128 # improves throughput: 7.3E8 (16384 32 12) up to 7.6E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 96 # degrades throughput: 4.1E8 (16384 32 12) up to 4.5E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 64 # degrades throughput: 1.7E8 (16384 32 12) flat at 1.7E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 160 # improves throughput: 6.9E8 (16384 32 12) up to 7.7E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 128 # improves throughput: 7.3E8 (16384 32 12) up to 7.6E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 96 # degrades throughput: 4.1E8 (16384 32 12) up to 4.5E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 64 # degrades throughput: 1.7E8 (16384 32 12) flat at 1.7E8 (65536 128 12)
+  CUBUILDRULEFLAGS = -Xcompiler -fPIC -c
+  CCBUILDRULEFLAGS = -Xcompiler -fPIC -c -x cu
+  CUDATESTFLAGS = -lcuda
+
+  # Set the host C++ compiler for GPUCC via "-ccbin <host-compiler>"
+  # (NB issue #505: this must be a single word, "clang++ --gcc-toolchain..." is not supported)
+  GPUFLAGS += -ccbin $(shell which $(subst ccache ,,$(CXX)))
+
+  # Allow newer (unsupported) C++ compilers with older versions of CUDA if ALLOW_UNSUPPORTED_COMPILER_IN_CUDA is set (#504)
+  ifneq ($(origin ALLOW_UNSUPPORTED_COMPILER_IN_CUDA),undefined)
+  GPUFLAGS += -allow-unsupported-compiler
+  endif
+
 else ifneq ($(origin REQUIRE_CUDA),undefined)
+
   # If REQUIRE_CUDA is set but no cuda is found, stop here (e.g. for CI tests on GPU #443)
-  $(error No cuda installation found (set CUDA_HOME or make nvcc visible in PATH))
+  $(error No cuda installation found (set CUDA_HOME or make GPUCC visible in PATH))
+
+#--- Option 2: CUDA does not exist, HIP exists -> use HIP
+
+# Set GPUCC as $(HIP_HOME)/bin/hipcc if it exists
+else ifneq ($(wildcard $(HIP_HOME)/bin/hipcc),)
+
+  GPUCC = $(HIP_HOME)/bin/hipcc
+  #USE_NVTX ?=-DUSE_NVTX # should maybe find something equivalent to this in HIP?
+  HIPARCHFLAGS = -target x86_64-linux-gnu --offload-arch=gfx90a
+  HIPINC = -I$(HIP_HOME)/include/
+  # Note: -DHIP_FAST_MATH is equivalent to -use_fast_math in HIP 
+  # (but only for single precision line 208: https://rocm-developer-tools.github.io/HIP/hcc__detail_2math__functions_8h_source.html)
+  # Note: CUOPTFLAGS should not be used for HIP, it had been added here but was then removed (#808)
+  GPUFLAGS = $(OPTFLAGS) $(INCFLAGS) $(HIPINC) $(HIPARCHFLAGS) -DHIP_FAST_MATH -DHIP_PLATFORM=amd -fPIC
+  ###GPUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
+  GPUFLAGS += -std=c++17
+  ###GPUFLAGS+= --maxrregcount 255 # (AV: is this option valid on HIP and meaningful on AMD GPUs?)
+  CUBUILDRULEFLAGS = -fPIC -c
+  CCBUILDRULEFLAGS = -fPIC -c -x hip
+
+else ifneq ($(origin REQUIRE_HIP),undefined)
+
+  # If REQUIRE_HIP is set but no HIP is found, stop here (e.g. for CI tests on GPU #443)
+  $(error No hip installation found (set HIP_HOME or make GPUCC visible in PATH))
+
+#--- Option 3: CUDA does not exist, HIP does not exist -> switch off both CUDA and HIP
+
 else
-  # No cuda. Switch cuda compilation off and go to common random numbers in C++
+
+  # No cudacc and no hipcc: switch CUDA and HIP compilation off and go to common random numbers in C++
   $(warning CUDA_HOME is not set or is invalid: export CUDA_HOME to compile with cuda)
-  override NVCC=
+  $(warning HIP_HOME is not set or is invalid: export HIP_HOME to compile with hip)
+  override GPUCC=
   override USE_NVTX=
   override CUINC=
   override CURANDLIBFLAGS=
-endif
-export NVCC
-export CUFLAGS
 
-# Set the host C++ compiler for nvcc via "-ccbin <host-compiler>"
-# (NB issue #505: this must be a single word, "clang++ --gcc-toolchain..." is not supported)
-CUFLAGS += -ccbin $(shell which $(subst ccache ,,$(CXX)))
-
-# Allow newer (unsupported) C++ compilers with older versions of CUDA if ALLOW_UNSUPPORTED_COMPILER_IN_CUDA is set (#504)
-ifneq ($(origin ALLOW_UNSUPPORTED_COMPILER_IN_CUDA),undefined)
-CUFLAGS += -allow-unsupported-compiler
 endif
 
+# Export GPUCC (so that it can also be used in cudacpp_src.mk?)
+export GPUCC
+export GPUFLAGS
+
 #-------------------------------------------------------------------------------
 
-#=== Configure ccache for C++ and CUDA builds
+#=== Configure ccache for C++ and CUDA/HIP builds
 
 # Enable ccache if USECCACHE=1
 ifeq ($(USECCACHE)$(shell echo $(CXX) | grep ccache),1)
@@ -201,15 +260,15 @@ endif
 #ifeq ($(USECCACHE)$(shell echo $(AR) | grep ccache),1)
 #  override AR:=ccache $(AR)
 #endif
-ifneq ($(NVCC),)
-  ifeq ($(USECCACHE)$(shell echo $(NVCC) | grep ccache),1)
-    override NVCC:=ccache $(NVCC)
+ifneq ($(GPUCC),)
+  ifeq ($(USECCACHE)$(shell echo $(GPUCC) | grep ccache),1)
+    override GPUCC:=ccache $(GPUCC)
   endif
 endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure PowerPC-specific compiler flags for C++ and CUDA
+#=== Configure PowerPC-specific compiler flags for C++ and CUDA/HIP
 
 # PowerPC-specific CXX compiler flags (being reviewed)
 ifeq ($(UNAME_P),ppc64le)
@@ -225,9 +284,9 @@ else
   ######CXXFLAGS+= -fno-semantic-interposition # no benefit (neither alone, nor combined with -flto)
 endif
 
-# PowerPC-specific CUDA compiler flags (to be reviewed!)
+# PowerPC-specific CUDA/HIP compiler flags (to be reviewed!)
 ifeq ($(UNAME_P),ppc64le)
-  CUFLAGS+= -Xcompiler -mno-float128
+  GPUFLAGS+= -Xcompiler -mno-float128
 endif
 
 #-------------------------------------------------------------------------------
@@ -235,9 +294,11 @@ endif
 #=== Configure defaults and check if user-defined choices exist for OMPFLAGS, AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
 
 # Set the default OMPFLAGS choice
-ifneq ($(shell $(CXX) --version | egrep '^Intel'),)
+ifneq ($(findstring hipcc,$(GPUCC)),)
+override OMPFLAGS = # disable OpenMP MT when using hipcc #802
+else ifneq ($(shell $(CXX) --version | egrep '^Intel'),)
 override OMPFLAGS = -fopenmp
-###override OMPFLAGS = # disable OpenMP MT on Intel (was ok without nvcc but not ok with nvcc before #578)
+###override OMPFLAGS = # disable OpenMP MT on Intel (was ok without GPUCC but not ok with GPUCC before #578)
 else ifneq ($(shell $(CXX) --version | egrep '^(clang)'),)
 override OMPFLAGS = -fopenmp
 ###override OMPFLAGS = # disable OpenMP MT on clang (was not ok without or with nvcc before #578)
@@ -293,7 +354,10 @@ endif
 
 # Set the default RNDGEN (random number generator) choice
 ifeq ($(RNDGEN),)
-  ifeq ($(NVCC),)
+  ifeq ($(GPUCC),)
+    override RNDGEN = hasNoCurand
+  # Edgecase for HIP compilation
+  else ifeq ($(findstring hipcc,$(GPUCC)),hipcc)
     override RNDGEN = hasNoCurand
   else ifeq ($(RNDGEN),)
     override RNDGEN = hasCurand
@@ -310,7 +374,7 @@ export OMPFLAGS
 
 #-------------------------------------------------------------------------------
 
-#=== Set the CUDA/C++ compiler flags appropriate to user-defined choices of AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
+#=== Set the CUDA/HIP/C++ compiler flags appropriate to user-defined choices of AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
 
 # Set the build flags appropriate to OMPFLAGS
 $(info OMPFLAGS=$(OMPFLAGS))
@@ -368,13 +432,13 @@ CXXFLAGS+= $(AVXFLAGS)
 $(info FPTYPE=$(FPTYPE))
 ifeq ($(FPTYPE),d)
   CXXFLAGS += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
-  CUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
+  GPUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
 else ifeq ($(FPTYPE),f)
   CXXFLAGS += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
-  CUFLAGS  += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
+  GPUFLAGS  += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
 else ifeq ($(FPTYPE),m)
   CXXFLAGS += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
-  CUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
+  GPUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
 else
   $(error Unknown FPTYPE='$(FPTYPE)': only 'd', 'f' and 'm' are supported)
 endif
@@ -383,7 +447,7 @@ endif
 $(info HELINL=$(HELINL))
 ifeq ($(HELINL),1)
   CXXFLAGS += -DMGONGPU_INLINE_HELAMPS
-  CUFLAGS  += -DMGONGPU_INLINE_HELAMPS
+  GPUFLAGS  += -DMGONGPU_INLINE_HELAMPS
 else ifneq ($(HELINL),0)
   $(error Unknown HELINL='$(HELINL)': only '0' and '1' are supported)
 endif
@@ -392,7 +456,7 @@ endif
 $(info HRDCOD=$(HRDCOD))
 ifeq ($(HRDCOD),1)
   CXXFLAGS += -DMGONGPU_HARDCODE_PARAM
-  CUFLAGS  += -DMGONGPU_HARDCODE_PARAM
+  GPUFLAGS  += -DMGONGPU_HARDCODE_PARAM
 else ifneq ($(HRDCOD),0)
   $(error Unknown HRDCOD='$(HRDCOD)': only '0' and '1' are supported)
 endif
@@ -444,11 +508,11 @@ ifeq ($(UNAME_S),Darwin)
   override CULIBFLAGSRPATH2 =
 else
   # RPATH to cuda/cpp libs when linking executables
-  override CXXLIBFLAGSRPATH = -Wl,-rpath,$(LIBDIRRPATH)
-  override CULIBFLAGSRPATH = -Xlinker -rpath,$(LIBDIRRPATH)
+  override CXXLIBFLAGSRPATH = -Wl,-rpath=$(LIBDIRRPATH)
+  override CULIBFLAGSRPATH = -Xlinker -rpath=$(LIBDIRRPATH)
   # RPATH to common lib when linking cuda/cpp libs
-  override CXXLIBFLAGSRPATH2 = -Wl,-rpath,'$$ORIGIN'
-  override CULIBFLAGSRPATH2 = -Xlinker -rpath,'$$ORIGIN'
+  override CXXLIBFLAGSRPATH2 = -Wl,-rpath='$$ORIGIN'
+  override CULIBFLAGSRPATH2 = -Xlinker -rpath='$$ORIGIN'
 endif
 
 # Setting LD_LIBRARY_PATH or DYLD_LIBRARY_PATH in the RUNTIME is no longer necessary (neither on Linux nor on Mac)
@@ -461,7 +525,7 @@ override RUNTIME =
 cxx_main=$(BUILDDIR)/check.exe
 fcxx_main=$(BUILDDIR)/fcheck.exe
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 cu_main=$(BUILDDIR)/gcheck.exe
 fcu_main=$(BUILDDIR)/fgcheck.exe
 else
@@ -492,14 +556,15 @@ $(BUILDDIR)/.build.$(TAG):
 	@touch $(BUILDDIR)/.build.$(TAG)
 
 # Generic target and build rules: objects from CUDA compilation
-ifneq ($(NVCC),)
+# NB: CCBUILDRULEFLAGS includes "-x cu" for nvcc and "-x hip" for hipcc (#810)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/%.o : %.cu *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $(CUBUILDRULEFLAGS) $< -o $@
 
 $(BUILDDIR)/%_cu.o : %.cc *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c -x cu $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $(CCBUILDRULEFLAGS) $< -o $@
 endif
 
 # Generic target and build rules: objects from C++ compilation
@@ -509,11 +574,14 @@ $(BUILDDIR)/%.o : %.cc *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	$(CXX) $(CPPFLAGS) $(CXXFLAGS) -fPIC -c $< -o $@
 
 # Apply special build flags only to CrossSectionKernel.cc and gCrossSectionKernel.cu (no fast math, see #117 and #516)
+# Added edgecase for HIP compilation
 ifeq ($(shell $(CXX) --version | grep ^nvc++),)
 $(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS := $(filter-out -ffast-math,$(CXXFLAGS))
 $(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS += -fno-fast-math
-ifneq ($(NVCC),)
-$(BUILDDIR)/gCrossSectionKernels.o: CUFLAGS += -Xcompiler -fno-fast-math
+ifeq ($(findstring nvcc,$(GPUCC)),nvcc)
+  $(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -Xcompiler -fno-fast-math
+else
+  $(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -fno-fast-math
 endif
 endif
 
@@ -530,10 +598,10 @@ ifeq ($(RNDGEN),hasCurand)
 $(BUILDDIR)/CurandRandomNumberKernel.o: CXXFLAGS += $(CUINC)
 endif
 
-# Avoid "warning: builtin __has_trivial_... is deprecated; use __is_trivially_... instead" in nvcc with icx2023 (#592)
+# Avoid "warning: builtin __has_trivial_... is deprecated; use __is_trivially_... instead" in GPUCC with icx2023 (#592)
 ifneq ($(shell $(CXX) --version | egrep '^(Intel)'),)
-ifneq ($(NVCC),)
-CUFLAGS += -Xcompiler -Wno-deprecated-builtins
+ifneq ($(GPUCC),)
+GPUFLAGS += -Wno-deprecated-builtins
 endif
 endif
 
@@ -541,8 +609,8 @@ endif
 # This patch does remove the warning, but I prefer to keep it disabled for the moment...
 ###ifneq ($(shell $(CXX) --version | egrep '^(clang|Apple clang|Intel)'),)
 ###$(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS += -Wno-overriding-t-option
-###ifneq ($(NVCC),)
-###$(BUILDDIR)/gCrossSectionKernels.o: CUFLAGS += -Xcompiler -Wno-overriding-t-option
+###ifneq ($(GPUCC),)
+###$(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -Xcompiler -Wno-overriding-t-option
 ###endif
 ###endif
 
@@ -569,7 +637,7 @@ MG5AMC_CXXLIB = mg5amc_$(processid_short)_cpp
 cxx_objects_lib=$(BUILDDIR)/CPPProcess.o $(BUILDDIR)/MatrixElementKernels.o $(BUILDDIR)/BridgeKernels.o $(BUILDDIR)/CrossSectionKernels.o
 cxx_objects_exe=$(BUILDDIR)/CommonRandomNumberKernel.o $(BUILDDIR)/RamboSamplingKernels.o
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 MG5AMC_CULIB = mg5amc_$(processid_short)_cuda
 cu_objects_lib=$(BUILDDIR)/gCPPProcess.o $(BUILDDIR)/gMatrixElementKernels.o $(BUILDDIR)/gBridgeKernels.o $(BUILDDIR)/gCrossSectionKernels.o
 cu_objects_exe=$(BUILDDIR)/gCommonRandomNumberKernel.o $(BUILDDIR)/gRamboSamplingKernels.o
@@ -581,11 +649,17 @@ $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: cxx_objects_lib += $(BUILDDIR)/fbridge.o
 $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib)
 	$(CXX) -shared -o $@ $(cxx_objects_lib) $(CXXLIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: $(BUILDDIR)/fbridge_cu.o
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: cu_objects_lib += $(BUILDDIR)/fbridge_cu.o
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cu_objects_lib)
-	$(NVCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+# Bypass std::filesystem completely to ease portability on LUMI #803
+#ifneq ($(findstring hipcc,$(GPUCC)),)
+#	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB) -lstdc++fs
+#else
+#	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+#endif
 endif
 
 #-------------------------------------------------------------------------------
@@ -602,16 +676,16 @@ $(cxx_main): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PAT
 $(cxx_main): $(BUILDDIR)/check_sa.o $(LIBDIR)/lib$(MG5AMC_CXXLIB).so $(cxx_objects_exe) $(BUILDDIR)/CurandRandomNumberKernel.o
 	$(CXX) -o $@ $(BUILDDIR)/check_sa.o $(OMPFLAGS) -ldl -pthread $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe) $(BUILDDIR)/CurandRandomNumberKernel.o $(CURANDLIBFLAGS)
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(cu_main): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(cu_main): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(cu_main): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(cu_main): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 else ifneq ($(shell $(CXX) --version | grep ^nvc++),) # support nvc++ #531
 $(cu_main): LIBFLAGS += -L$(patsubst %bin/nvc++,%lib,$(subst ccache ,,$(CXX))) -lnvhpcatm -lnvcpumath -lnvc
 endif
 $(cu_main): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(cu_main): $(BUILDDIR)/gcheck_sa.o $(LIBDIR)/lib$(MG5AMC_CULIB).so $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o
-	$(NVCC) -o $@ $(BUILDDIR)/gcheck_sa.o $(CUARCHFLAGS) $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o $(CURANDLIBFLAGS)
+	$(GPUCC) -o $@ $(BUILDDIR)/gcheck_sa.o $(CUARCHFLAGS) $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o $(CURANDLIBFLAGS)
 endif
 
 #-------------------------------------------------------------------------------
@@ -635,19 +709,27 @@ $(fcxx_main): LIBFLAGS += -L$(shell dirname $(shell $(FC) --print-file-name libg
 endif
 $(fcxx_main): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(fcxx_main): $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler.o $(LIBDIR)/lib$(MG5AMC_CXXLIB).so $(cxx_objects_exe)
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(BUILDDIR)/fcheck_sa.o $(OMPFLAGS) $(BUILDDIR)/fsampler.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe) -lstdc++
+else
 	$(CXX) -o $@ $(BUILDDIR)/fcheck_sa.o $(OMPFLAGS) $(BUILDDIR)/fsampler.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe)
+endif
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(fcu_main): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(fcu_main): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(fcu_main): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(fcu_main): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 endif
 ifeq ($(UNAME_S),Darwin)
 $(fcu_main): LIBFLAGS += -L$(shell dirname $(shell $(FC) --print-file-name libgfortran.dylib)) # add path to libgfortran on Mac #375
 endif
 $(fcu_main): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(fcu_main): $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBDIR)/lib$(MG5AMC_CULIB).so $(cu_objects_exe)
-	$(NVCC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe)
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) -lstdc++ -L$(shell dirname $(shell $(GPUCC) -print-prog-name=clang))/../../lib -lamdhip64
+else
+	$(GPUCC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe)
+endif
 endif
 
 #-------------------------------------------------------------------------------
@@ -659,7 +741,7 @@ $(BUILDDIR)/testxxx.o: testxxx_cc_ref.txt
 $(testmain): $(BUILDDIR)/testxxx.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/testxxx.o # Comment out this line to skip the C++ test of xxx functions
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/testxxx_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/testxxx_cu.o: INCFLAGS += $(GTESTINC)
 $(BUILDDIR)/testxxx_cu.o: testxxx_cc_ref.txt
@@ -672,7 +754,7 @@ $(BUILDDIR)/testmisc.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/testmisc.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/testmisc.o # Comment out this line to skip the C++ miscellaneous tests
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/testmisc_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/testmisc_cu.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/testmisc_cu.o
@@ -684,12 +766,12 @@ $(BUILDDIR)/runTest.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/runTest.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/runTest.o
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/runTest_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/runTest_cu.o: INCFLAGS += $(GTESTINC)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(testmain): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(testmain): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(testmain): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(testmain): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 else ifneq ($(shell $(CXX) --version | grep ^nvc++),) # support nvc++ #531
 $(testmain): LIBFLAGS += -L$(patsubst %bin/nvc++,%lib,$(subst ccache ,,$(CXX))) -lnvhpcatm -lnvcpumath -lnvc
 endif
@@ -713,14 +795,23 @@ $(testmain): LIBFLAGS += -lgomp
 endif
 endif
 
-ifeq ($(NVCC),) # link only runTest.o
+# Bypass std::filesystem completely to ease portability on LUMI #803
+#ifneq ($(findstring hipcc,$(GPUCC)),)
+#$(testmain): LIBFLAGS += -lstdc++fs
+#endif
+
+ifeq ($(GPUCC),) # link only runTest.o
 $(testmain): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(testmain): $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib) $(cxx_objects_exe) $(GTESTLIBS)
 	$(CXX) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) -ldl -pthread $(LIBFLAGS)
 else # link both runTest.o and runTest_cu.o
 $(testmain): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(testmain): $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) $(GTESTLIBS)
-	$(NVCC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) -lcuda
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) $(CUDATESTFLAGS) -lstdc++ -lpthread  -L$(shell dirname $(shell $(GPUCC) -print-prog-name=clang))/../../lib -lamdhip64
+else
+	$(GPUCC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) $(CUDATESTFLAGS)
+endif
 endif
 
 # Use target gtestlibs to build only googletest
@@ -829,9 +920,9 @@ ifeq ($(USECCACHE),1)
 	ccache --version | head -1
 endif
 	@echo ""
-	@echo NVCC=$(NVCC)
-ifneq ($(NVCC),)
-	$(NVCC) --version
+	@echo GPUCC=$(GPUCC)
+ifneq ($(GPUCC),)
+	$(GPUCC) --version
 endif
 	@echo ""
 	@echo CXX=$(CXX)
@@ -850,7 +941,7 @@ endif
 
 # Target: check (run the C++ test executable)
 # [NB THIS IS WHAT IS USED IN THE GITHUB CI!]
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 check: runTest cmpFcheck cmpFGcheck
 else
 check: runTest cmpFcheck
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/fbridge.cc b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/fbridge.cc
index 2d2b36d560..22ce3f5115 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/fbridge.cc
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/fbridge.cc
@@ -1,11 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Roiser (Oct 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "Bridge.h"
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 
 extern "C"
 {
@@ -22,7 +22,7 @@ extern "C"
    * Using the same Fortran MadEvent code, linking to the hetrerogeneous library would allow access to both CPU and GPU implementations.
    * The specific heterogeneous configuration (how many GPUs, how many threads on each CPU, etc) could be loaded in CUDA/C++ from a data file.
    */
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -46,8 +46,8 @@ extern "C"
    */
   void fbridgecreate_( CppObjectInFortran** ppbridge, const int* pnevtF, const int* pnparF, const int* pnp4F )
   {
-#ifdef __CUDACC__
-    CudaRuntime::setUp();
+#ifdef MGONGPUCPP_GPUIMPL
+    GpuRuntime::setUp();
 #endif
     // (NB: CPPProcess::initProc no longer needs to be executed here because it is called in the Bridge constructor)
     // FIXME: disable OMP in Bridge when called from Fortran
@@ -65,8 +65,8 @@ extern "C"
     Bridge<FORTRANFPTYPE>* pbridge = dynamic_cast<Bridge<FORTRANFPTYPE>*>( *ppbridge );
     if( pbridge == 0 ) throw std::runtime_error( "fbridgedelete_: invalid Bridge address" );
     delete pbridge;
-#ifdef __CUDACC__
-    CudaRuntime::tearDown();
+#ifdef MGONGPUCPP_GPUIMPL
+    GpuRuntime::tearDown();
 #endif
   }
 
@@ -96,7 +96,7 @@ extern "C"
   {
     Bridge<FORTRANFPTYPE>* pbridge = dynamic_cast<Bridge<FORTRANFPTYPE>*>( *ppbridge );
     if( pbridge == 0 ) throw std::runtime_error( "fbridgesequence_: invalid Bridge address" );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Use the device/GPU implementation in the CUDA library
     // (there is also a host implementation in this library)
     pbridge->gpu_sequence( momenta, gs, rndhel, rndcol, *pchannelId, mes, selhel, selcol );
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/fsampler.cc b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/fsampler.cc
index 2fb445372d..3743934f41 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/fsampler.cc
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/fsampler.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Feb 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "mgOnGpuConfig.h"
 
@@ -13,7 +13,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -40,7 +40,7 @@ namespace mg5amcCpu
   private:
     const int m_nevt; // The number of events in each iteration
     int m_iiter;      // The iteration counter (for random number seeding)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     HostBufferRndNumMomenta m_hstRndmom; // Memory buffers for random numbers
     HostBufferMomenta m_hstMomenta;      // Memory buffers for momenta
     HostBufferWeights m_hstWeights;      // Memory buffers for sampling weights
@@ -105,7 +105,7 @@ namespace mg5amcCpu
 
 extern "C"
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/runTest.cc b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/runTest.cc
index d4a760a71b..de327f2321 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/runTest.cc
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/runTest.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Hageboeck (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*xxx to run only testxxx.cc tests
 //----------------------------------------------------------------------------
@@ -18,7 +18,7 @@
 #include "RandomNumberKernels.h"
 #include "epoch_process_id.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using namespace mg5amcGpu;
 #else
 using namespace mg5amcCpu;
@@ -35,7 +35,7 @@ struct CUDA_CPU_TestBase : public TestDriverBase
     : TestDriverBase( npar, refFileName ) {}
 };
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 struct CPUTest : public CUDA_CPU_TestBase
 {
   // Struct data members (process, and memory structures for random numbers, momenta, matrix elements and weights on host and device)
@@ -119,7 +119,7 @@ struct CPUTest : public CUDA_CPU_TestBase
 };
 #endif
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 struct CUDATest : public CUDA_CPU_TestBase
 {
   // Reset the device when our test goes out of scope. Note that this should happen after
@@ -128,7 +128,7 @@ struct CUDATest : public CUDA_CPU_TestBase
   {
     ~DeviceReset()
     {
-      checkCuda( cudaDeviceReset() ); // this is needed by cuda-memcheck --leak-check full
+      checkGpu( gpuDeviceReset() ); // this is needed by cuda-memcheck --leak-check full
     }
   } deviceResetter;
 
@@ -256,7 +256,7 @@ INSTANTIATE_TEST_SUITE_P( prefix, \
                           test_suite_name, \
                           testing::Values( new CUDATest( MG_EPOCH_REFERENCE_FILE_NAME ) ) );
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 MG_INSTANTIATE_TEST_SUITE_GPU( XTESTID_GPU( MG_EPOCH_PROCESS_ID ), MadgraphTest );
 #else
 MG_INSTANTIATE_TEST_SUITE_CPU( XTESTID_CPU( MG_EPOCH_PROCESS_ID ), MadgraphTest );
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/testmisc.cc b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/testmisc.cc
index 895d6eeb56..ba9e59a8a3 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/testmisc.cc
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/testmisc.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*misc to run only testmisc.cc tests
 //----------------------------------------------------------------------------
@@ -17,7 +17,7 @@
 #include <sstream>
 #include <typeinfo>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #define TESTID( s ) s##_GPU_MISC
 #else
 #define TESTID( s ) s##_CPU_MISC
@@ -26,7 +26,7 @@
 #define XTESTID( s ) TESTID( s )
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -59,7 +59,7 @@ namespace mg5amcCpu
 
 TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testmisc )
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/testxxx.cc b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/testxxx.cc
index a1c3cdc238..688cb8167b 100644
--- a/epochX/cudacpp/heft_gg_h.sa/SubProcesses/testxxx.cc
+++ b/epochX/cudacpp/heft_gg_h.sa/SubProcesses/testxxx.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Apr 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*xxx to run only testxxx.cc tests
 //----------------------------------------------------------------------------
@@ -24,7 +24,7 @@
 #include <iomanip>
 #include <iostream>
 #include <vector>
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #define TESTID( s ) s##_GPU_XXX
 #else
 #define TESTID( s ) s##_CPU_XXX
@@ -32,7 +32,7 @@
 
 #define XTESTID( s ) TESTID( s )
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -42,7 +42,7 @@ namespace mg5amcCpu
   int FPEhandlerIevt = -1;
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU): '" << FPEhandlerMessage << "' ievt=" << FPEhandlerIevt << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU neppV=" << neppV << "): '" << FPEhandlerMessage << "' ievt=" << FPEhandlerIevt << std::endl;
@@ -53,7 +53,7 @@ namespace mg5amcCpu
 
 TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -77,7 +77,7 @@ TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
   assert( nevt % neppM == 0 ); // nevt must be a multiple of neppM
   assert( nevt % neppV == 0 ); // nevt must be a multiple of neppV
   // Fill in the input momenta
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   mg5amcGpu::PinnedHostBufferMomenta hstMomenta( nevt ); // AOSOA[npagM][npar=4][np4=4][neppM]
 #else
   mg5amcCpu::HostBufferMomenta hstMomenta( nevt ); // AOSOA[npagM][npar=4][np4=4][neppM]
@@ -322,7 +322,7 @@ TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
   {
     for( int ievt = 0; ievt < nevt; ievt++ )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       using namespace mg5amcGpu;
 #else
       using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/heft_gg_h.sa/src/HelAmps_heft.h b/epochX/cudacpp/heft_gg_h.sa/src/HelAmps_heft.h
index eae9ff5242..dbff117235 100644
--- a/epochX/cudacpp/heft_gg_h.sa/src/HelAmps_heft.h
+++ b/epochX/cudacpp/heft_gg_h.sa/src/HelAmps_heft.h
@@ -5,7 +5,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -28,7 +28,7 @@
 //#include <iomanip>
 //#include <iostream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/heft_gg_h.sa/src/Parameters_heft.cc b/epochX/cudacpp/heft_gg_h.sa/src/Parameters_heft.cc
index e5442756b1..d3d6058b46 100644
--- a/epochX/cudacpp/heft_gg_h.sa/src/Parameters_heft.cc
+++ b/epochX/cudacpp/heft_gg_h.sa/src/Parameters_heft.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -17,7 +17,7 @@
 #include <iomanip>
 #include <iostream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using namespace mg5amcGpu;
 #else
 using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/heft_gg_h.sa/src/Parameters_heft.h b/epochX/cudacpp/heft_gg_h.sa/src/Parameters_heft.h
index 790485fee0..c2be5bba97 100644
--- a/epochX/cudacpp/heft_gg_h.sa/src/Parameters_heft.h
+++ b/epochX/cudacpp/heft_gg_h.sa/src/Parameters_heft.h
@@ -28,7 +28,7 @@
 #include "read_slha.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -94,7 +94,7 @@ namespace mg5amcCpu
 #include <limits>
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -230,7 +230,7 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -247,7 +247,7 @@ namespace mg5amcCpu
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wunused-variable"  // e.g. <<warning: unused variable ‘mdl_G__exp__2’ [-Wunused-variable]>>
 #pragma GCC diagnostic ignored "-Wunused-parameter" // e.g. <<warning: unused parameter ‘G’ [-Wunused-parameter]>>
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #pragma nv_diagnostic push
 #pragma nv_diag_suppress 177 // e.g. <<warning #177-D: variable "mdl_G__exp__2" was declared but never referenced>>
 #endif
@@ -298,7 +298,7 @@ namespace mg5amcCpu
       // End non-SM (e.g. EFT) implementation - special handling of vectors of floats (#439)
       return out;
     }
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #pragma GCC diagnostic pop
 #pragma nv_diagnostic pop
 #endif
diff --git a/epochX/cudacpp/heft_gg_h.sa/src/cudacpp_src.mk b/epochX/cudacpp/heft_gg_h.sa/src/cudacpp_src.mk
index 0bd815c9b3..fb8da8830b 100644
--- a/epochX/cudacpp/heft_gg_h.sa/src/cudacpp_src.mk
+++ b/epochX/cudacpp/heft_gg_h.sa/src/cudacpp_src.mk
@@ -19,7 +19,7 @@ SHELL := /bin/bash
 #=== Configure common compiler flags for CUDA and C++
 
 INCFLAGS = -I.
-OPTFLAGS = -O3 # this ends up in CUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
+OPTFLAGS = -O3 # this ends up in GPUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
 
 #-------------------------------------------------------------------------------
 
@@ -45,13 +45,13 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure the CUDA compiler (note: NVCC is already exported including ccache)
+#=== Configure the CUDA compiler (note: GPUCC is already exported including ccache)
 
-###$(info NVCC=$(NVCC))
+###$(info GPUCC=$(GPUCC))
 
 #-------------------------------------------------------------------------------
 
-#=== Configure ccache for C++ builds (note: NVCC is already exported including ccache)
+#=== Configure ccache for C++ builds (note: GPUCC is already exported including ccache)
 
 # Enable ccache if USECCACHE=1
 ifeq ($(USECCACHE)$(shell echo $(CXX) | grep ccache),1)
@@ -92,6 +92,13 @@ endif
 ###$(info OMPFLAGS=$(OMPFLAGS))
 CXXFLAGS += $(OMPFLAGS)
 
+# Add correct flags for nvcc (-x cu) and hipcc (-x hip) for GPU code (see #810)
+ifeq ($(findstring nvcc,$(GPUCC)),nvcc)
+  GPUFLAGS += -Xcompiler -fPIC -c -x cu
+else ifeq ($(findstring hipcc,$(GPUCC)),hipcc)
+  GPUFLAGS += -fPIC -c -x hip
+endif
+
 # Set the build flags appropriate to each AVX choice (example: "make AVX=none")
 # [NB MGONGPU_PVW512 is needed because "-mprefer-vector-width=256" is not exposed in a macro]
 # [See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96476]
@@ -253,20 +260,20 @@ $(BUILDDIR)/%.o : %.cc *.h $(BUILDDIR)/.build.$(TAG)
 # Generic target and build rules: objects from CUDA compilation
 $(BUILDDIR)/%_cu.o : %.cc *.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c -x cu $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $< -o $@
 
 #-------------------------------------------------------------------------------
 
 cxx_objects=$(addprefix $(BUILDDIR)/, Parameters_heft.o read_slha.o)
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 cu_objects=$(addprefix $(BUILDDIR)/, Parameters_heft_cu.o)
 endif
 
 # Target (and build rules): common (src) library
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so : $(cxx_objects) $(cu_objects)
 	@if [ ! -d $(LIBDIR) ]; then echo "mkdir -p $(LIBDIR)"; mkdir -p $(LIBDIR); fi
-	$(NVCC) -shared -o $@ $(cxx_objects) $(cu_objects) $(LDFLAGS)
+	$(GPUCC) -shared -o $@ $(cxx_objects) $(cu_objects) $(LDFLAGS)
 else
 $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so : $(cxx_objects)
 	@if [ ! -d $(LIBDIR) ]; then echo "mkdir -p $(LIBDIR)"; mkdir -p $(LIBDIR); fi
diff --git a/epochX/cudacpp/heft_gg_h.sa/src/mgOnGpuConfig.h b/epochX/cudacpp/heft_gg_h.sa/src/mgOnGpuConfig.h
index b247654dcf..475749ca7c 100644
--- a/epochX/cudacpp/heft_gg_h.sa/src/mgOnGpuConfig.h
+++ b/epochX/cudacpp/heft_gg_h.sa/src/mgOnGpuConfig.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jul 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUCONFIG_H
 #define MGONGPUCONFIG_H 1
@@ -10,12 +10,26 @@
 // There are two different code bases for standalone_cudacpp (without multichannel) and madevent+cudacpp (with multichannel)
 #undef MGONGPU_SUPPORTS_MULTICHANNEL
 
+// Is this a GPU (CUDA, HIP) or CPU implementation?
+#ifdef __CUDACC__
+#define MGONGPUCPP_GPUIMPL cuda
+#elif defined __HIPCC__
+#define MGONGPUCPP_GPUIMPL hip
+#include "hip/hip_runtime.h" // needed for blockDim, blockIdx, threadIdx: better in mgOnGpuConfig.h than in GpuAbstraction.h
+#else
+#undef MGONGPUCPP_GPUIMPL
+#endif
+
 // ** NB1 Throughputs (e.g. 6.8E8) are events/sec for "./gcheck.exe -p 65536 128 12"
 // ** NB2 Baseline on b7g47n0004 fluctuates (probably depends on load on other VMs)
 
 // Choose if curand is supported for generating random numbers
+// For HIP, by default, do not use curand (common random numbers will be used instead)
 // For both CUDA and C++, by default, do not inline, but allow this macro to be set from outside with e.g. -DMGONGPU_HAS_NO_CURAND
-// (there exist CUDA installations, e.g. using the HPC package, which do not include curand - see PR #784)
+// (there exist CUDA installations, e.g. using the HPC package, which do not include curand - see PR #784 and #785)
+#if defined __HIPCC__
+#define MGONGPU_HAS_NO_CURAND 1
+#else
 //#ifdef __CUDACC__
 //#undef MGONGPU_HAS_NO_CURAND // default
 ////#define MGONGPU_HAS_NO_CURAND 1
@@ -23,6 +37,7 @@
 //#undef MGONGPU_HAS_NO_CURAND // default
 ////#define MGONGPU_HAS_NO_CURAND 1
 //#endif
+#endif
 
 // Choose floating point precision (for everything but color algebra #537)
 // If one of these macros has been set from outside with e.g. -DMGONGPU_FPTYPE_FLOAT, nothing happens (issue #167)
@@ -54,23 +69,31 @@
 //#undef MGONGPU_HARDCODE_PARAM // default
 ////#define MGONGPU_HARDCODE_PARAM 1
 
-// Complex type in c++: std::complex or cxsmpl (CHOOSE ONLY ONE)
-#ifndef __CUDACC__
-//#define MGONGPU_CPPCXTYPE_STDCOMPLEX 1 // ~8 percent slower on float, same on double (5.1E6/double, 9.4E6/float)
-#define MGONGPU_CPPCXTYPE_CXSMPL 1 // new default (5.1E6/double, 10.2E6/float)
-#endif
-
-// Complex type in cuda: thrust or cucomplex or cxsmpl (CHOOSE ONLY ONE)
+// Complex type in CUDA: thrust or cucomplex or cxsmpl (CHOOSE ONLY ONE)
+// (NB THIS IS MGONGPU_*CU*CXTYPE_xxx)
 #ifdef __CUDACC__
 #define MGONGPU_CUCXTYPE_THRUST 1 // default (~1.15E9/double, ~3.2E9/float)
 //#define MGONGPU_CUCXTYPE_CUCOMPLEX 1 // ~10 percent slower (1.03E9/double, ~2.8E9/float)
 //#define MGONGPU_CUCXTYPE_CXSMPL 1 // ~10 percent slower (1.00E9/double, ~2.9E9/float)
+
+// Complex type in HIP: cxsmpl (ONLY ONE OPTION POSSIBLE? #810)
+// (NB THIS IS MGONGPU_*HIP*CXTYPE_xxx)
+#elif defined __HIPCC__
+#define MGONGPU_HIPCXTYPE_CXSMPL 1 // default for HIP
+
+// Complex type in C++: std::complex or cxsmpl (CHOOSE ONLY ONE)
+// (NB THIS IS MGONGPU_*CPP*CXTYPE_xxx)
+#else
+//#define MGONGPU_CPPCXTYPE_STDCOMPLEX 1 // ~8 percent slower on float, same on double (5.1E6/double, 9.4E6/float)
+#define MGONGPU_CPPCXTYPE_CXSMPL 1 // new default (5.1E6/double, 10.2E6/float)
 #endif
 
-// Cuda nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
+// CUDA nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
 #ifdef __CUDACC__
-#undef MGONGPU_NSIGHT_DEBUG // default
+#undef MGONGPU_NSIGHT_DEBUG // default in CUDA
 //#define MGONGPU_NSIGHT_DEBUG 1
+#else
+#undef MGONGPU_NSIGHT_DEBUG // only option in HIP or C++
 #endif
 
 // SANITY CHECKS (floating point precision for everything but color algebra #537)
@@ -86,17 +109,21 @@
 #error You cannot use double precision for color algebra and single precision elsewhere
 #endif
 
-// SANITY CHECKS (c++ complex number implementation)
-#ifndef __CUDACC__
-#if defined MGONGPU_CPPCXTYPE_STDCOMPLEX and defined MGONGPU_CPPCXTYPE_CXSMPL
-#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CPPCXTYPE_STDCOMPLEX or MGONGPU_CPPCXTYPE_CXSMPL
+// SANITY CHECKS (CUDA complex number implementation)
+#ifdef __CUDACC__
+#if defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CUCOMPLEX
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX for CUDA
+#elif defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CXSMPL for CUDA
+#elif defined MGONGPU_CUCXTYPE_CUCOMPLEX and defined MGONGPU_CUCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE OF MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL for CUDA
 #endif
 #endif
 
-// SANITY CHECKS (cuda complex number implementation)
-#ifdef __CUDACC__
-#if defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CUCOMPLEX and defined MGONGPU_CUCXTYPE_CXSMPL
-#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL
+// SANITY CHECKS (C++ complex number implementation)
+#ifndef MGONGPUCPP_GPUIMPL
+#if defined MGONGPU_CPPCXTYPE_STDCOMPLEX and defined MGONGPU_CPPCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CPPCXTYPE_STDCOMPLEX or MGONGPU_CPPCXTYPE_CXSMPL for C++
 #endif
 #endif
 
@@ -134,7 +161,7 @@ namespace mgOnGpu
   // Alignment requirement for using reinterpret_cast with SIMD vectorized code
   // (using reinterpret_cast with non aligned memory may lead to segmentation faults!)
   // Only needed for C++ code but can be enforced also in NVCC builds of C++ code using CUDA>=11.2 and C++17 (#318, #319, #333)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   constexpr int cppAlign = 64; // alignment requirement for SIMD vectorization (64-byte i.e. 512-bit)
 #endif
 
@@ -145,7 +172,7 @@ using mgOnGpu::fptype;
 using mgOnGpu::fptype2;
 
 // C++ SIMD vectorization width (this will be used to set neppV)
-#ifdef __CUDACC__ // CUDA implementation has no SIMD
+#ifdef MGONGPUCPP_GPUIMPL // CUDA and HIP implementations have no SIMD
 #undef MGONGPU_CPPSIMD
 #elif defined __AVX512VL__ && defined MGONGPU_PVW512 // C++ "512z" AVX512 with 512 width (512-bit ie 64-byte): 8 (DOUBLE) or 16 (FLOAT)
 #ifdef MGONGPU_FPTYPE_DOUBLE
@@ -175,9 +202,9 @@ using mgOnGpu::fptype2;
 #undef MGONGPU_CPPSIMD
 #endif
 
-// Cuda nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
+// CUDA nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
 // Arguments (not used so far): text is __FUNCTION__, code is 0 (start) or 1 (end)
-#if defined __CUDACC__ && defined MGONGPU_NSIGHT_DEBUG /* clang-format off */
+#if defined __CUDA__ && defined MGONGPU_NSIGHT_DEBUG /* clang-format off */
 #define mgDebugDeclare() __shared__ float mgDebugCounter[mgOnGpu::ntpbMAX];
 #define mgDebugInitialise() { mgDebugCounter[threadIdx.x] = 0; }
 #define mgDebug( code, text ) { mgDebugCounter[threadIdx.x] += 1; }
@@ -189,8 +216,8 @@ using mgOnGpu::fptype2;
 #define mgDebugFinalise() { /*noop*/ }
 #endif /* clang-format on */
 
-// Define empty CUDA declaration specifiers for C++
-#ifndef __CUDACC__
+// Define empty CUDA/HIP declaration specifiers for C++
+#ifndef MGONGPUCPP_GPUIMPL
 #define __global__
 #define __host__
 #define __device__
diff --git a/epochX/cudacpp/heft_gg_h.sa/src/mgOnGpuCxtypes.h b/epochX/cudacpp/heft_gg_h.sa/src/mgOnGpuCxtypes.h
index ca9a9f00c0..7ede1dbfae 100644
--- a/epochX/cudacpp/heft_gg_h.sa/src/mgOnGpuCxtypes.h
+++ b/epochX/cudacpp/heft_gg_h.sa/src/mgOnGpuCxtypes.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022, based on earlier work by D. Smith) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUCXTYPES_H
 #define MGONGPUCXTYPES_H 1
@@ -19,7 +19,7 @@
 #include <complex>
 
 // Complex type in cuda: thrust or cucomplex or cxsmpl
-#ifdef __CUDACC__
+#ifdef __CUDACC__ // this must be __CUDAC__ (not MGONGPUCPP_GPUIMPL)
 #if defined MGONGPU_CUCXTYPE_THRUST
 #pragma clang diagnostic push
 #pragma clang diagnostic ignored "-Wtautological-compare" // for icpx2021/clang13 (https://stackoverflow.com/a/15864661)
@@ -30,8 +30,13 @@
 #elif not defined MGONGPU_CUCXTYPE_CXSMPL
 #error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL
 #endif
+// Complex type in HIP: cxsmpl
+#elif defined __HIPCC__
+#if not defined MGONGPU_HIPCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_HIPCXTYPE_CXSMPL
+#endif
 #else
-// Complex type in c++: std::complex or cxsmpl
+// Complex type in c++ or HIP: std::complex or cxsmpl
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
 #include <cmath>
 #elif not defined MGONGPU_CPPCXTYPE_CXSMPL
@@ -82,7 +87,7 @@ namespace mgOnGpu /* clang-format off */
 using mgOnGpu::cxsmpl;
 
 // Printout to stream for user defined types
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -92,7 +97,7 @@ namespace mg5amcCpu
   inline __host__ std::ostream&
   operator<<( std::ostream& out, const cxsmpl<FP>& c )
   {
-    out << std::complex( c.real(), c.imag() );
+    out << std::complex<FP>( c.real(), c.imag() );
     return out;
   }
 
@@ -215,14 +220,14 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
   // --- Type definitions (complex type: cxtype)
-#ifdef __CUDACC__ // cuda
+#ifdef __CUDACC__ // this must be __CUDAC__ (not MGONGPUCPP_GPUIMPL)
 #if defined MGONGPU_CUCXTYPE_THRUST
   typedef thrust::complex<fptype> cxtype;
 #elif defined MGONGPU_CUCXTYPE_CUCOMPLEX
@@ -255,13 +260,13 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
-#if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+#if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
 
   //------------------------------
   // CUDA or C++ - using cxsmpl
@@ -303,11 +308,11 @@ namespace mg5amcCpu
     return cxmake( c.real(), c.imag() );
   }
 
-#endif // #if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+#endif // #if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
 
   //==========================================================================
 
-#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_THRUST // cuda + thrust
+#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_THRUST // cuda + thrust (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // CUDA - using thrust::complex
@@ -347,7 +352,7 @@ namespace mg5amcCpu
 
   //==========================================================================
 
-#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_CUCOMPLEX // cuda + cucomplex
+#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_CUCOMPLEX // cuda + cucomplex (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // CUDA - using cuComplex
@@ -566,7 +571,7 @@ namespace mg5amcCpu
 
   //==========================================================================
 
-#if not defined __CUDACC__ and defined MGONGPU_CPPCXTYPE_STDCOMPLEX // c++ + stdcomplex
+#if not defined __CUDACC__ and defined MGONGPU_CPPCXTYPE_STDCOMPLEX // c++/hip + stdcomplex (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // C++ - using std::complex
@@ -633,7 +638,7 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/heft_gg_h.sa/src/mgOnGpuFptypes.h b/epochX/cudacpp/heft_gg_h.sa/src/mgOnGpuFptypes.h
index 905c97d700..fa3a02664b 100644
--- a/epochX/cudacpp/heft_gg_h.sa/src/mgOnGpuFptypes.h
+++ b/epochX/cudacpp/heft_gg_h.sa/src/mgOnGpuFptypes.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUFPTYPES_H
 #define MGONGPUFPTYPES_H 1
@@ -12,7 +12,7 @@
 #include <cmath>
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL // cuda
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -20,7 +20,7 @@ namespace mg5amcCpu
 {
   //==========================================================================
 
-#ifdef __CUDACC__ // cuda
+#ifdef MGONGPUCPP_GPUIMPL // cuda
 
   //------------------------------
   // Floating point types - Cuda
@@ -64,11 +64,11 @@ namespace mg5amcCpu
 #endif
   }
 
-#endif // #ifdef __CUDACC__
+#endif // #ifdef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
   //------------------------------
   // Floating point types - C++
@@ -92,7 +92,7 @@ namespace mg5amcCpu
     return std::sqrt( f );
   }
 
-#endif // #ifndef __CUDACC__
+#endif // #ifndef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
diff --git a/epochX/cudacpp/heft_gg_h.sa/src/mgOnGpuVectors.h b/epochX/cudacpp/heft_gg_h.sa/src/mgOnGpuVectors.h
index e1299ba81e..cdae04326b 100644
--- a/epochX/cudacpp/heft_gg_h.sa/src/mgOnGpuVectors.h
+++ b/epochX/cudacpp/heft_gg_h.sa/src/mgOnGpuVectors.h
@@ -32,7 +32,7 @@
 #endif
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -131,7 +131,7 @@ namespace mg5amcCpu
 #endif
 #endif
 
-#else // i.e #ifndef MGONGPU_CPPSIMD (this includes #ifdef __CUDACC__)
+#else // i.e #ifndef MGONGPU_CPPSIMD (this includes #ifdef MGONGPUCPP_GPUIMPL)
 
   const int neppV = 1;
 
@@ -153,13 +153,13 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
   // Printout to stream for user defined types
 
@@ -805,11 +805,11 @@ namespace mg5amcCpu
 
 #endif // #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
 
-#endif // #ifndef __CUDACC__
+#endif // #ifndef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
   //------------------------------
   // Vector types - CUDA
@@ -853,12 +853,12 @@ namespace mg5amcCpu
     return mask;
   }
 
-#endif // #ifdef __CUDACC__
+#endif // #ifdef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
   // Scalar-or-vector types: scalar in CUDA, vector or scalar in C++
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   typedef bool bool_sv;
   typedef fptype fptype_sv;
   typedef fptype2 fptype2_sv;
@@ -879,7 +879,7 @@ namespace mg5amcCpu
 #endif
 
   // Scalar-or-vector zeros: scalar in CUDA, vector or scalar in C++
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   inline __host__ __device__ cxtype cxzero_sv(){ return cxtype( 0, 0 ); }
 #elif defined MGONGPU_CPPSIMD
   inline cxtype_v cxzero_sv() { return cxtype_v(); } // RRRR=0000 IIII=0000
diff --git a/epochX/cudacpp/heft_gg_h.sa/src/rambo.h b/epochX/cudacpp/heft_gg_h.sa/src/rambo.h
index e02ea52496..cd7e1008ea 100644
--- a/epochX/cudacpp/heft_gg_h.sa/src/rambo.h
+++ b/epochX/cudacpp/heft_gg_h.sa/src/rambo.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -18,7 +18,7 @@
 #include <iostream>
 
 // Simplified rambo version for 2 to N (with N>=2) processes with massless particles
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +83,7 @@ namespace mg5amcCpu
       static bool first = true;
       if( first )
       {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
         if constexpr( M_ACCESS::isOnDevice() ) // avoid
         {
           const int ievt0 = 0;
@@ -166,7 +166,7 @@ namespace mg5amcCpu
     wt = po2log;
     if( nparf != 2 ) wt = ( 2. * nparf - 4. ) * log( energy ) + z[nparf - 1];
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     // issue warnings if weight is too small or too large
     static int iwarn[5] = { 0, 0, 0, 0, 0 };
     if( wt < -180. )
diff --git a/epochX/cudacpp/heft_gg_h.sa/src/read_slha.cc b/epochX/cudacpp/heft_gg_h.sa/src/read_slha.cc
index 055b19a779..f8e46f2e66 100644
--- a/epochX/cudacpp/heft_gg_h.sa/src/read_slha.cc
+++ b/epochX/cudacpp/heft_gg_h.sa/src/read_slha.cc
@@ -11,7 +11,11 @@
 
 #include <algorithm>
 #include <cstdlib>
-#include <filesystem>
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
 #include <fstream>
 #include <iostream>
 
@@ -60,7 +64,14 @@ SLHAReader::read_slha_file( std::string file_name, bool verbose )
     {
       std::cout << "WARNING! Card file '" << file_name << "' does not exist:"
                 << " look for the file in directory $" << envpath << "='" << getenv( envpath ) << "'" << std::endl;
+      /*
+#ifdef __HIPCC__
+      const std::string file_name2 = std::experimental::filesystem::path( getenv( envpath ) ) / std::experimental::filesystem::path( file_name ).filename();
+#else
       const std::string file_name2 = std::filesystem::path( getenv( envpath ) ) / std::filesystem::path( file_name ).filename();
+#endif
+      */
+      const std::string file_name2 = std::string( getenv( envpath ) ) + "/" + file_name; // bypass std::filesystem #803
       param_card.open( file_name2.c_str(), std::ifstream::in );
       if( param_card.good() )
       {
diff --git a/epochX/cudacpp/pp_tt012j.mad/CODEGEN_mad_pp_tt012j_log.txt b/epochX/cudacpp/pp_tt012j.mad/CODEGEN_mad_pp_tt012j_log.txt
index ff161c336f..e01d29e02f 100644
--- a/epochX/cudacpp/pp_tt012j.mad/CODEGEN_mad_pp_tt012j_log.txt
+++ b/epochX/cudacpp/pp_tt012j.mad/CODEGEN_mad_pp_tt012j_log.txt
@@ -52,7 +52,7 @@ Note that you can still compile and run aMC@NLO with the built-in PDFs
 
 Using default text editor "vi". Set another one in ./input/mg5_configuration.txt
 Using default eps viewer "evince". Set another one in ./input/mg5_configuration.txt
-No valid web browser found. Please set in ./input/mg5_configuration.txt
+Using default web browser "firefox". Set another one in ./input/mg5_configuration.txt
 import /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_mad_pp_tt012j.mg
 The import format was not given, so we guess it as command
 set stdout_level DEBUG
@@ -61,7 +61,7 @@ set zerowidth_tchannel F
 define j = p
 INFO: load particles 
 INFO: load vertices 
-DEBUG: model prefixing  takes 0.005424976348876953 
+DEBUG: model prefixing  takes 0.005494594573974609 
 INFO: Restrict model sm with file models/sm/restrict_default.dat . 
 DEBUG: Simplifying conditional expressions 
 DEBUG: remove interactions: u s w+ at order: QED=1 
@@ -172,7 +172,7 @@ INFO: Process u~ u > t t~ added to mirror process u u~ > t t~
 INFO: Process c~ c > t t~ added to mirror process c c~ > t t~ 
 INFO: Process d~ d > t t~ added to mirror process d d~ > t t~ 
 INFO: Process s~ s > t t~ added to mirror process s s~ > t t~ 
-5 processes with 7 diagrams generated in 0.030 s
+5 processes with 7 diagrams generated in 0.029 s
 Total: 5 processes with 7 diagrams
 add process p p > t t~ j @1
 INFO: Checking for minimal orders which gives processes. 
@@ -212,7 +212,7 @@ INFO: Process d~ g > t t~ d~ added to mirror process g d~ > t t~ d~
 INFO: Process d~ d > t t~ g added to mirror process d d~ > t t~ g 
 INFO: Process s~ g > t t~ s~ added to mirror process g s~ > t t~ s~ 
 INFO: Process s~ s > t t~ g added to mirror process s s~ > t t~ g 
-13 processes with 76 diagrams generated in 0.139 s
+13 processes with 76 diagrams generated in 0.136 s
 Total: 18 processes with 83 diagrams
 add process p p > t t~ j j @2
 INFO: Checking for minimal orders which gives processes. 
@@ -378,7 +378,7 @@ INFO: Process s~ u~ > t t~ u~ s~ added to mirror process u~ s~ > t t~ u~ s~
 INFO: Process s~ c~ > t t~ c~ s~ added to mirror process c~ s~ > t t~ c~ s~ 
 INFO: Process s~ d~ > t t~ d~ s~ added to mirror process d~ s~ > t t~ d~ s~ 
 INFO: Crossed process found for s~ s~ > t t~ s~ s~, reuse diagrams. 
-65 processes with 1119 diagrams generated in 1.876 s
+65 processes with 1119 diagrams generated in 1.826 s
 Total: 83 processes with 1202 diagrams
 output madevent ../TMPOUT/CODEGEN_mad_pp_tt012j --hel_recycling=False --vector_size=32 --me_exporter=standalone_cudacpp
 Load PLUGIN.CUDACPP_OUTPUT
@@ -497,7 +497,7 @@ INFO: Combined process d d~ > t t~ WEIGHTED<=2 with process u u~ > t t~ WEIGHTED
 INFO: Combined process s s~ > t t~ WEIGHTED<=2 with process u u~ > t t~ WEIGHTED<=2 
 INFO: Creating files in directory P2_gg_ttxgg 
 DEBUG:  kwargs[prefix] = 0 [model_handling.py at line 1058] 
-DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7ff282794c40> [export_v4.py at line 6262] 
+DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7f5bfe753e80> [export_v4.py at line 6262] 
 INFO: Creating files in directory . 
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.h
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.cc
@@ -514,7 +514,7 @@ INFO: Generating Feynman diagrams for Process: g g > t t~ g g WEIGHTED<=4 @2
 INFO: Finding symmetric diagrams for subprocess group gg_ttxgg 
 INFO: Creating files in directory P2_gg_ttxuux 
 DEBUG:  kwargs[prefix] = 0 [model_handling.py at line 1058] 
-DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7ff282794430> [export_v4.py at line 6262] 
+DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7f5bfdf5dd90> [export_v4.py at line 6262] 
 INFO: Creating files in directory . 
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.h
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.cc
@@ -531,7 +531,7 @@ INFO: Generating Feynman diagrams for Process: g g > t t~ u u~ WEIGHTED<=4 @2
 INFO: Finding symmetric diagrams for subprocess group gg_ttxuux 
 INFO: Creating files in directory P2_gu_ttxgu 
 DEBUG:  kwargs[prefix] = 0 [model_handling.py at line 1058] 
-DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7ff282addeb0> [export_v4.py at line 6262] 
+DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7f5bfe2af190> [export_v4.py at line 6262] 
 INFO: Creating files in directory . 
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.h
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.cc
@@ -548,7 +548,7 @@ INFO: Generating Feynman diagrams for Process: g u > t t~ g u WEIGHTED<=4 @2
 INFO: Finding symmetric diagrams for subprocess group gu_ttxgu 
 INFO: Creating files in directory P2_gux_ttxgux 
 DEBUG:  kwargs[prefix] = 0 [model_handling.py at line 1058] 
-DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7ff2832e91c0> [export_v4.py at line 6262] 
+DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7f5bfdeee490> [export_v4.py at line 6262] 
 INFO: Creating files in directory . 
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.h
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.cc
@@ -565,7 +565,7 @@ INFO: Generating Feynman diagrams for Process: g u~ > t t~ g u~ WEIGHTED<=4 @2
 INFO: Finding symmetric diagrams for subprocess group gux_ttxgux 
 INFO: Creating files in directory P2_uux_ttxgg 
 DEBUG:  kwargs[prefix] = 0 [model_handling.py at line 1058] 
-DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7ff282919250> [export_v4.py at line 6262] 
+DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7f5bfe2af190> [export_v4.py at line 6262] 
 INFO: Creating files in directory . 
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.h
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.cc
@@ -582,7 +582,7 @@ INFO: Generating Feynman diagrams for Process: u u~ > t t~ g g WEIGHTED<=4 @2
 INFO: Finding symmetric diagrams for subprocess group uux_ttxgg 
 INFO: Creating files in directory P1_gg_ttxg 
 DEBUG:  kwargs[prefix] = 0 [model_handling.py at line 1058] 
-DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7ff282919250> [export_v4.py at line 6262] 
+DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7f5bfdf677c0> [export_v4.py at line 6262] 
 INFO: Creating files in directory . 
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.h
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.cc
@@ -599,7 +599,7 @@ INFO: Generating Feynman diagrams for Process: g g > t t~ g WEIGHTED<=3 @1
 INFO: Finding symmetric diagrams for subprocess group gg_ttxg 
 INFO: Creating files in directory P2_uu_ttxuu 
 DEBUG:  kwargs[prefix] = 0 [model_handling.py at line 1058] 
-DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7ff282ae3ca0> [export_v4.py at line 6262] 
+DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7f5bfdf677c0> [export_v4.py at line 6262] 
 INFO: Creating files in directory . 
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.h
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.cc
@@ -616,7 +616,7 @@ INFO: Generating Feynman diagrams for Process: u u > t t~ u u WEIGHTED<=4 @2
 INFO: Finding symmetric diagrams for subprocess group uu_ttxuu 
 INFO: Creating files in directory P2_uux_ttxuux 
 DEBUG:  kwargs[prefix] = 0 [model_handling.py at line 1058] 
-DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7ff282aa5be0> [export_v4.py at line 6262] 
+DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7f5bfdf677c0> [export_v4.py at line 6262] 
 INFO: Creating files in directory . 
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.h
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.cc
@@ -633,7 +633,7 @@ INFO: Generating Feynman diagrams for Process: u u~ > t t~ u u~ WEIGHTED<=4 @2
 INFO: Finding symmetric diagrams for subprocess group uux_ttxuux 
 INFO: Creating files in directory P2_uxux_ttxuxux 
 DEBUG:  kwargs[prefix] = 0 [model_handling.py at line 1058] 
-DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7ff282fa8f70> [export_v4.py at line 6262] 
+DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7f5bfdf677c0> [export_v4.py at line 6262] 
 INFO: Creating files in directory . 
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.h
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.cc
@@ -650,7 +650,7 @@ INFO: Generating Feynman diagrams for Process: u~ u~ > t t~ u~ u~ WEIGHTED<=4 @2
 INFO: Finding symmetric diagrams for subprocess group uxux_ttxuxux 
 INFO: Creating files in directory P2_uc_ttxuc 
 DEBUG:  kwargs[prefix] = 0 [model_handling.py at line 1058] 
-DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7ff282e2d850> [export_v4.py at line 6262] 
+DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7f5bfe753e80> [export_v4.py at line 6262] 
 INFO: Creating files in directory . 
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.h
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.cc
@@ -667,7 +667,7 @@ INFO: Generating Feynman diagrams for Process: u c > t t~ u c WEIGHTED<=4 @2
 INFO: Finding symmetric diagrams for subprocess group uc_ttxuc 
 INFO: Creating files in directory P2_uux_ttxccx 
 DEBUG:  kwargs[prefix] = 0 [model_handling.py at line 1058] 
-DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7ff282addd60> [export_v4.py at line 6262] 
+DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7f5bfe753e80> [export_v4.py at line 6262] 
 INFO: Creating files in directory . 
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.h
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.cc
@@ -684,7 +684,7 @@ INFO: Generating Feynman diagrams for Process: u u~ > t t~ c c~ WEIGHTED<=4 @2
 INFO: Finding symmetric diagrams for subprocess group uux_ttxccx 
 INFO: Creating files in directory P2_ucx_ttxucx 
 DEBUG:  kwargs[prefix] = 0 [model_handling.py at line 1058] 
-DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7ff282a84d30> [export_v4.py at line 6262] 
+DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7f5bfdf5dd00> [export_v4.py at line 6262] 
 INFO: Creating files in directory . 
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.h
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.cc
@@ -701,7 +701,7 @@ INFO: Generating Feynman diagrams for Process: u c~ > t t~ u c~ WEIGHTED<=4 @2
 INFO: Finding symmetric diagrams for subprocess group ucx_ttxucx 
 INFO: Creating files in directory P2_uxcx_ttxuxcx 
 DEBUG:  kwargs[prefix] = 0 [model_handling.py at line 1058] 
-DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7ff282e2d850> [export_v4.py at line 6262] 
+DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7f5bfdf5dd00> [export_v4.py at line 6262] 
 INFO: Creating files in directory . 
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.h
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.cc
@@ -718,7 +718,7 @@ INFO: Generating Feynman diagrams for Process: u~ c~ > t t~ u~ c~ WEIGHTED<=4 @2
 INFO: Finding symmetric diagrams for subprocess group uxcx_ttxuxcx 
 INFO: Creating files in directory P1_gu_ttxu 
 DEBUG:  kwargs[prefix] = 0 [model_handling.py at line 1058] 
-DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7ff282e2d850> [export_v4.py at line 6262] 
+DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7f5bfdf5dd90> [export_v4.py at line 6262] 
 INFO: Creating files in directory . 
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.h
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.cc
@@ -735,7 +735,7 @@ INFO: Generating Feynman diagrams for Process: g u > t t~ u WEIGHTED<=3 @1
 INFO: Finding symmetric diagrams for subprocess group gu_ttxu 
 INFO: Creating files in directory P1_gux_ttxux 
 DEBUG:  kwargs[prefix] = 0 [model_handling.py at line 1058] 
-DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7ff282e2d850> [export_v4.py at line 6262] 
+DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7f5bfdeb2730> [export_v4.py at line 6262] 
 INFO: Creating files in directory . 
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.h
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.cc
@@ -752,7 +752,7 @@ INFO: Generating Feynman diagrams for Process: g u~ > t t~ u~ WEIGHTED<=3 @1
 INFO: Finding symmetric diagrams for subprocess group gux_ttxux 
 INFO: Creating files in directory P1_uux_ttxg 
 DEBUG:  kwargs[prefix] = 0 [model_handling.py at line 1058] 
-DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7ff282aa3a00> [export_v4.py at line 6262] 
+DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7f5bfe3240a0> [export_v4.py at line 6262] 
 INFO: Creating files in directory . 
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.h
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.cc
@@ -769,7 +769,7 @@ INFO: Generating Feynman diagrams for Process: u u~ > t t~ g WEIGHTED<=3 @1
 INFO: Finding symmetric diagrams for subprocess group uux_ttxg 
 INFO: Creating files in directory P0_gg_ttx 
 DEBUG:  kwargs[prefix] = 0 [model_handling.py at line 1058] 
-DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7ff282addd60> [export_v4.py at line 6262] 
+DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7f5bfe753e80> [export_v4.py at line 6262] 
 INFO: Creating files in directory . 
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.h
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.cc
@@ -786,7 +786,7 @@ INFO: Generating Feynman diagrams for Process: g g > t t~ WEIGHTED<=2
 INFO: Finding symmetric diagrams for subprocess group gg_ttx 
 INFO: Creating files in directory P0_uux_ttx 
 DEBUG:  kwargs[prefix] = 0 [model_handling.py at line 1058] 
-DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7ff282794c40> [export_v4.py at line 6262] 
+DEBUG:  process_exporter_cpp =  <PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_OneProcessExporter object at 0x7f5bfe2ad940> [export_v4.py at line 6262] 
 INFO: Creating files in directory . 
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.h
 FileWriter <class 'PLUGIN.CUDACPP_OUTPUT.model_handling.PLUGIN_CPPWriter'> for ././CPPProcess.cc
@@ -801,15 +801,15 @@ INFO: Created files CPPProcess.h and CPPProcess.cc in directory ./.
 DEBUG:  vector, subproc_group,self.opt['vector_size'] =  32 True 32 [export_v4.py at line 1872] 
 INFO: Generating Feynman diagrams for Process: u u~ > t t~ WEIGHTED<=2 
 INFO: Finding symmetric diagrams for subprocess group uux_ttx 
-Generated helas calls for 18 subprocesses (372 diagrams) in 1.312 s
-Wrote files for 810 helas calls in 3.308 s
+Generated helas calls for 18 subprocesses (372 diagrams) in 1.285 s
+Wrote files for 810 helas calls in 3.231 s
 ALOHA: aloha starts to compute helicity amplitudes
 ALOHA: aloha creates VVV1 routines
 ALOHA: aloha creates FFV1 routines
 ALOHA: aloha creates VVVV1 routines
 ALOHA: aloha creates VVVV3 routines
 ALOHA: aloha creates VVVV4 routines
-ALOHA: aloha creates 5 routines in  0.342 s
+ALOHA: aloha creates 5 routines in  0.334 s
 DEBUG:  Entering PLUGIN_ProcessExporter.convert_model (create the model) [output.py at line 202] 
 ALOHA: aloha starts to compute helicity amplitudes
 ALOHA: aloha creates VVV1 routines
@@ -817,7 +817,7 @@ ALOHA: aloha creates FFV1 routines
 ALOHA: aloha creates VVVV1 routines
 ALOHA: aloha creates VVVV3 routines
 ALOHA: aloha creates VVVV4 routines
-ALOHA: aloha creates 10 routines in  0.321 s
+ALOHA: aloha creates 10 routines in  0.314 s
 <class 'aloha.create_aloha.AbstractRoutine'> VVV1
 <class 'aloha.create_aloha.AbstractRoutine'> VVV1
 <class 'aloha.create_aloha.AbstractRoutine'> FFV1
@@ -1028,9 +1028,9 @@ Type "launch" to generate events from this process, or see
 Run "open index.html" to see more information about this process.
 quit
 
-real	0m9.073s
-user	0m8.514s
-sys	0m0.464s
+real	0m8.840s
+user	0m8.307s
+sys	0m0.499s
 Code generation completed in 9 seconds
 ************************************************************
 *                                                          *
@@ -1057,7 +1057,7 @@ INFO: load configuration from /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/mg5amc
 INFO: load configuration from /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_mad_pp_tt012j/Cards/me5_configuration.txt  
 Using default text editor "vi". Set another one in ./input/mg5_configuration.txt
 Using default eps viewer "evince". Set another one in ./input/mg5_configuration.txt
-No valid web browser found. Please set in ./input/mg5_configuration.txt
+Using default web browser "firefox". Set another one in ./input/mg5_configuration.txt
 treatcards run
 quit
 INFO:  
@@ -1087,7 +1087,7 @@ INFO: load configuration from /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/mg5amc
 INFO: load configuration from /data/avalassi/GPU2023/madgraph4gpuX/MG5aMC/TMPOUT/CODEGEN_mad_pp_tt012j/Cards/me5_configuration.txt  
 Using default text editor "vi". Set another one in ./input/mg5_configuration.txt
 Using default eps viewer "evince". Set another one in ./input/mg5_configuration.txt
-No valid web browser found. Please set in ./input/mg5_configuration.txt
+Using default web browser "firefox". Set another one in ./input/mg5_configuration.txt
 treatcards param
 quit
 INFO:  
diff --git a/epochX/cudacpp/pp_tt012j.mad/COPYRIGHT b/epochX/cudacpp/pp_tt012j.mad/COPYRIGHT
index a134b5fef9..9036d9260a 100644
--- a/epochX/cudacpp/pp_tt012j.mad/COPYRIGHT
+++ b/epochX/cudacpp/pp_tt012j.mad/COPYRIGHT
@@ -15,6 +15,7 @@ The full development team currently includes the following authors :
   Stephan Hageboeck (CERN)
   Olivier Mattelaer (Universite Catholique de Louvain, original author)
   Stefan Roiser (CERN, original author)
+  Jorgen Teig (CERN)
   Andrea Valassi (CERN, original author)
   Zenny Wettersten (CERN)
 See https://github.com/madgraph5/madgraph4gpu for more details. For the full
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/Bridge.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/Bridge.h
index bf8b5e024d..f9ed70dfde 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/Bridge.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/Bridge.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Roiser (Nov 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef BRIDGE_H
 #define BRIDGE_H 1
@@ -14,16 +14,23 @@
 #include "MemoryAccessMomenta.h"  // for MemoryAccessMomenta::neppM
 #include "MemoryBuffers.h"        // for HostBufferMomenta, DeviceBufferMomenta etc
 
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
+
+#include <sys/stat.h> // bypass std::filesystem #803
+
 #include <algorithm>
 #include <cassert>
 #include <cmath>
 #include <cstring>
-#include <filesystem>
 #include <iostream>
 #include <memory>
 #include <type_traits>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +90,7 @@ namespace mg5amcCpu
     Bridge& operator=( const Bridge& ) = delete;
     Bridge& operator=( Bridge&& ) = delete;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     /**
      * Set the gpublocks and gputhreads for the gpusequence - throws if evnt != gpublocks*gputhreads
      * (this is needed for BridgeKernel tests rather than for actual production use in Fortran)
@@ -150,7 +157,7 @@ namespace mg5amcCpu
     unsigned int m_nevt; // number of events
     int m_nGoodHel;      // the number of good helicities (-1 initially when they have not yet been calculated)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     int m_gputhreads; // number of gpu threads (default set from number of events, can be modified)
     int m_gpublocks;  // number of gpu blocks (default set from number of events, can be modified)
     DeviceBuffer<FORTRANFPTYPE, sizePerEventMomenta> m_devMomentaF;
@@ -187,12 +194,12 @@ namespace mg5amcCpu
   // Forward declare transposition methods
   //
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
   template<typename Tin, typename Tout>
   __global__ void dev_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt );
 
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
 
   template<typename Tin, typename Tout>
   void hst_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt );
@@ -209,7 +216,7 @@ namespace mg5amcCpu
   Bridge<FORTRANFPTYPE>::Bridge( unsigned int nevtF, unsigned int nparF, unsigned int np4F )
     : m_nevt( nevtF )
     , m_nGoodHel( -1 )
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     , m_gputhreads( 256 )                  // default number of gpu threads
     , m_gpublocks( m_nevt / m_gputhreads ) // this ensures m_nevt <= m_gpublocks*m_gputhreads
     , m_devMomentaF( m_nevt )
@@ -233,7 +240,7 @@ namespace mg5amcCpu
   {
     if( nparF != CPPProcess::npar ) throw std::runtime_error( "Bridge constructor: npar mismatch" );
     if( np4F != CPPProcess::np4 ) throw std::runtime_error( "Bridge constructor: np4 mismatch" );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( ( m_nevt < s_gputhreadsmin ) || ( m_nevt % s_gputhreadsmin != 0 ) )
       throw std::runtime_error( "Bridge constructor: nevt should be a multiple of " + std::to_string( s_gputhreadsmin ) );
     while( m_nevt != m_gpublocks * m_gputhreads )
@@ -249,20 +256,28 @@ namespace mg5amcCpu
 #else
     std::cout << "WARNING! Instantiate host Bridge (nevt=" << m_nevt << ")" << std::endl;
     m_pmek.reset( new MatrixElementKernelHost( m_hstMomentaC, m_hstGs, m_hstRndHel, m_hstRndCol, m_hstMEs, m_hstSelHel, m_hstSelCol, m_nevt ) );
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
     // Create a process object, read param card and set parameters
     // FIXME: the process instance can happily go out of scope because it is only needed to read parameters?
     // FIXME: the CPPProcess should really be a singleton? what if fbridgecreate is called from several Fortran threads?
     CPPProcess process( /*verbose=*/false );
     std::string paramCard = "../../Cards/param_card.dat";
-    if( !std::filesystem::exists( paramCard ) )
-    {
-      paramCard = "../" + paramCard;
-    }
+    /*
+#ifdef __HIPCC__
+    if( !std::experimental::filesystem::exists( paramCard ) ) paramCard = "../" + paramCard;
+#else
+    if( !std::filesystem::exists( paramCard ) ) paramCard = "../" + paramCard;
+#endif
+    */
+    //struct stat dummybuffer; // bypass std::filesystem #803
+    //if( !( stat( paramCard.c_str(), &dummyBuffer ) == 0 ) ) paramCard = "../" + paramCard; //
+    auto fileExists = []( std::string& fileName )
+    { struct stat buffer; return stat( fileName.c_str(), &buffer ) == 0; };
+    if( !fileExists( paramCard ) ) paramCard = "../" + paramCard; // bypass std::filesystem #803
     process.initProc( paramCard );
   }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::set_gpugrid( const int gpublocks, const int gputhreads )
   {
@@ -276,7 +291,7 @@ namespace mg5amcCpu
   }
 #endif
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::gpu_sequence( const FORTRANFPTYPE* momenta,
                                             const FORTRANFPTYPE* gs,
@@ -291,14 +306,14 @@ namespace mg5amcCpu
     constexpr int neppM = MemoryAccessMomenta::neppM;
     if constexpr( neppM == 1 && std::is_same_v<FORTRANFPTYPE, fptype> )
     {
-      checkCuda( cudaMemcpy( m_devMomentaC.data(), momenta, m_devMomentaC.bytes(), cudaMemcpyHostToDevice ) );
+      gpuMemcpy( m_devMomentaC.data(), momenta, m_devMomentaC.bytes(), gpuMemcpyHostToDevice );
     }
     else
     {
-      checkCuda( cudaMemcpy( m_devMomentaF.data(), momenta, m_devMomentaF.bytes(), cudaMemcpyHostToDevice ) );
+      gpuMemcpy( m_devMomentaF.data(), momenta, m_devMomentaF.bytes(), gpuMemcpyHostToDevice );
       const int thrPerEvt = CPPProcess::npar * CPPProcess::np4; // AV: transpose alg does 1 element per thread (NOT 1 event per thread)
       //const int thrPerEvt = 1; // AV: try new alg with 1 event per thread... this seems slower
-      dev_transposeMomentaF2C<<<m_gpublocks * thrPerEvt, m_gputhreads>>>( m_devMomentaF.data(), m_devMomentaC.data(), m_nevt );
+      gpuLaunchKernel( dev_transposeMomentaF2C, m_gpublocks * thrPerEvt, m_gputhreads, m_devMomentaF.data(), m_devMomentaC.data(), m_nevt );
     }
     if constexpr( std::is_same_v<FORTRANFPTYPE, fptype> )
     {
@@ -341,7 +356,7 @@ namespace mg5amcCpu
   }
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   template<typename FORTRANFPTYPE>
   void Bridge<FORTRANFPTYPE>::cpu_sequence( const FORTRANFPTYPE* momenta,
                                             const FORTRANFPTYPE* gs,
@@ -396,7 +411,7 @@ namespace mg5amcCpu
   // - C++ array: momenta[npagM][npar][np4][neppM] with nevt=npagM*neppM (AOSOA)
   //
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<typename Tin, typename Tout>
   __global__ void dev_transposeMomentaF2C( const Tin* in, Tout* out, const unsigned int nevt )
   {
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/BridgeKernels.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/BridgeKernels.cc
index d58066c9c1..eaf4037a24 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/BridgeKernels.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/BridgeKernels.cc
@@ -1,17 +1,18 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "BridgeKernels.h"
 
+#include "GpuAbstraction.h"
 #include "MemoryAccessMomenta.h"
 
 #include <sstream>
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -45,7 +46,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu
 {
 
@@ -96,7 +97,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/BridgeKernels.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/BridgeKernels.h
index 15eb4bff4d..3efef8ce97 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/BridgeKernels.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/BridgeKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef BRIDGEKERNELS_H
 #define BRIDGEKERNELS_H 1
@@ -12,7 +12,7 @@
 #include "MatrixElementKernels.h"
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -49,7 +49,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A Bridge wrapper class encapsulating matrix element calculations on a CPU host
   class BridgeKernelHost final : public BridgeKernelBase
   {
@@ -89,7 +89,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A Bridge wrapper class encapsulating matrix element calculations on a GPU device
   class BridgeKernelDevice : public BridgeKernelBase
   {
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/CommonRandomNumberKernel.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/CommonRandomNumberKernel.cc
index 985b39f576..010bc4cbd0 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/CommonRandomNumberKernel.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/CommonRandomNumberKernel.cc
@@ -1,15 +1,16 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "CommonRandomNumbers.h"
+#include "GpuAbstraction.h"
 #include "MemoryBuffers.h"
 #include "RandomNumberKernels.h"
 
 #include <cassert>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/CrossSectionKernels.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/CrossSectionKernels.cc
index 0b355a3c8d..c15b39844d 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/CrossSectionKernels.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/CrossSectionKernels.cc
@@ -1,10 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "CrossSectionKernels.h"
 
+#include "GpuAbstraction.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessWeights.h"
 #include "MemoryBuffers.h"
@@ -77,7 +78,7 @@ debug_me_is_abnormal( const fptype& me, size_t ievtALL )
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -185,7 +186,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/CrossSectionKernels.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/CrossSectionKernels.h
index 7933ca4bbf..4d9659e04e 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/CrossSectionKernels.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/CrossSectionKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef CROSSSECTIONKERNELS_H
 #define CROSSSECTIONKERNELS_H 1
@@ -13,7 +13,7 @@
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -96,7 +96,7 @@ namespace mg5amcCpu
   //--------------------------------------------------------------------------
 
   /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating the calculation of event statistics on a GPU device
   class CrossSectionKernelDevice : public CrossSectionKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/CudaRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/CudaRuntime.h
deleted file mode 100644
index 64ce52f4b3..0000000000
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/CudaRuntime.h
+++ /dev/null
@@ -1,85 +0,0 @@
-// Copyright (C) 2020-2023 CERN and UCLouvain.
-// Licensed under the GNU Lesser General Public License (version 3 or later).
-// Created by: S. Roiser (Jul 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
-
-#ifndef MG5AMC_CUDARUNTIME_H
-#define MG5AMC_CUDARUNTIME_H 1
-
-// MG5AMC on GPU uses the CUDA runtime API, not the lower level CUDA driver API
-// See https://docs.nvidia.com/cuda/cuda-runtime-api/driver-vs-runtime-api.html#driver-vs-runtime-api
-
-#include <cassert>
-#include <iostream>
-
-//--------------------------------------------------------------------------
-
-// See https://stackoverflow.com/a/14038590
-#ifdef __CUDACC__ /* clang-format off */
-#define checkCuda( code ) { assertCuda( code, __FILE__, __LINE__ ); }
-inline void assertCuda( cudaError_t code, const char* file, int line, bool abort = true )
-{
-  if( code != cudaSuccess )
-  {
-    printf( "ERROR! assertCuda: '%s' (%d) in %s:%d\n", cudaGetErrorString( code ), code, file, line );
-    if( abort ) assert( code == cudaSuccess );
-  }
-}
-#endif /* clang-format on */
-
-//--------------------------------------------------------------------------
-
-#ifdef __CUDACC__
-namespace mg5amcGpu
-{
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  // *** FIXME! This will all need to be designed differently when going to multi-GPU nodes! ***
-  struct CudaRuntime final
-  {
-    CudaRuntime( const bool debug = true )
-      : m_debug( debug ) { setUp( m_debug ); }
-    ~CudaRuntime() { tearDown( m_debug ); }
-    CudaRuntime( const CudaRuntime& ) = delete;
-    CudaRuntime( CudaRuntime&& ) = delete;
-    CudaRuntime& operator=( const CudaRuntime& ) = delete;
-    CudaRuntime& operator=( CudaRuntime&& ) = delete;
-    bool m_debug;
-
-    // Set up CUDA application
-    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
-    // Calling cudaSetDevice on startup is useful to properly book-keep the time spent in CUDA initialization
-    static void setUp( const bool debug = true )
-    {
-      // ** NB: it is useful to call cudaSetDevice, or cudaFree, to properly book-keep the time spent in CUDA initialization
-      // ** NB: otherwise, the first CUDA operation (eg a cudaMemcpyToSymbol in CPPProcess ctor) appears to take much longer!
-      /*
-      // [We initially added cudaFree(0) to "ease profile analysis" only because it shows up as a big recognizable block!]
-      // No explicit initialization is needed: https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#initialization
-      // It is not clear what cudaFree(0) does at all: https://stackoverflow.com/questions/69967813/
-      if ( debug ) std::cout << "__CudaRuntime: calling cudaFree(0)" << std::endl;
-      checkCuda( cudaFree( 0 ) ); // SLOW!
-      */
-      // Replace cudaFree(0) by cudaSetDevice(0), even if it is not really needed either
-      // (but see https://developer.nvidia.com/blog/cuda-pro-tip-always-set-current-device-avoid-multithreading-bugs)
-      if( debug ) std::cout << "__CudaRuntime: calling cudaSetDevice(0)" << std::endl;
-      checkCuda( cudaSetDevice( 0 ) ); // SLOW!
-    }
-
-    // Tear down CUDA application (call cudaDeviceReset)
-    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
-    // Calling cudaDeviceReset on shutdown is only needed for checking memory leaks in cuda-memcheck
-    // See https://docs.nvidia.com/cuda/cuda-memcheck/index.html#leak-checking
-    static void tearDown( const bool debug = true )
-    {
-      if( debug ) std::cout << "__CudaRuntime: calling cudaDeviceReset()" << std::endl;
-      checkCuda( cudaDeviceReset() );
-    }
-  };
-
-}
-#endif
-
-//--------------------------------------------------------------------------
-
-#endif // MG5AMC_CUDARUNTIME_H
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/CurandRandomNumberKernel.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/CurandRandomNumberKernel.cc
index eb56333b03..08a16f6f2c 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/CurandRandomNumberKernel.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/CurandRandomNumberKernel.cc
@@ -1,9 +1,9 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "MemoryBuffers.h"
 #include "RandomNumberKernels.h"
 
@@ -22,7 +22,7 @@ inline void assertCurand( curandStatus_t code, const char *file, int line, bool
 }
 #endif /* clang-format on */
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -36,7 +36,7 @@ namespace mg5amcCpu
   {
     if( m_isOnDevice )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if( !m_rnarray.isOnDevice() )
         throw std::runtime_error( "CurandRandomNumberKernel on device with a host random number array" );
 #else
@@ -114,7 +114,7 @@ namespace mg5amcCpu
     /*
     printf( "\nCurandRandomNumberKernel::generateRnarray size = %d\n", (int)m_rnarray.size() );
     fptype* data = m_rnarray.data();
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( m_rnarray.isOnDevice() )
     {
       data = new fptype[m_rnarray.size()]();
@@ -123,7 +123,7 @@ namespace mg5amcCpu
 #endif
     for( int i = 0; i < ( (int)m_rnarray.size() / 4 ); i++ )
       printf( "[%4d] %f %f %f %f\n", i * 4, data[i * 4], data[i * 4 + 2], data[i * 4 + 2], data[i * 4 + 3] );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( m_rnarray.isOnDevice() ) delete[] data;
 #endif
     */
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/EventStatistics.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/EventStatistics.h
index 48b51e0a49..b425a5bade 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/EventStatistics.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/EventStatistics.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef EventStatistics_H
 #define EventStatistics_H 1
@@ -16,7 +16,7 @@
 #include <limits>
 #include <string>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/GpuAbstraction.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/GpuAbstraction.h
new file mode 100644
index 0000000000..9c467b1e04
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/GpuAbstraction.h
@@ -0,0 +1,69 @@
+// Copyright (C) 2020-2023 CERN and UCLouvain.
+// Licensed under the GNU Lesser General Public License (version 3 or later).
+// Created by: J. Teig (Jul 2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+
+#ifndef MG5AMC_GPUABSTRACTION_H
+#define MG5AMC_GPUABSTRACTION_H 1
+
+#include <cassert>
+
+//--------------------------------------------------------------------------
+
+#ifdef __CUDACC__
+
+#define gpuError_t cudaError_t
+#define gpuPeekAtLastError cudaPeekAtLastError
+#define gpuGetErrorString cudaGetErrorString
+#define gpuSuccess cudaSuccess
+
+#define gpuMallocHost( ptr, size ) checkGpu( cudaMallocHost( ptr, size ) )
+#define gpuMalloc( ptr, size ) checkGpu( cudaMalloc( ptr, size ) )
+
+#define gpuMemcpy( dstData, srcData, srcBytes, func ) checkGpu( cudaMemcpy( dstData, srcData, srcBytes, func ) )
+#define gpuMemcpyHostToDevice cudaMemcpyHostToDevice
+#define gpuMemcpyDeviceToHost cudaMemcpyDeviceToHost
+#define gpuMemcpyToSymbol( type1, type2, size ) checkGpu( cudaMemcpyToSymbol( type1, type2, size ) )
+
+#define gpuFree( ptr ) checkGpu( cudaFree( ptr ) )
+#define gpuFreeHost( ptr ) checkGpu( cudaFreeHost( ptr ) )
+
+#define gpuSetDevice cudaSetDevice
+#define gpuDeviceSynchronize cudaDeviceSynchronize
+#define gpuDeviceReset cudaDeviceReset
+
+#define gpuLaunchKernel( kernel, blocks, threads, ... ) kernel<<<blocks, threads>>>( __VA_ARGS__ )
+#define gpuLaunchKernelSharedMem( kernel, blocks, threads, sharedMem, ... ) kernel<<<blocks, threads, sharedMem>>>( __VA_ARGS__ )
+
+//--------------------------------------------------------------------------
+
+#elif defined __HIPCC__
+
+#define gpuError_t hipError_t
+#define gpuPeekAtLastError hipPeekAtLastError
+#define gpuGetErrorString hipGetErrorString
+#define gpuSuccess hipSuccess
+
+#define gpuMallocHost( ptr, size ) checkGpu( hipHostMalloc( ptr, size ) ) // HostMalloc better
+#define gpuMalloc( ptr, size ) checkGpu( hipMalloc( ptr, size ) )
+
+#define gpuMemcpy( dstData, srcData, srcBytes, func ) checkGpu( hipMemcpy( dstData, srcData, srcBytes, func ) )
+#define gpuMemcpyHostToDevice hipMemcpyHostToDevice
+#define gpuMemcpyDeviceToHost hipMemcpyDeviceToHost
+#define gpuMemcpyToSymbol( type1, type2, size ) checkGpu( hipMemcpyToSymbol( type1, type2, size ) )
+
+#define gpuFree( ptr ) checkGpu( hipFree( ptr ) )
+#define gpuFreeHost( ptr ) checkGpu( hipHostFree( ptr ) )
+
+#define gpuSetDevice hipSetDevice
+#define gpuDeviceSynchronize hipDeviceSynchronize
+#define gpuDeviceReset hipDeviceReset
+
+#define gpuLaunchKernel( kernel, blocks, threads, ... ) kernel<<<blocks, threads>>>( __VA_ARGS__ )
+#define gpuLaunchKernelSharedMem( kernel, blocks, threads, sharedMem, ... ) kernel<<<blocks, threads, sharedMem>>>( __VA_ARGS__ )
+
+//--------------------------------------------------------------------------
+
+#endif
+
+#endif // MG5AMC_GPUABSTRACTION_H
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/GpuRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/GpuRuntime.h
new file mode 100644
index 0000000000..93579ef08b
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/GpuRuntime.h
@@ -0,0 +1,85 @@
+// Copyright (C) 2020-2023 CERN and UCLouvain.
+// Licensed under the GNU Lesser General Public License (version 3 or later).
+// Created by: J. Teig (Jun 2023, based on earlier work by S. Roiser) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+
+#ifndef MG5AMC_GPURUNTIME_H
+#define MG5AMC_GPURUNTIME_H 1
+
+// MG5AMC on GPU uses the CUDA runtime API, not the lower level CUDA driver API
+// See https://docs.nvidia.com/cuda/cuda-runtime-api/driver-vs-runtime-api.html#driver-vs-runtime-api
+
+#include "GpuAbstraction.h"
+
+#include <iostream>
+
+//--------------------------------------------------------------------------
+
+// See https://stackoverflow.com/a/14038590
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
+#define checkGpu( code ) { assertGpu( code, __FILE__, __LINE__ ); }
+inline void assertGpu( gpuError_t code, const char* file, int line, bool abort = true )
+{
+  if( code != gpuSuccess )
+  {
+    printf( "ERROR! assertGpu: '%s' (%d) in %s:%d\n", gpuGetErrorString( code ), code, file, line );
+    if( abort ) assert( code == gpuSuccess );
+  }
+}
+#endif /* clang-format on */
+
+//--------------------------------------------------------------------------
+
+#ifdef MGONGPUCPP_GPUIMPL
+namespace mg5amcGpu
+{
+  // Instantiate a GpuRuntime at the beginnining of the application's main to
+  // invoke gpuSetDevice(0) in the constructor and book a gpuDeviceReset() call in the destructor
+  // *** FIXME! This will all need to be designed differently when going to multi-GPU nodes! ***
+  struct GpuRuntime final
+  {
+    GpuRuntime( const bool debug = true )
+      : m_debug( debug ) { setUp( m_debug ); }
+    ~GpuRuntime() { tearDown( m_debug ); }
+    GpuRuntime( const GpuRuntime& ) = delete;
+    GpuRuntime( GpuRuntime&& ) = delete;
+    GpuRuntime& operator=( const GpuRuntime& ) = delete;
+    GpuRuntime& operator=( GpuRuntime&& ) = delete;
+    bool m_debug;
+
+    // Set up CUDA application
+    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
+    // Calling cudaSetDevice on startup is useful to properly book-keep the time spent in CUDA initialization
+    static void setUp( const bool debug = true )
+    {
+      // ** NB: it is useful to call cudaSetDevice, or cudaFree, to properly book-keep the time spent in CUDA initialization
+      // ** NB: otherwise, the first CUDA operation (eg a cudaMemcpyToSymbol in CPPProcess ctor) appears to take much longer!
+      /*
+      // [We initially added cudaFree(0) to "ease profile analysis" only because it shows up as a big recognizable block!]
+      // No explicit initialization is needed: https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#initialization
+      // It is not clear what cudaFree(0) does at all: https://stackoverflow.com/questions/69967813/
+      if ( debug ) std::cout << "__CudaRuntime: calling cudaFree(0)" << std::endl;
+      checkCuda( cudaFree( 0 ) ); // SLOW!
+      */
+      // Replace cudaFree(0) by cudaSetDevice(0), even if it is not really needed either
+      // (but see https://developer.nvidia.com/blog/cuda-pro-tip-always-set-current-device-avoid-multithreading-bugs)
+      if( debug ) std::cout << "__GpuRuntime: calling GpuSetDevice(0)" << std::endl;
+      checkGpu( gpuSetDevice( 0 ) ); // SLOW!
+    }
+
+    // Tear down CUDA application (call cudaDeviceReset)
+    // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
+    // Calling cudaDeviceReset on shutdown is only needed for checking memory leaks in cuda-memcheck
+    // See https://docs.nvidia.com/cuda/cuda-memcheck/index.html#leak-checking
+    static void tearDown( const bool debug = true )
+    {
+      if( debug ) std::cout << "__GpuRuntime: calling GpuDeviceReset()" << std::endl;
+      checkGpu( gpuDeviceReset() );
+    }
+  };
+}
+#endif
+
+//--------------------------------------------------------------------------
+
+#endif // MG5AMC_GPURUNTIME_H
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MadgraphTest.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MadgraphTest.h
index ef40624c88..6054185300 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MadgraphTest.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MadgraphTest.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Hageboeck (Dec 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MADGRAPHTEST_H_
 #define MADGRAPHTEST_H_ 1
@@ -14,7 +14,11 @@
 
 #include <array>
 #include <cmath>
-#include <filesystem>
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
 #include <fstream>
 #include <iomanip>
 #include <memory>
@@ -22,7 +26,7 @@
 #include <string>
 #include <vector>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using mg5amcGpu::CPPProcess;
 #else
 using mg5amcCpu::CPPProcess;
@@ -201,7 +205,7 @@ class MadgraphTest : public testing::TestWithParam<TestDriverBase*>
 
 // Since we link both the CPU-only and GPU tests into the same executable, we prevent
 // a multiply defined symbol by only compiling this in the non-CUDA phase:
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
 /// Compare momenta and matrix elements.
 /// This uses an implementation of TestDriverBase to run a madgraph workflow,
@@ -219,7 +223,14 @@ TEST_P( MadgraphTest, CompareMomentaAndME )
   const char* dumpEventsC = getenv( "CUDACPP_RUNTEST_DUMPEVENTS" );
   const bool dumpEvents = ( dumpEventsC != 0 ) && ( std::string( dumpEventsC ) != "" );
   const std::string refFileName = testDriver->getRefFileName();
+  /*
+#ifdef __HIPCC__
+  const std::string dumpFileName = std::experimental::filesystem::path( refFileName ).filename();
+#else
   const std::string dumpFileName = std::filesystem::path( refFileName ).filename();
+#endif
+  */
+  const std::string dumpFileName = refFileName; // bypass std::filesystem #803
   std::ofstream dumpFile;
   if( dumpEvents )
   {
@@ -307,6 +318,6 @@ TEST_P( MadgraphTest, CompareMomentaAndME )
   }
 }
 
-#endif // __CUDACC__
+#endif // MGONGPUCPP_GPUIMPL
 
 #endif /* MADGRAPHTEST_H_ */
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MatrixElementKernels.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MatrixElementKernels.cc
index 74b5239ebf..81699dfea9 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MatrixElementKernels.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MatrixElementKernels.cc
@@ -1,12 +1,12 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "MatrixElementKernels.h"
 
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h" // Includes the abstraction for Nvidia/AMD compilation
 #include "MemoryAccessMomenta.h"
 #include "MemoryBuffers.h"
 
@@ -14,7 +14,7 @@
 
 //============================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu
 {
 
@@ -150,7 +150,7 @@ namespace mg5amcCpu
 
 //============================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 {
 
@@ -209,13 +209,13 @@ namespace mg5amcGpu
     PinnedHostBufferHelicityMask hstIsGoodHel( ncomb );
     DeviceBufferHelicityMask devIsGoodHel( ncomb );
     // ... 0d1. Compute good helicity mask on the device
-    computeDependentCouplings<<<m_gpublocks, m_gputhreads>>>( m_gs.data(), m_couplings.data() );
+    gpuLaunchKernel( computeDependentCouplings, m_gpublocks, m_gputhreads, m_gs.data(), m_couplings.data() );
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-    sigmaKin_getGoodHel<<<m_gpublocks, m_gputhreads>>>( m_momenta.data(), m_couplings.data(), m_matrixElements.data(), m_numerators.data(), m_denominators.data(), devIsGoodHel.data() );
+    gpuLaunchKernel( sigmaKin_getGoodHel, m_gpublocks, m_gputhreads, m_momenta.data(), m_couplings.data(), m_matrixElements.data(), m_numerators.data(), m_denominators.data(), devIsGoodHel.data() );
 #else
-    sigmaKin_getGoodHel<<<m_gpublocks, m_gputhreads>>>( m_momenta.data(), m_couplings.data(), m_matrixElements.data(), devIsGoodHel.data() );
+    gpuLaunchKernel( sigmaKin_getGoodHel, m_gpublocks, m_gputhreads, m_momenta.data(), m_couplings.data(), m_matrixElements.data(), devIsGoodHel.data() );
 #endif
-    checkCuda( cudaPeekAtLastError() );
+    checkGpu( gpuPeekAtLastError() );
     // ... 0d2. Copy back good helicity mask to the host
     copyHostFromDevice( hstIsGoodHel, devIsGoodHel );
     // ... 0d3. Copy back good helicity list to constant memory on the device
@@ -226,19 +226,19 @@ namespace mg5amcGpu
 
   void MatrixElementKernelDevice::computeMatrixElements( const unsigned int channelId )
   {
-    computeDependentCouplings<<<m_gpublocks, m_gputhreads>>>( m_gs.data(), m_couplings.data() );
+    gpuLaunchKernel( computeDependentCouplings, m_gpublocks, m_gputhreads, m_gs.data(), m_couplings.data() );
 #ifndef MGONGPU_NSIGHT_DEBUG
     constexpr unsigned int sharedMemSize = 0;
 #else
     constexpr unsigned int sharedMemSize = ntpbMAX * sizeof( float );
 #endif
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-    sigmaKin<<<m_gpublocks, m_gputhreads, sharedMemSize>>>( m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), channelId, m_numerators.data(), m_denominators.data(), m_selhel.data(), m_selcol.data() );
+    gpuLaunchKernelSharedMem( sigmaKin, m_gpublocks, m_gputhreads, sharedMemSize, m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), channelId, m_numerators.data(), m_denominators.data(), m_selhel.data(), m_selcol.data() );
 #else
-    sigmaKin<<<m_gpublocks, m_gputhreads, sharedMemSize>>>( m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), m_selhel.data(), m_selcol.data() );
+    gpuLaunchKernelSharedMem( sigmaKin, m_gpublocks, m_gputhreads, sharedMemSize, m_momenta.data(), m_couplings.data(), m_rndhel.data(), m_rndcol.data(), m_matrixElements.data(), m_selhel.data(), m_selcol.data() );
 #endif
-    checkCuda( cudaPeekAtLastError() );
-    checkCuda( cudaDeviceSynchronize() );
+    checkGpu( gpuPeekAtLastError() );
+    checkGpu( gpuDeviceSynchronize() );
   }
 
   //--------------------------------------------------------------------------
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MatrixElementKernels.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MatrixElementKernels.h
index 23e84757a2..72bd8f195b 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MatrixElementKernels.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MatrixElementKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MATRIXELEMENTKERNELS_H
 #define MATRIXELEMENTKERNELS_H 1
@@ -10,7 +10,7 @@
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -81,7 +81,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating matrix element calculations on a CPU host
   class MatrixElementKernelHost final : public MatrixElementKernelBase, public NumberOfEvents
   {
@@ -130,7 +130,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating matrix element calculations on a GPU device
   class MatrixElementKernelDevice : public MatrixElementKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessAmplitudes.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessAmplitudes.h
index 573b3bbbc9..ffb76e93de 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessAmplitudes.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessAmplitudes.h
@@ -15,7 +15,7 @@
 #define MGONGPU_TRIVIAL_AMPLITUDES 1
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessCouplings.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessCouplings.h
index 35a3af42e0..3afdf3e554 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessCouplings.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessCouplings.h
@@ -15,7 +15,7 @@
 #include "MemoryBuffers.h"       // for HostBufferCouplings::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessCouplingsFixed.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessCouplingsFixed.h
index dc0d93afff..ffcdf4dbef 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessCouplingsFixed.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessCouplingsFixed.h
@@ -14,7 +14,7 @@
 //#include "MemoryAccessHelpers.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessDenominators.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessDenominators.h
index 3bce635718..66f2d32a6b 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessDenominators.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessDenominators.h
@@ -10,7 +10,7 @@
 #include "MemoryAccessGs.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessGs.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessGs.h
index 31311aa375..4c726b30f3 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessGs.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessGs.h
@@ -13,7 +13,7 @@
 #include "MemoryBuffers.h" // for HostBufferMatrixElements::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessHelpers.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessHelpers.h
index c82a6c7635..db73e4e064 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessHelpers.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessHelpers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessHelpers_H
 #define MemoryAccessHelpers_H 1
@@ -105,7 +105,7 @@ class KernelAccessHelper : public MemoryAccessHelper<T>
     }
     else
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
       //printf( "kernelAccessRecord: ievt=%d threadId=%d\n", ievt, threadIdx.x );
       return T::ieventAccessRecord( buffer, ievt ); // NB fptype and fptype_sv coincide for CUDA
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessMatrixElements.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessMatrixElements.h
index f32e6fea5b..3741011971 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessMatrixElements.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessMatrixElements.h
@@ -13,7 +13,7 @@
 #include "MemoryBuffers.h" // for HostBufferMatrixElements::isaligned
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessMomenta.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessMomenta.h
index 29266de32c..3be229d392 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessMomenta.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessMomenta.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessMomenta_H
 #define MemoryAccessMomenta_H 1
@@ -13,7 +13,7 @@
 #include "MemoryAccessVectors.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -30,7 +30,7 @@ namespace mg5amcCpu
 
     // Number of Events Per Page in the momenta AOSOA memory buffer layout
     // (these are all best kept as a compile-time constants: see issue #23)
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
     // -----------------------------------------------------------------------------------------------
     // --- GPUs: neppM is best set to a power of 2 times the number of fptype's in a 32-byte cacheline
     // --- This is relevant to ensure coalesced access to momenta in global memory
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessNumerators.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessNumerators.h
index b152183b28..18991f4fa6 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessNumerators.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessNumerators.h
@@ -10,7 +10,7 @@
 #include "MemoryAccessGs.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessRandomNumbers.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessRandomNumbers.h
index e2988d39f3..40cb089135 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessRandomNumbers.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessRandomNumbers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessRandomNumbers_H
 #define MemoryAccessRandomNumbers_H 1
@@ -11,7 +11,7 @@
 #include "CPPProcess.h"
 #include "MemoryAccessHelpers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using mg5amcGpu::CPPProcess;
 #else
 using mg5amcCpu::CPPProcess;
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessVectors.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessVectors.h
index e9b197368e..08faccff0f 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessVectors.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessVectors.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryAccessVectors_H
 #define MemoryAccessVectors_H 1
@@ -10,7 +10,7 @@
 
 #include "mgOnGpuVectors.h"
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 namespace mg5amcCpu // this is only needed for CPU SIMD vectorization
 {
 
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessWavefunctions.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessWavefunctions.h
index 5428aaf933..33bef4559e 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessWavefunctions.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryAccessWavefunctions.h
@@ -15,7 +15,7 @@
 #define MGONGPU_TRIVIAL_WAVEFUNCTIONS 1
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryBuffers.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryBuffers.h
index 3093e6ed18..7756a71621 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryBuffers.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/MemoryBuffers.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021, based on earlier work by S. Hageboeck) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MemoryBuffers_H
 #define MemoryBuffers_H 1
@@ -11,12 +11,12 @@
 #include "mgOnGpuCxtypes.h"
 
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "Parameters_sm.h"
 
 #include <sstream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -87,7 +87,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   constexpr bool HostBufferALIGNED = false;   // ismisaligned=false
   constexpr bool HostBufferMISALIGNED = true; // ismisaligned=true
 
@@ -119,7 +119,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA pinned host buffer
   template<typename T>
   class PinnedHostBufferBase : public BufferBase<T>
@@ -128,18 +128,18 @@ namespace mg5amcCpu
     PinnedHostBufferBase( const size_t size )
       : BufferBase<T>( size, false )
     {
-      checkCuda( cudaMallocHost( &( this->m_data ), this->bytes() ) );
+      gpuMallocHost( &( this->m_data ), this->bytes() );
     }
     virtual ~PinnedHostBufferBase()
     {
-      checkCuda( cudaFreeHost( this->m_data ) );
+      gpuFreeHost( this->m_data );
     }
   };
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA device buffer
   template<typename T>
   class DeviceBufferBase : public BufferBase<T>
@@ -148,18 +148,18 @@ namespace mg5amcCpu
     DeviceBufferBase( const size_t size )
       : BufferBase<T>( size, true )
     {
-      checkCuda( cudaMalloc( &( this->m_data ), this->bytes() ) );
+      gpuMalloc( &( this->m_data ), this->bytes() );
     }
     virtual ~DeviceBufferBase()
     {
-      checkCuda( cudaFree( this->m_data ) );
+      gpuFree( this->m_data );
     }
   };
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for a given number of events
   template<typename T, size_t sizePerEvent, bool ismisaligned>
   class HostBuffer : public HostBufferBase<T, ismisaligned>, virtual private NumberOfEvents
@@ -175,7 +175,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA pinned host buffer for a given number of events
   template<typename T, size_t sizePerEvent>
   class PinnedHostBuffer : public PinnedHostBufferBase<T>, virtual private NumberOfEvents
@@ -191,7 +191,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating a CUDA device buffer for a given number of events
   template<typename T, size_t sizePerEvent>
   class DeviceBuffer : public DeviceBufferBase<T>, virtual private NumberOfEvents
@@ -213,7 +213,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for momenta random numbers
   constexpr size_t sizePerEventRndNumMomenta = MemoryBuffers::np4 * MemoryBuffers::nparf;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for momenta random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumMomenta, HostBufferALIGNED> HostBufferRndNumMomenta;
 #else
@@ -232,7 +232,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer with ONE fptype per event
   constexpr size_t sizePerEventOneFp = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer with ONE fptype per event
   typedef HostBuffer<fptype, sizePerEventOneFp, HostBufferALIGNED> HostBufferOneFp;
 #else
@@ -257,7 +257,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for Gs
   constexpr size_t sizePerEventGs = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventGs, HostBufferALIGNED> HostBufferGs;
 #else
@@ -276,7 +276,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for numerators
   constexpr size_t sizePerEventNumerators = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventNumerators, HostBufferALIGNED> HostBufferNumerators;
 #else
@@ -296,7 +296,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for denominators
   constexpr size_t sizePerEventDenominators = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventDenominators, HostBufferALIGNED> HostBufferDenominators;
 #else
@@ -315,7 +315,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for random numbers
   constexpr size_t sizePerEventCouplings = MemoryBuffers::ndcoup * MemoryBuffers::nx2;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for gs
   typedef HostBuffer<fptype, sizePerEventCouplings, HostBufferALIGNED> HostBufferCouplings;
 #else
@@ -333,7 +333,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for momenta
   constexpr size_t sizePerEventMomenta = MemoryBuffers::np4 * MemoryBuffers::npar;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for momenta
   typedef HostBuffer<fptype, sizePerEventMomenta, HostBufferALIGNED> HostBufferMomenta;
   //typedef HostBuffer<fptype, sizePerEventMomenta, HostBufferMISALIGNED> HostBufferMomenta; // TEST MISALIGNMENT!
@@ -352,7 +352,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for sampling weights
   constexpr size_t sizePerEventWeights = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for sampling weights
   typedef HostBuffer<fptype, sizePerEventWeights, HostBufferALIGNED> HostBufferWeights;
 #else
@@ -370,7 +370,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for matrix elements
   constexpr size_t sizePerEventMatrixElements = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for matrix elements
   typedef HostBuffer<fptype, sizePerEventMatrixElements, HostBufferALIGNED> HostBufferMatrixElements;
 #else
@@ -385,7 +385,7 @@ namespace mg5amcCpu
   // A base class encapsulating a memory buffer for the helicity mask
   typedef BufferBase<bool> BufferHelicityMask;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for the helicity mask
   typedef HostBufferBase<bool, HostBufferALIGNED> HostBufferHelicityMask;
 #else
@@ -403,7 +403,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for wavefunctions
   constexpr size_t sizePerEventWavefunctions = MemoryBuffers::nw6 * MemoryBuffers::nx2;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for wavefunctions
   typedef HostBuffer<fptype, sizePerEventWavefunctions, HostBufferALIGNED> HostBufferWavefunctions;
 #else
@@ -421,7 +421,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for helicity random numbers
   constexpr size_t sizePerEventRndNumHelicity = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for helicity random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumHelicity, HostBufferALIGNED> HostBufferRndNumHelicity;
 #else
@@ -439,7 +439,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for color random numbers
   constexpr size_t sizePerEventRndNumColor = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for color random numbers
   typedef HostBuffer<fptype, sizePerEventRndNumColor, HostBufferALIGNED> HostBufferRndNumColor;
 #else
@@ -457,7 +457,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for helicity selection
   constexpr size_t sizePerEventSelectedHelicity = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for helicity selection
   typedef HostBuffer<int, sizePerEventSelectedHelicity, HostBufferALIGNED> HostBufferSelectedHelicity;
 #else
@@ -475,7 +475,7 @@ namespace mg5amcCpu
   // The size (number of elements) per event in a memory buffer for color selection
   constexpr size_t sizePerEventSelectedColor = 1;
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // A class encapsulating a C++ host buffer for color selection
   typedef HostBuffer<int, sizePerEventSelectedColor, HostBufferALIGNED> HostBufferSelectedColor;
 #else
@@ -487,7 +487,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<class Tdst, class Tsrc>
   void copyDeviceFromHost( Tdst& dst, const Tsrc& src ) // keep the same order of arguments as in memcpy
   {
@@ -504,13 +504,13 @@ namespace mg5amcCpu
       throw std::runtime_error( sstr.str() );
     }
     // NB (PR #45): cudaMemcpy involves an intermediate memcpy to pinned memory if host array is a not a pinned host array
-    checkCuda( cudaMemcpy( dst.data(), src.data(), src.bytes(), cudaMemcpyHostToDevice ) );
+    gpuMemcpy( dst.data(), src.data(), src.bytes(), gpuMemcpyHostToDevice );
   }
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   template<class Tdst, class Tsrc>
   void copyHostFromDevice( Tdst& dst, const Tsrc& src ) // keep the same order of arguments as in memcpy
   {
@@ -527,7 +527,7 @@ namespace mg5amcCpu
       throw std::runtime_error( sstr.str() );
     }
     // NB (PR #45): cudaMemcpy involves an intermediate memcpy to pinned memory if host array is a not a pinned host array
-    checkCuda( cudaMemcpy( dst.data(), src.data(), src.bytes(), cudaMemcpyDeviceToHost ) );
+    gpuMemcpy( dst.data(), src.data(), src.bytes(), gpuMemcpyDeviceToHost );
   }
 #endif
 
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_gg_ttx/CPPProcess.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_gg_ttx/CPPProcess.cc
index 7f14b5e299..c465192676 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_gg_ttx/CPPProcess.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_gg_ttx/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_sm.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -46,7 +45,7 @@
 // Class member functions for calculating the matrix elements for
 // Process: g g > t t~ WEIGHTED<=2
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -80,7 +79,7 @@ namespace mg5amcCpu
   __device__ const fptype cIPD[2] = { (fptype)Parameters_sm::mdl_MT, (fptype)Parameters_sm::mdl_WT };
   __device__ const fptype* cIPC = nullptr; // unused as nicoup=0
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ fptype cIPD[2];
   __device__ __constant__ fptype* cIPC = nullptr; // unused as nicoup=0
 #else
@@ -90,7 +89,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -118,13 +117,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -151,7 +150,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -187,7 +186,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -200,8 +199,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -302,7 +303,7 @@ namespace mg5amcCpu
         { 16, -2 },
         { -2, 16 } }; // 2-D array[2][2]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -359,7 +360,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -418,7 +419,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -465,8 +466,8 @@ namespace mg5amcCpu
       { 1, 1, -1, -1 },
       { 1, 1, 1, 1 },
       { 1, 1, 1, -1 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -506,9 +507,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     const fptype tIPD[2] = { (fptype)m_pars->mdl_MT, (fptype)m_pars->mdl_WT };
     //const cxtype tIPC[0] = { ... }; // nicoup=0
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) ) );
-    //checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ) ); // nicoup=0
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) );
+    //gpuMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
 #else
     memcpy( cIPD, tIPD, 2 * sizeof( fptype ) );
     //memcpy( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
@@ -544,7 +545,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -573,6 +574,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -586,6 +591,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -609,12 +616,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -635,7 +642,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -761,9 +768,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -787,7 +794,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -807,7 +814,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 256 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -821,9 +828,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -851,7 +861,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -1061,7 +1071,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_gg_ttx/CPPProcess.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_gg_ttx/CPPProcess.h
index 448175be9d..f8a20b77fc 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_gg_ttx/CPPProcess.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_gg_ttx/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -107,7 +107,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -120,7 +120,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -150,7 +150,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_gg_ttx/CudaRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_gg_ttx/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_gg_ttx/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_gg_ttx/GpuAbstraction.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_gg_ttx/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_gg_ttx/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_gg_ttx/GpuRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_gg_ttx/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_gg_ttx/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_gg_ttx/check_sa.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_gg_ttx/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_gg_ttx/check_sa.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_gg_ttx/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_uux_ttx/CPPProcess.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_uux_ttx/CPPProcess.cc
index 20496eaa70..85bdc6bf24 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_uux_ttx/CPPProcess.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_uux_ttx/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_sm.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -49,7 +48,7 @@
 // Process: d d~ > t t~ WEIGHTED<=2
 // Process: s s~ > t t~ WEIGHTED<=2
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +82,7 @@ namespace mg5amcCpu
   __device__ const fptype cIPD[2] = { (fptype)Parameters_sm::mdl_MT, (fptype)Parameters_sm::mdl_WT };
   __device__ const fptype* cIPC = nullptr; // unused as nicoup=0
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ fptype cIPD[2];
   __device__ __constant__ fptype* cIPC = nullptr; // unused as nicoup=0
 #else
@@ -93,7 +92,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -121,13 +120,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -154,7 +153,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -190,7 +189,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -203,8 +202,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -279,7 +280,7 @@ namespace mg5amcCpu
         { 9, 3 },
         { 3, 9 } }; // 2-D array[2][2]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -336,7 +337,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -395,7 +396,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -442,8 +443,8 @@ namespace mg5amcCpu
       { -1, 1, -1, -1 },
       { -1, 1, 1, 1 },
       { -1, 1, 1, -1 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -483,9 +484,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     const fptype tIPD[2] = { (fptype)m_pars->mdl_MT, (fptype)m_pars->mdl_WT };
     //const cxtype tIPC[0] = { ... }; // nicoup=0
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) ) );
-    //checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ) ); // nicoup=0
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) );
+    //gpuMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
 #else
     memcpy( cIPD, tIPD, 2 * sizeof( fptype ) );
     //memcpy( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
@@ -521,7 +522,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -550,6 +551,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -563,6 +568,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -586,12 +593,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -612,7 +619,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -738,9 +745,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -764,7 +771,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -784,7 +791,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 36 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -798,9 +805,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -828,7 +838,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -1038,7 +1048,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_uux_ttx/CPPProcess.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_uux_ttx/CPPProcess.h
index e166fa1652..6498b91441 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_uux_ttx/CPPProcess.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_uux_ttx/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -110,7 +110,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -123,7 +123,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -153,7 +153,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_uux_ttx/CudaRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_uux_ttx/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_uux_ttx/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_uux_ttx/GpuAbstraction.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_uux_ttx/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_uux_ttx/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_uux_ttx/GpuRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_uux_ttx/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_uux_ttx/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_uux_ttx/check_sa.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_uux_ttx/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_uux_ttx/check_sa.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P0_uux_ttx/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gg_ttxg/CPPProcess.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gg_ttxg/CPPProcess.cc
index afeebde3c6..2fa9b4f651 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gg_ttxg/CPPProcess.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gg_ttxg/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_sm.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -46,7 +45,7 @@
 // Class member functions for calculating the matrix elements for
 // Process: g g > t t~ g WEIGHTED<=3 @1
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -80,7 +79,7 @@ namespace mg5amcCpu
   __device__ const fptype cIPD[2] = { (fptype)Parameters_sm::mdl_MT, (fptype)Parameters_sm::mdl_WT };
   __device__ const fptype* cIPC = nullptr; // unused as nicoup=0
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ fptype cIPD[2];
   __device__ __constant__ fptype* cIPC = nullptr; // unused as nicoup=0
 #else
@@ -90,7 +89,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -118,13 +117,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -151,7 +150,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -187,7 +186,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -200,8 +199,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -505,7 +506,7 @@ namespace mg5amcCpu
         { 1, -8, 10, 1, 64, -8 },
         { 10, 1, 1, -8, -8, 64 } }; // 2-D array[6][6]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -562,7 +563,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -621,7 +622,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -684,8 +685,8 @@ namespace mg5amcCpu
       { 1, 1, 1, 1, 1 },
       { 1, 1, 1, -1, -1 },
       { 1, 1, 1, -1, 1 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -726,9 +727,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     const fptype tIPD[2] = { (fptype)m_pars->mdl_MT, (fptype)m_pars->mdl_WT };
     //const cxtype tIPC[0] = { ... }; // nicoup=0
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) ) );
-    //checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ) ); // nicoup=0
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) );
+    //gpuMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
 #else
     memcpy( cIPD, tIPD, 2 * sizeof( fptype ) );
     //memcpy( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
@@ -765,7 +766,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -794,6 +795,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -807,6 +812,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -830,12 +837,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -856,7 +863,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -982,9 +989,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -1008,7 +1015,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -1028,7 +1035,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 256 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -1042,9 +1049,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -1072,7 +1082,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -1282,7 +1292,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gg_ttxg/CPPProcess.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gg_ttxg/CPPProcess.h
index 37d6ebe981..11f562273e 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gg_ttxg/CPPProcess.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gg_ttxg/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -107,7 +107,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -120,7 +120,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -150,7 +150,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gg_ttxg/CudaRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gg_ttxg/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gg_ttxg/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gg_ttxg/GpuAbstraction.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gg_ttxg/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gg_ttxg/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gg_ttxg/GpuRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gg_ttxg/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gg_ttxg/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gg_ttxg/check_sa.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gg_ttxg/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gg_ttxg/check_sa.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gg_ttxg/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gu_ttxu/CPPProcess.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gu_ttxu/CPPProcess.cc
index b7e3475679..f505a0d8c0 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gu_ttxu/CPPProcess.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gu_ttxu/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_sm.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -49,7 +48,7 @@
 // Process: g d > t t~ d WEIGHTED<=3 @1
 // Process: g s > t t~ s WEIGHTED<=3 @1
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +82,7 @@ namespace mg5amcCpu
   __device__ const fptype cIPD[2] = { (fptype)Parameters_sm::mdl_MT, (fptype)Parameters_sm::mdl_WT };
   __device__ const fptype* cIPC = nullptr; // unused as nicoup=0
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ fptype cIPD[2];
   __device__ __constant__ fptype* cIPC = nullptr; // unused as nicoup=0
 #else
@@ -93,7 +92,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -121,13 +120,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -154,7 +153,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -190,7 +189,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -203,8 +202,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -341,7 +342,7 @@ namespace mg5amcCpu
         { 4, 0, 12, 4 },
         { 0, 4, 4, 12 } }; // 2-D array[4][4]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -398,7 +399,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -457,7 +458,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -520,8 +521,8 @@ namespace mg5amcCpu
       { 1, -1, 1, 1, 1 },
       { 1, -1, 1, -1, -1 },
       { 1, -1, 1, -1, 1 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -562,9 +563,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     const fptype tIPD[2] = { (fptype)m_pars->mdl_MT, (fptype)m_pars->mdl_WT };
     //const cxtype tIPC[0] = { ... }; // nicoup=0
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) ) );
-    //checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ) ); // nicoup=0
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) );
+    //gpuMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
 #else
     memcpy( cIPD, tIPD, 2 * sizeof( fptype ) );
     //memcpy( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
@@ -601,7 +602,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -630,6 +631,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -643,6 +648,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -666,12 +673,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -692,7 +699,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -818,9 +825,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -844,7 +851,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -864,7 +871,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 96 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -878,9 +885,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -908,7 +918,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -1118,7 +1128,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gu_ttxu/CPPProcess.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gu_ttxu/CPPProcess.h
index bf037c6c28..ce22572055 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gu_ttxu/CPPProcess.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gu_ttxu/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -110,7 +110,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -123,7 +123,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -153,7 +153,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gu_ttxu/CudaRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gu_ttxu/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gu_ttxu/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gu_ttxu/GpuAbstraction.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gu_ttxu/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gu_ttxu/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gu_ttxu/GpuRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gu_ttxu/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gu_ttxu/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gu_ttxu/check_sa.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gu_ttxu/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gu_ttxu/check_sa.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gu_ttxu/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gux_ttxux/CPPProcess.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gux_ttxux/CPPProcess.cc
index 0f999663da..9cba1a6d8c 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gux_ttxux/CPPProcess.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gux_ttxux/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_sm.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -49,7 +48,7 @@
 // Process: g d~ > t t~ d~ WEIGHTED<=3 @1
 // Process: g s~ > t t~ s~ WEIGHTED<=3 @1
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +82,7 @@ namespace mg5amcCpu
   __device__ const fptype cIPD[2] = { (fptype)Parameters_sm::mdl_MT, (fptype)Parameters_sm::mdl_WT };
   __device__ const fptype* cIPC = nullptr; // unused as nicoup=0
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ fptype cIPD[2];
   __device__ __constant__ fptype* cIPC = nullptr; // unused as nicoup=0
 #else
@@ -93,7 +92,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -121,13 +120,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -154,7 +153,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -190,7 +189,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -203,8 +202,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -341,7 +342,7 @@ namespace mg5amcCpu
         { 4, 0, 12, 4 },
         { 0, 4, 4, 12 } }; // 2-D array[4][4]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -398,7 +399,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -457,7 +458,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -520,8 +521,8 @@ namespace mg5amcCpu
       { 1, 1, 1, 1, -1 },
       { 1, 1, 1, -1, 1 },
       { 1, 1, 1, -1, -1 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -562,9 +563,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     const fptype tIPD[2] = { (fptype)m_pars->mdl_MT, (fptype)m_pars->mdl_WT };
     //const cxtype tIPC[0] = { ... }; // nicoup=0
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) ) );
-    //checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ) ); // nicoup=0
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) );
+    //gpuMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
 #else
     memcpy( cIPD, tIPD, 2 * sizeof( fptype ) );
     //memcpy( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
@@ -601,7 +602,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -630,6 +631,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -643,6 +648,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -666,12 +673,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -692,7 +699,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -818,9 +825,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -844,7 +851,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -864,7 +871,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 96 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -878,9 +885,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -908,7 +918,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -1118,7 +1128,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gux_ttxux/CPPProcess.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gux_ttxux/CPPProcess.h
index 0f49f5247b..46c4347506 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gux_ttxux/CPPProcess.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gux_ttxux/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -110,7 +110,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -123,7 +123,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -153,7 +153,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gux_ttxux/CudaRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gux_ttxux/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gux_ttxux/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gux_ttxux/GpuAbstraction.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gux_ttxux/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gux_ttxux/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gux_ttxux/GpuRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gux_ttxux/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gux_ttxux/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gux_ttxux/check_sa.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gux_ttxux/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gux_ttxux/check_sa.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_gux_ttxux/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_uux_ttxg/CPPProcess.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_uux_ttxg/CPPProcess.cc
index 87830582d7..222171c5cc 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_uux_ttxg/CPPProcess.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_uux_ttxg/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_sm.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -49,7 +48,7 @@
 // Process: d d~ > t t~ g WEIGHTED<=3 @1
 // Process: s s~ > t t~ g WEIGHTED<=3 @1
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +82,7 @@ namespace mg5amcCpu
   __device__ const fptype cIPD[2] = { (fptype)Parameters_sm::mdl_MT, (fptype)Parameters_sm::mdl_WT };
   __device__ const fptype* cIPC = nullptr; // unused as nicoup=0
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ fptype cIPD[2];
   __device__ __constant__ fptype* cIPC = nullptr; // unused as nicoup=0
 #else
@@ -93,7 +92,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -121,13 +120,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -154,7 +153,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -190,7 +189,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -203,8 +202,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -341,7 +342,7 @@ namespace mg5amcCpu
         { 4, 0, 12, 4 },
         { 0, 4, 4, 12 } }; // 2-D array[4][4]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -398,7 +399,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -457,7 +458,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -520,8 +521,8 @@ namespace mg5amcCpu
       { -1, 1, 1, 1, 1 },
       { -1, 1, 1, -1, -1 },
       { -1, 1, 1, -1, 1 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -562,9 +563,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     const fptype tIPD[2] = { (fptype)m_pars->mdl_MT, (fptype)m_pars->mdl_WT };
     //const cxtype tIPC[0] = { ... }; // nicoup=0
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) ) );
-    //checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ) ); // nicoup=0
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) );
+    //gpuMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
 #else
     memcpy( cIPD, tIPD, 2 * sizeof( fptype ) );
     //memcpy( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
@@ -601,7 +602,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -630,6 +631,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -643,6 +648,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -666,12 +673,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -692,7 +699,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -818,9 +825,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -844,7 +851,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -864,7 +871,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 36 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -878,9 +885,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -908,7 +918,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -1118,7 +1128,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_uux_ttxg/CPPProcess.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_uux_ttxg/CPPProcess.h
index f8bdb38aee..fc7c0d8196 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_uux_ttxg/CPPProcess.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_uux_ttxg/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -110,7 +110,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -123,7 +123,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -153,7 +153,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_uux_ttxg/CudaRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_uux_ttxg/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_uux_ttxg/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_uux_ttxg/GpuAbstraction.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_uux_ttxg/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_uux_ttxg/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_uux_ttxg/GpuRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_uux_ttxg/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_uux_ttxg/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_uux_ttxg/check_sa.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_uux_ttxg/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_uux_ttxg/check_sa.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P1_uux_ttxg/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxgg/CPPProcess.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxgg/CPPProcess.cc
index 9051b3108d..c374ce3189 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxgg/CPPProcess.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxgg/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_sm.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -46,7 +45,7 @@
 // Class member functions for calculating the matrix elements for
 // Process: g g > t t~ g g WEIGHTED<=4 @2
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -80,7 +79,7 @@ namespace mg5amcCpu
   __device__ const fptype cIPD[2] = { (fptype)Parameters_sm::mdl_MT, (fptype)Parameters_sm::mdl_WT };
   __device__ const fptype* cIPC = nullptr; // unused as nicoup=0
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ fptype cIPD[2];
   __device__ __constant__ fptype* cIPC = nullptr; // unused as nicoup=0
 #else
@@ -90,7 +89,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -118,13 +117,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -151,7 +150,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -187,7 +186,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -200,8 +199,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -2417,7 +2418,7 @@ namespace mg5amcCpu
         { 62, 71, -10, 80, -1, 8, -28, 62, 62, -10, -10, -1, -1, 8, -10, -1, -64, 8, 8, -64, 80, 8, 512, -64 },
         { -28, 62, 62, -10, -10, -1, 62, 71, -10, 80, -1, 8, -10, -1, -1, 8, 8, -64, 80, 8, 8, -64, -64, 512 } }; // 2-D array[24][24]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -2474,7 +2475,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -2533,7 +2534,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -2628,8 +2629,8 @@ namespace mg5amcCpu
       { 1, 1, 1, -1, -1, 1 },
       { 1, 1, 1, -1, 1, -1 },
       { 1, 1, 1, -1, 1, 1 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -2671,9 +2672,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     const fptype tIPD[2] = { (fptype)m_pars->mdl_MT, (fptype)m_pars->mdl_WT };
     //const cxtype tIPC[0] = { ... }; // nicoup=0
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) ) );
-    //checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ) ); // nicoup=0
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) );
+    //gpuMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
 #else
     memcpy( cIPD, tIPD, 2 * sizeof( fptype ) );
     //memcpy( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
@@ -2711,7 +2712,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -2740,6 +2741,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -2753,6 +2758,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -2776,12 +2783,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -2802,7 +2809,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -2928,9 +2935,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -2954,7 +2961,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -2974,7 +2981,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 512 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -2988,9 +2995,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -3018,7 +3028,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -3228,7 +3238,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxgg/CPPProcess.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxgg/CPPProcess.h
index 9f43559181..511b053c2a 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxgg/CPPProcess.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxgg/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -107,7 +107,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -120,7 +120,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -150,7 +150,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxgg/CudaRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxgg/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxgg/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxgg/GpuAbstraction.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxgg/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxgg/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxgg/GpuRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxgg/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxgg/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxgg/check_sa.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxgg/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxgg/check_sa.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxgg/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxuux/CPPProcess.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxuux/CPPProcess.cc
index 866433ae8b..6e93814e8e 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxuux/CPPProcess.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxuux/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_sm.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -49,7 +48,7 @@
 // Process: g g > t t~ d d~ WEIGHTED<=4 @2
 // Process: g g > t t~ s s~ WEIGHTED<=4 @2
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +82,7 @@ namespace mg5amcCpu
   __device__ const fptype cIPD[2] = { (fptype)Parameters_sm::mdl_MT, (fptype)Parameters_sm::mdl_WT };
   __device__ const fptype* cIPC = nullptr; // unused as nicoup=0
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ fptype cIPD[2];
   __device__ __constant__ fptype* cIPC = nullptr; // unused as nicoup=0
 #else
@@ -93,7 +92,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -121,13 +120,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -154,7 +153,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -190,7 +189,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -203,8 +202,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -812,7 +813,7 @@ namespace mg5amcCpu
         { -2, 6, -6, -2, 16, 0, 0, -2, 16, 6, 48, 16 },
         { 6, -2, -2, -6, 0, 16, -2, 0, 6, 16, 16, 48 } }; // 2-D array[12][12]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -869,7 +870,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -928,7 +929,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -1023,8 +1024,8 @@ namespace mg5amcCpu
       { 1, 1, 1, -1, -1, -1 },
       { 1, 1, 1, -1, 1, 1 },
       { 1, 1, 1, -1, 1, -1 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -1066,9 +1067,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     const fptype tIPD[2] = { (fptype)m_pars->mdl_MT, (fptype)m_pars->mdl_WT };
     //const cxtype tIPC[0] = { ... }; // nicoup=0
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) ) );
-    //checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ) ); // nicoup=0
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) );
+    //gpuMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
 #else
     memcpy( cIPD, tIPD, 2 * sizeof( fptype ) );
     //memcpy( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
@@ -1106,7 +1107,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -1135,6 +1136,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -1148,6 +1153,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -1171,12 +1178,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -1197,7 +1204,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -1323,9 +1330,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -1349,7 +1356,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -1369,7 +1376,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 256 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -1383,9 +1390,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -1413,7 +1423,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -1623,7 +1633,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxuux/CPPProcess.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxuux/CPPProcess.h
index f26b60c5bb..c411623fc8 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxuux/CPPProcess.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxuux/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -110,7 +110,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -123,7 +123,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -153,7 +153,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxuux/CudaRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxuux/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxuux/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxuux/GpuAbstraction.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxuux/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxuux/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxuux/GpuRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxuux/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxuux/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxuux/check_sa.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxuux/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxuux/check_sa.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gg_ttxuux/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gu_ttxgu/CPPProcess.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gu_ttxgu/CPPProcess.cc
index 1be98364ee..ed55777a81 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gu_ttxgu/CPPProcess.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gu_ttxgu/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_sm.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -49,7 +48,7 @@
 // Process: g d > t t~ g d WEIGHTED<=4 @2
 // Process: g s > t t~ g s WEIGHTED<=4 @2
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +82,7 @@ namespace mg5amcCpu
   __device__ const fptype cIPD[2] = { (fptype)Parameters_sm::mdl_MT, (fptype)Parameters_sm::mdl_WT };
   __device__ const fptype* cIPC = nullptr; // unused as nicoup=0
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ fptype cIPD[2];
   __device__ __constant__ fptype* cIPC = nullptr; // unused as nicoup=0
 #else
@@ -93,7 +92,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -121,13 +120,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -154,7 +153,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -190,7 +189,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -203,8 +202,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -812,7 +813,7 @@ namespace mg5amcCpu
         { 0, -2, -6, -2, -2, 6, 16, 0, 6, 16, 48, 16 },
         { -2, 0, -2, -6, 6, -2, 0, 16, 16, 6, 16, 48 } }; // 2-D array[12][12]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -869,7 +870,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -928,7 +929,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -1023,8 +1024,8 @@ namespace mg5amcCpu
       { 1, -1, 1, -1, -1, 1 },
       { 1, -1, 1, -1, 1, -1 },
       { 1, -1, 1, -1, 1, 1 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -1066,9 +1067,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     const fptype tIPD[2] = { (fptype)m_pars->mdl_MT, (fptype)m_pars->mdl_WT };
     //const cxtype tIPC[0] = { ... }; // nicoup=0
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) ) );
-    //checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ) ); // nicoup=0
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) );
+    //gpuMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
 #else
     memcpy( cIPD, tIPD, 2 * sizeof( fptype ) );
     //memcpy( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
@@ -1106,7 +1107,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -1135,6 +1136,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -1148,6 +1153,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -1171,12 +1178,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -1197,7 +1204,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -1323,9 +1330,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -1349,7 +1356,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -1369,7 +1376,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 96 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -1383,9 +1390,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -1413,7 +1423,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -1623,7 +1633,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gu_ttxgu/CPPProcess.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gu_ttxgu/CPPProcess.h
index 853175b477..9c820a5ddb 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gu_ttxgu/CPPProcess.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gu_ttxgu/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -110,7 +110,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -123,7 +123,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -153,7 +153,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gu_ttxgu/CudaRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gu_ttxgu/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gu_ttxgu/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gu_ttxgu/GpuAbstraction.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gu_ttxgu/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gu_ttxgu/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gu_ttxgu/GpuRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gu_ttxgu/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gu_ttxgu/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gu_ttxgu/check_sa.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gu_ttxgu/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gu_ttxgu/check_sa.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gu_ttxgu/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gux_ttxgux/CPPProcess.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gux_ttxgux/CPPProcess.cc
index dfb05016f5..1e9d03033f 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gux_ttxgux/CPPProcess.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gux_ttxgux/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_sm.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -49,7 +48,7 @@
 // Process: g d~ > t t~ g d~ WEIGHTED<=4 @2
 // Process: g s~ > t t~ g s~ WEIGHTED<=4 @2
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +82,7 @@ namespace mg5amcCpu
   __device__ const fptype cIPD[2] = { (fptype)Parameters_sm::mdl_MT, (fptype)Parameters_sm::mdl_WT };
   __device__ const fptype* cIPC = nullptr; // unused as nicoup=0
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ fptype cIPD[2];
   __device__ __constant__ fptype* cIPC = nullptr; // unused as nicoup=0
 #else
@@ -93,7 +92,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -121,13 +120,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -154,7 +153,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -190,7 +189,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -203,8 +202,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -812,7 +813,7 @@ namespace mg5amcCpu
         { -2, 0, 0, 16, -2, -6, 6, -2, 16, 6, 48, 16 },
         { 0, -2, 16, 0, -6, -2, -2, 6, 6, 16, 16, 48 } }; // 2-D array[12][12]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -869,7 +870,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -928,7 +929,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -1023,8 +1024,8 @@ namespace mg5amcCpu
       { 1, 1, 1, -1, -1, -1 },
       { 1, 1, 1, -1, 1, 1 },
       { 1, 1, 1, -1, 1, -1 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -1066,9 +1067,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     const fptype tIPD[2] = { (fptype)m_pars->mdl_MT, (fptype)m_pars->mdl_WT };
     //const cxtype tIPC[0] = { ... }; // nicoup=0
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) ) );
-    //checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ) ); // nicoup=0
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) );
+    //gpuMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
 #else
     memcpy( cIPD, tIPD, 2 * sizeof( fptype ) );
     //memcpy( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
@@ -1106,7 +1107,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -1135,6 +1136,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -1148,6 +1153,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -1171,12 +1178,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -1197,7 +1204,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -1323,9 +1330,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -1349,7 +1356,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -1369,7 +1376,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 96 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -1383,9 +1390,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -1413,7 +1423,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -1623,7 +1633,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gux_ttxgux/CPPProcess.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gux_ttxgux/CPPProcess.h
index e60cb5b6d7..a5a285b22d 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gux_ttxgux/CPPProcess.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gux_ttxgux/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -110,7 +110,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -123,7 +123,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -153,7 +153,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gux_ttxgux/CudaRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gux_ttxgux/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gux_ttxgux/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gux_ttxgux/GpuAbstraction.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gux_ttxgux/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gux_ttxgux/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gux_ttxgux/GpuRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gux_ttxgux/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gux_ttxgux/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gux_ttxgux/check_sa.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gux_ttxgux/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gux_ttxgux/check_sa.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_gux_ttxgux/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uc_ttxuc/CPPProcess.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uc_ttxuc/CPPProcess.cc
index ecef3e57ca..75930b65cd 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uc_ttxuc/CPPProcess.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uc_ttxuc/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_sm.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -51,7 +50,7 @@
 // Process: c s > t t~ c s WEIGHTED<=4 @2
 // Process: d s > t t~ d s WEIGHTED<=4 @2
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -85,7 +84,7 @@ namespace mg5amcCpu
   __device__ const fptype cIPD[2] = { (fptype)Parameters_sm::mdl_MT, (fptype)Parameters_sm::mdl_WT };
   __device__ const fptype* cIPC = nullptr; // unused as nicoup=0
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ fptype cIPD[2];
   __device__ __constant__ fptype* cIPC = nullptr; // unused as nicoup=0
 #else
@@ -95,7 +94,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -123,13 +122,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -156,7 +155,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -192,7 +191,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -205,8 +204,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -387,7 +388,7 @@ namespace mg5amcCpu
         { 3, 9, 9, 3, 27, 9 },
         { 9, 3, 3, 9, 9, 27 } }; // 2-D array[6][6]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -444,7 +445,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -503,7 +504,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -598,8 +599,8 @@ namespace mg5amcCpu
       { -1, -1, 1, -1, -1, 1 },
       { -1, -1, 1, -1, 1, -1 },
       { -1, -1, 1, -1, 1, 1 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -641,9 +642,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     const fptype tIPD[2] = { (fptype)m_pars->mdl_MT, (fptype)m_pars->mdl_WT };
     //const cxtype tIPC[0] = { ... }; // nicoup=0
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) ) );
-    //checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ) ); // nicoup=0
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) );
+    //gpuMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
 #else
     memcpy( cIPD, tIPD, 2 * sizeof( fptype ) );
     //memcpy( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
@@ -681,7 +682,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -710,6 +711,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -723,6 +728,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -746,12 +753,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -772,7 +779,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -898,9 +905,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -924,7 +931,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -944,7 +951,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 36 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -958,9 +965,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -988,7 +998,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -1198,7 +1208,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uc_ttxuc/CPPProcess.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uc_ttxuc/CPPProcess.h
index 5329710b87..8c84687f8a 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uc_ttxuc/CPPProcess.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uc_ttxuc/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -112,7 +112,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -125,7 +125,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -155,7 +155,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uc_ttxuc/CudaRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uc_ttxuc/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uc_ttxuc/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uc_ttxuc/GpuAbstraction.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uc_ttxuc/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uc_ttxuc/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uc_ttxuc/GpuRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uc_ttxuc/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uc_ttxuc/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uc_ttxuc/check_sa.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uc_ttxuc/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uc_ttxuc/check_sa.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uc_ttxuc/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_ucx_ttxucx/CPPProcess.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_ucx_ttxucx/CPPProcess.cc
index e4f9dee3a2..b12263362e 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_ucx_ttxucx/CPPProcess.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_ucx_ttxucx/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_sm.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -57,7 +56,7 @@
 // Process: s c~ > t t~ s c~ WEIGHTED<=4 @2
 // Process: s d~ > t t~ s d~ WEIGHTED<=4 @2
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -91,7 +90,7 @@ namespace mg5amcCpu
   __device__ const fptype cIPD[2] = { (fptype)Parameters_sm::mdl_MT, (fptype)Parameters_sm::mdl_WT };
   __device__ const fptype* cIPC = nullptr; // unused as nicoup=0
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ fptype cIPD[2];
   __device__ __constant__ fptype* cIPC = nullptr; // unused as nicoup=0
 #else
@@ -101,7 +100,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -129,13 +128,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -162,7 +161,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -198,7 +197,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -211,8 +210,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -393,7 +394,7 @@ namespace mg5amcCpu
         { 3, 9, 9, 3, 27, 9 },
         { 9, 3, 3, 9, 9, 27 } }; // 2-D array[6][6]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -450,7 +451,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -509,7 +510,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -604,8 +605,8 @@ namespace mg5amcCpu
       { -1, 1, 1, -1, -1, -1 },
       { -1, 1, 1, -1, 1, 1 },
       { -1, 1, 1, -1, 1, -1 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -647,9 +648,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     const fptype tIPD[2] = { (fptype)m_pars->mdl_MT, (fptype)m_pars->mdl_WT };
     //const cxtype tIPC[0] = { ... }; // nicoup=0
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) ) );
-    //checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ) ); // nicoup=0
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) );
+    //gpuMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
 #else
     memcpy( cIPD, tIPD, 2 * sizeof( fptype ) );
     //memcpy( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
@@ -687,7 +688,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -716,6 +717,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -729,6 +734,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -752,12 +759,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -778,7 +785,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -904,9 +911,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -930,7 +937,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -950,7 +957,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 36 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -964,9 +971,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -994,7 +1004,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -1204,7 +1214,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_ucx_ttxucx/CPPProcess.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_ucx_ttxucx/CPPProcess.h
index 391789dc81..da747c3465 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_ucx_ttxucx/CPPProcess.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_ucx_ttxucx/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -118,7 +118,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -131,7 +131,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -161,7 +161,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_ucx_ttxucx/CudaRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_ucx_ttxucx/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_ucx_ttxucx/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_ucx_ttxucx/GpuAbstraction.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_ucx_ttxucx/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_ucx_ttxucx/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_ucx_ttxucx/GpuRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_ucx_ttxucx/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_ucx_ttxucx/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_ucx_ttxucx/check_sa.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_ucx_ttxucx/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_ucx_ttxucx/check_sa.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_ucx_ttxucx/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uu_ttxuu/CPPProcess.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uu_ttxuu/CPPProcess.cc
index 302d63e31d..d55c7270e6 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uu_ttxuu/CPPProcess.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uu_ttxuu/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_sm.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -49,7 +48,7 @@
 // Process: d d > t t~ d d WEIGHTED<=4 @2
 // Process: s s > t t~ s s WEIGHTED<=4 @2
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +82,7 @@ namespace mg5amcCpu
   __device__ const fptype cIPD[2] = { (fptype)Parameters_sm::mdl_MT, (fptype)Parameters_sm::mdl_WT };
   __device__ const fptype* cIPC = nullptr; // unused as nicoup=0
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ fptype cIPD[2];
   __device__ __constant__ fptype* cIPC = nullptr; // unused as nicoup=0
 #else
@@ -93,7 +92,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -121,13 +120,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -154,7 +153,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -190,7 +189,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -203,8 +202,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -497,7 +498,7 @@ namespace mg5amcCpu
         { 3, 9, 9, 3, 27, 9 },
         { 9, 3, 3, 9, 9, 27 } }; // 2-D array[6][6]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -554,7 +555,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -613,7 +614,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -708,8 +709,8 @@ namespace mg5amcCpu
       { -1, -1, 1, -1, -1, 1 },
       { -1, -1, 1, -1, 1, -1 },
       { -1, -1, 1, -1, 1, 1 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -751,9 +752,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     const fptype tIPD[2] = { (fptype)m_pars->mdl_MT, (fptype)m_pars->mdl_WT };
     //const cxtype tIPC[0] = { ... }; // nicoup=0
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) ) );
-    //checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ) ); // nicoup=0
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) );
+    //gpuMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
 #else
     memcpy( cIPD, tIPD, 2 * sizeof( fptype ) );
     //memcpy( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
@@ -791,7 +792,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -820,6 +821,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -833,6 +838,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -856,12 +863,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -882,7 +889,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -1008,9 +1015,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -1034,7 +1041,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -1054,7 +1061,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 72 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -1068,9 +1075,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -1098,7 +1108,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -1308,7 +1318,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uu_ttxuu/CPPProcess.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uu_ttxuu/CPPProcess.h
index 2d95f4b170..d8232ea652 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uu_ttxuu/CPPProcess.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uu_ttxuu/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -110,7 +110,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -123,7 +123,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -153,7 +153,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uu_ttxuu/CudaRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uu_ttxuu/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uu_ttxuu/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uu_ttxuu/GpuAbstraction.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uu_ttxuu/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uu_ttxuu/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uu_ttxuu/GpuRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uu_ttxuu/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uu_ttxuu/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uu_ttxuu/check_sa.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uu_ttxuu/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uu_ttxuu/check_sa.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uu_ttxuu/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxccx/CPPProcess.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxccx/CPPProcess.cc
index d0be5131af..ed38fae08c 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxccx/CPPProcess.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxccx/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_sm.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -57,7 +56,7 @@
 // Process: s s~ > t t~ c c~ WEIGHTED<=4 @2
 // Process: s s~ > t t~ d d~ WEIGHTED<=4 @2
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -91,7 +90,7 @@ namespace mg5amcCpu
   __device__ const fptype cIPD[2] = { (fptype)Parameters_sm::mdl_MT, (fptype)Parameters_sm::mdl_WT };
   __device__ const fptype* cIPC = nullptr; // unused as nicoup=0
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ fptype cIPD[2];
   __device__ __constant__ fptype* cIPC = nullptr; // unused as nicoup=0
 #else
@@ -101,7 +100,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -129,13 +128,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -162,7 +161,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -198,7 +197,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -211,8 +210,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -393,7 +394,7 @@ namespace mg5amcCpu
         { 3, 9, 9, 3, 27, 9 },
         { 9, 3, 3, 9, 9, 27 } }; // 2-D array[6][6]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -450,7 +451,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -509,7 +510,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -604,8 +605,8 @@ namespace mg5amcCpu
       { -1, 1, 1, -1, -1, -1 },
       { -1, 1, 1, -1, 1, 1 },
       { -1, 1, 1, -1, 1, -1 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -647,9 +648,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     const fptype tIPD[2] = { (fptype)m_pars->mdl_MT, (fptype)m_pars->mdl_WT };
     //const cxtype tIPC[0] = { ... }; // nicoup=0
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) ) );
-    //checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ) ); // nicoup=0
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) );
+    //gpuMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
 #else
     memcpy( cIPD, tIPD, 2 * sizeof( fptype ) );
     //memcpy( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
@@ -687,7 +688,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -716,6 +717,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -729,6 +734,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -752,12 +759,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -778,7 +785,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -904,9 +911,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -930,7 +937,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -950,7 +957,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 36 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -964,9 +971,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -994,7 +1004,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -1204,7 +1214,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxccx/CPPProcess.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxccx/CPPProcess.h
index 14490d782f..71fdc6e547 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxccx/CPPProcess.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxccx/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -118,7 +118,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -131,7 +131,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -161,7 +161,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxccx/CudaRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxccx/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxccx/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxccx/GpuAbstraction.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxccx/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxccx/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxccx/GpuRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxccx/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxccx/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxccx/check_sa.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxccx/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxccx/check_sa.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxccx/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxgg/CPPProcess.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxgg/CPPProcess.cc
index 3a2178d534..b127281504 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxgg/CPPProcess.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxgg/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_sm.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -49,7 +48,7 @@
 // Process: d d~ > t t~ g g WEIGHTED<=4 @2
 // Process: s s~ > t t~ g g WEIGHTED<=4 @2
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +82,7 @@ namespace mg5amcCpu
   __device__ const fptype cIPD[2] = { (fptype)Parameters_sm::mdl_MT, (fptype)Parameters_sm::mdl_WT };
   __device__ const fptype* cIPC = nullptr; // unused as nicoup=0
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ fptype cIPD[2];
   __device__ __constant__ fptype* cIPC = nullptr; // unused as nicoup=0
 #else
@@ -93,7 +92,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -121,13 +120,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -154,7 +153,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -190,7 +189,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -203,8 +202,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -812,7 +813,7 @@ namespace mg5amcCpu
         { 16, -2, 0, 0, 0, 0, -2, 16, 16, 6, 48, 16 },
         { 0, 0, 16, -2, -2, 16, 0, 0, 6, 16, 16, 48 } }; // 2-D array[12][12]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -869,7 +870,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -928,7 +929,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -1023,8 +1024,8 @@ namespace mg5amcCpu
       { -1, 1, 1, -1, -1, 1 },
       { -1, 1, 1, -1, 1, -1 },
       { -1, 1, 1, -1, 1, 1 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -1066,9 +1067,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     const fptype tIPD[2] = { (fptype)m_pars->mdl_MT, (fptype)m_pars->mdl_WT };
     //const cxtype tIPC[0] = { ... }; // nicoup=0
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) ) );
-    //checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ) ); // nicoup=0
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) );
+    //gpuMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
 #else
     memcpy( cIPD, tIPD, 2 * sizeof( fptype ) );
     //memcpy( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
@@ -1106,7 +1107,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -1135,6 +1136,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -1148,6 +1153,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -1171,12 +1178,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -1197,7 +1204,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -1323,9 +1330,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -1349,7 +1356,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -1369,7 +1376,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 72 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -1383,9 +1390,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -1413,7 +1423,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -1623,7 +1633,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxgg/CPPProcess.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxgg/CPPProcess.h
index 1543c29649..e9a24f516d 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxgg/CPPProcess.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxgg/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -110,7 +110,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -123,7 +123,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -153,7 +153,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxgg/CudaRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxgg/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxgg/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxgg/GpuAbstraction.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxgg/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxgg/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxgg/GpuRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxgg/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxgg/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxgg/check_sa.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxgg/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxgg/check_sa.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxgg/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxuux/CPPProcess.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxuux/CPPProcess.cc
index 70fbbee59f..80d2682458 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxuux/CPPProcess.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxuux/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_sm.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -49,7 +48,7 @@
 // Process: d d~ > t t~ d d~ WEIGHTED<=4 @2
 // Process: s s~ > t t~ s s~ WEIGHTED<=4 @2
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +82,7 @@ namespace mg5amcCpu
   __device__ const fptype cIPD[2] = { (fptype)Parameters_sm::mdl_MT, (fptype)Parameters_sm::mdl_WT };
   __device__ const fptype* cIPC = nullptr; // unused as nicoup=0
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ fptype cIPD[2];
   __device__ __constant__ fptype* cIPC = nullptr; // unused as nicoup=0
 #else
@@ -93,7 +92,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -121,13 +120,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -154,7 +153,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -190,7 +189,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -203,8 +202,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -497,7 +498,7 @@ namespace mg5amcCpu
         { 3, 9, 9, 3, 27, 9 },
         { 9, 3, 3, 9, 9, 27 } }; // 2-D array[6][6]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -554,7 +555,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -613,7 +614,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -708,8 +709,8 @@ namespace mg5amcCpu
       { -1, 1, 1, -1, -1, -1 },
       { -1, 1, 1, -1, 1, 1 },
       { -1, 1, 1, -1, 1, -1 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -751,9 +752,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     const fptype tIPD[2] = { (fptype)m_pars->mdl_MT, (fptype)m_pars->mdl_WT };
     //const cxtype tIPC[0] = { ... }; // nicoup=0
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) ) );
-    //checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ) ); // nicoup=0
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) );
+    //gpuMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
 #else
     memcpy( cIPD, tIPD, 2 * sizeof( fptype ) );
     //memcpy( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
@@ -791,7 +792,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -820,6 +821,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -833,6 +838,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -856,12 +863,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -882,7 +889,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -1008,9 +1015,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -1034,7 +1041,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -1054,7 +1061,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 36 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -1068,9 +1075,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -1098,7 +1108,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -1308,7 +1318,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxuux/CPPProcess.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxuux/CPPProcess.h
index 58cece5c62..d8d3d481ea 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxuux/CPPProcess.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxuux/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -110,7 +110,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -123,7 +123,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -153,7 +153,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxuux/CudaRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxuux/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxuux/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxuux/GpuAbstraction.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxuux/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxuux/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxuux/GpuRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxuux/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxuux/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxuux/check_sa.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxuux/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxuux/check_sa.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uux_ttxuux/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxcx_ttxuxcx/CPPProcess.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxcx_ttxuxcx/CPPProcess.cc
index 7df13a2341..8dd445388c 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxcx_ttxuxcx/CPPProcess.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxcx_ttxuxcx/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_sm.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -51,7 +50,7 @@
 // Process: c~ s~ > t t~ c~ s~ WEIGHTED<=4 @2
 // Process: d~ s~ > t t~ d~ s~ WEIGHTED<=4 @2
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -85,7 +84,7 @@ namespace mg5amcCpu
   __device__ const fptype cIPD[2] = { (fptype)Parameters_sm::mdl_MT, (fptype)Parameters_sm::mdl_WT };
   __device__ const fptype* cIPC = nullptr; // unused as nicoup=0
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ fptype cIPD[2];
   __device__ __constant__ fptype* cIPC = nullptr; // unused as nicoup=0
 #else
@@ -95,7 +94,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -123,13 +122,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -156,7 +155,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -192,7 +191,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -205,8 +204,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -387,7 +388,7 @@ namespace mg5amcCpu
         { 3, 9, 9, 3, 27, 9 },
         { 9, 3, 3, 9, 9, 27 } }; // 2-D array[6][6]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -444,7 +445,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -503,7 +504,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -598,8 +599,8 @@ namespace mg5amcCpu
       { 1, 1, 1, -1, 1, -1 },
       { 1, 1, 1, -1, -1, 1 },
       { 1, 1, 1, -1, -1, -1 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -641,9 +642,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     const fptype tIPD[2] = { (fptype)m_pars->mdl_MT, (fptype)m_pars->mdl_WT };
     //const cxtype tIPC[0] = { ... }; // nicoup=0
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) ) );
-    //checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ) ); // nicoup=0
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) );
+    //gpuMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
 #else
     memcpy( cIPD, tIPD, 2 * sizeof( fptype ) );
     //memcpy( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
@@ -681,7 +682,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -710,6 +711,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -723,6 +728,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -746,12 +753,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -772,7 +779,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -898,9 +905,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -924,7 +931,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -944,7 +951,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 36 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -958,9 +965,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -988,7 +998,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -1198,7 +1208,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxcx_ttxuxcx/CPPProcess.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxcx_ttxuxcx/CPPProcess.h
index 6bd3135c3c..901c6dfcc9 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxcx_ttxuxcx/CPPProcess.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxcx_ttxuxcx/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -112,7 +112,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -125,7 +125,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -155,7 +155,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxcx_ttxuxcx/CudaRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxcx_ttxuxcx/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxcx_ttxuxcx/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxcx_ttxuxcx/GpuAbstraction.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxcx_ttxuxcx/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxcx_ttxuxcx/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxcx_ttxuxcx/GpuRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxcx_ttxuxcx/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxcx_ttxuxcx/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxcx_ttxuxcx/check_sa.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxcx_ttxuxcx/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxcx_ttxuxcx/check_sa.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxcx_ttxuxcx/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxux_ttxuxux/CPPProcess.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxux_ttxuxux/CPPProcess.cc
index f464c27160..08e9ed321f 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxux_ttxuxux/CPPProcess.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxux_ttxuxux/CPPProcess.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -16,7 +16,6 @@
 
 #include "mgOnGpuConfig.h"
 
-#include "CudaRuntime.h"
 #include "HelAmps_sm.h"
 #include "MemoryAccessAmplitudes.h"
 #include "MemoryAccessCouplings.h"
@@ -49,7 +48,7 @@
 // Process: d~ d~ > t t~ d~ d~ WEIGHTED<=4 @2
 // Process: s~ s~ > t t~ s~ s~ WEIGHTED<=4 @2
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +82,7 @@ namespace mg5amcCpu
   __device__ const fptype cIPD[2] = { (fptype)Parameters_sm::mdl_MT, (fptype)Parameters_sm::mdl_WT };
   __device__ const fptype* cIPC = nullptr; // unused as nicoup=0
 #else
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ fptype cIPD[2];
   __device__ __constant__ fptype* cIPC = nullptr; // unused as nicoup=0
 #else
@@ -93,7 +92,7 @@ namespace mg5amcCpu
 #endif
 
   // Helicity combinations (and filtering of "good" helicity combinations)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __device__ __constant__ short cHel[ncomb][npar];
   __device__ __constant__ int cNGoodHel;
   __device__ __constant__ int cGoodHel[ncomb];
@@ -121,13 +120,13 @@ namespace mg5amcCpu
                            fptype* allDenominators,       // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
                            fptype_sv* jamp2_sv            // output: jamp2[nParity][ncolor][neppV] for color choice (nullptr if disabled)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                            , const int ievt00             // input: first event number in current C++ event page (for CUDA, ievt depends on threadid)
 #endif
                            )
   //ALWAYS_INLINE // attributes are not permitted in a function definition
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using M_ACCESS = DeviceAccessMomenta;         // non-trivial access: buffer includes all events
     using E_ACCESS = DeviceAccessMatrixElements;  // non-trivial access: buffer includes all events
@@ -154,7 +153,7 @@ namespace mg5amcCpu
 #endif /* clang-format on */
     mgDebug( 0, __FUNCTION__ );
     //printf( "calculate_wavefunctions: ihel=%2d\n", ihel );
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     //printf( "calculate_wavefunctions: ievt00=%d\n", ievt00 );
 #endif
 
@@ -190,7 +189,7 @@ namespace mg5amcCpu
 #endif
     for( int iParity = 0; iParity < nParity; ++iParity )
     { // START LOOP ON IPARITY
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       const int ievt0 = ievt00 + iParity * neppV;
 #endif
       constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -203,8 +202,10 @@ namespace mg5amcCpu
         allCOUPs[idcoup] = CD_ACCESS::idcoupAccessBufferConst( allcouplings, idcoup ); // dependent couplings, vary event-by-event
       for( size_t iicoup = 0; iicoup < nicoup; iicoup++ )
         allCOUPs[ndcoup + iicoup] = CI_ACCESS::iicoupAccessBufferConst( cIPC, iicoup ); // independent couplings, fixed for all events
+#ifdef MGONGPUCPP_GPUIMPL
 #ifdef __CUDACC__
 #pragma nv_diagnostic pop
+#endif
       // CUDA kernels take input/output buffers with momenta/MEs for all events
       const fptype* momenta = allmomenta;
       const fptype* COUPs[nxcoup];
@@ -497,7 +498,7 @@ namespace mg5amcCpu
         { 3, 9, 9, 3, 27, 9 },
         { 9, 3, 3, 9, 9, 27 } }; // 2-D array[6][6]
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
       // Pre-compute a constexpr triangular color matrix properly normalized #475
       struct TriangularNormalizedColorMatrix
       {
@@ -554,7 +555,7 @@ namespace mg5amcCpu
 #endif
       for( int icol = 0; icol < ncolor; icol++ )
       {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
         // === C++ START ===
         // Diagonal terms
 #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
@@ -613,7 +614,7 @@ namespace mg5amcCpu
       MEs_sv_previous += deltaMEs_previous;
 #endif
       /*
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       if ( cNGoodHel > 0 ) printf( "calculate_wavefunctions: ievt=%6d ihel=%2d me_running=%f\n", blockDim.x * blockIdx.x + threadIdx.x, ihel, MEs_sv );
 #else
 #ifdef MGONGPU_CPPSIMD
@@ -708,8 +709,8 @@ namespace mg5amcCpu
       { 1, 1, 1, -1, 1, -1 },
       { 1, 1, 1, -1, -1, 1 },
       { 1, 1, 1, -1, -1, -1 } };
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cHel, tHel, ncomb * npar * sizeof( short ) );
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
@@ -751,9 +752,9 @@ namespace mg5amcCpu
     // Then copy them to CUDA constant memory (issue #39) or its C++ emulation in file-scope static memory
     const fptype tIPD[2] = { (fptype)m_pars->mdl_MT, (fptype)m_pars->mdl_WT };
     //const cxtype tIPC[0] = { ... }; // nicoup=0
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) ) );
-    //checkCuda( cudaMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ) ); // nicoup=0
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cIPD, tIPD, 2 * sizeof( fptype ) );
+    //gpuMemcpyToSymbol( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
 #else
     memcpy( cIPD, tIPD, 2 * sizeof( fptype ) );
     //memcpy( cIPC, tIPC, 0 * sizeof( cxtype ) ); // nicoup=0
@@ -791,7 +792,7 @@ namespace mg5amcCpu
   {
     std::stringstream out;
     // CUDA version (NVCC)
-    // [Use __NVCC__ instead of __CUDACC__ here!]
+    // [Use __NVCC__ instead of MGONGPUCPP_GPUIMPL here!]
     // [This tests if 'nvcc' was used even to build a .cc file, even if not necessarily 'nvcc -x cu' for a .cu file]
     // [Check 'nvcc --compiler-options -dM -E dummy.c | grep CUDA': see https://stackoverflow.com/a/53713712]
 #ifdef __NVCC__
@@ -820,6 +821,10 @@ namespace mg5amcCpu
     out << "Apple clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
 #else
     out << "clang " << __clang_major__ << "." << __clang_minor__ << "." << __clang_patchlevel__;
+    /*
+    // === AV 26-Jan-2024 DISABLE THIS CODE (START)
+    // === AV 26-Jan-2024 First, it is totally wrong to assume that the CXX environment variable is used in the build!
+    // === AV 26-Jan-2024 Second and worse, here we need build time values, while CXX in this code is evaluated at runtime!
     // GCC toolchain version inside CLANG
     std::string tchainout;
     std::string tchaincmd = "readelf -p .comment $(${CXX} -print-libgcc-file-name) |& grep 'GCC: (GNU)' | grep -v Warning | sort -u | awk '{print $5}'";
@@ -833,6 +838,8 @@ namespace mg5amcCpu
 #else
     out << " (gcc " << tchainout << ")";
 #endif
+    // === AV 26-Jan-2024 DISABLE THIS CODE (END)
+    */
 #endif
 #else
     out << "clang UNKNOWKN";
@@ -856,12 +863,12 @@ namespace mg5amcCpu
   __global__ void /* clang-format off */
   computeDependentCouplings( const fptype* allgs, // input: Gs[nevt]
                              fptype* allcouplings // output: couplings[nevt*ndcoup*2]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
                              , const int nevt     // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
   ) /* clang-format on */
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     using namespace mg5amcGpu;
     using G_ACCESS = DeviceAccessGs;
     using C_ACCESS = DeviceAccessCouplings;
@@ -882,7 +889,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -1008,9 +1015,9 @@ namespace mg5amcCpu
         nGoodHel++;
       }
     }
-#ifdef __CUDACC__
-    checkCuda( cudaMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) ) );
-    checkCuda( cudaMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) ) );
+#ifdef MGONGPUCPP_GPUIMPL
+    gpuMemcpyToSymbol( cNGoodHel, &nGoodHel, sizeof( int ) );
+    gpuMemcpyToSymbol( cGoodHel, goodHel, ncomb * sizeof( int ) );
 #else
     cNGoodHel = nGoodHel;
     for( int ihel = 0; ihel < ncomb; ihel++ ) cGoodHel[ihel] = goodHel[ihel];
@@ -1034,7 +1041,7 @@ namespace mg5amcCpu
 #endif
             int* allselhel,                // output: helicity selection[nevt]
             int* allselcol                 // output: helicity selection[nevt]
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
             , const int nevt               // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
 #endif
             ) /* clang-format on */
@@ -1054,7 +1061,7 @@ namespace mg5amcCpu
     // Denominators: spins, colors and identical particles
     constexpr int helcolDenominators[1] = { 72 }; // assume nprocesses == 1 (#272 and #343)
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Remember: in CUDA this is a kernel for one event, in c++ this processes n events
     const int ievt = blockDim.x * blockIdx.x + threadIdx.x; // index of event (thread) in grid
 #else
@@ -1068,9 +1075,12 @@ namespace mg5amcCpu
 #endif
 
     // Start sigmaKin_lines
+
+#include "GpuAbstraction.h"
+
     // === PART 0 - INITIALISATION (before calculate_wavefunctions) ===
     // Reset the "matrix elements" - running sums of |M|^2 over helicities for the given event
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] = 0;
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     allNumerators[ievt] = 0;
@@ -1098,7 +1108,7 @@ namespace mg5amcCpu
     // === PART 1 - HELICITY LOOP: CALCULATE WAVEFUNCTIONS ===
     // (in both CUDA and C++, using precomputed good helicities)
 
-#ifdef __CUDACC__ // CUDA OR C++
+#ifdef MGONGPUCPP_GPUIMPL // CUDA OR C++
 
     // *** START OF PART 1a - CUDA (one event per CPU thread) ***
     // Running sum of partial amplitudes squared for event by event color selection (#402)
@@ -1308,7 +1318,7 @@ namespace mg5amcCpu
     // Get the final |M|^2 as an average over helicities/colors of the running sum of |M|^2 over helicities for the given event
     // [NB 'sum over final spins, average over initial spins', eg see
     // https://www.uzh.ch/cmsssl/physik/dam/jcr:2e24b7b1-f4d7-4160-817e-47b13dbf1d7c/Handout_4_2016-UZH.pdf]
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     allMEs[ievt] /= helcolDenominators[0];
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
     if( channelId > 0 ) allMEs[ievt] *= allNumerators[ievt] / allDenominators[ievt];
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxux_ttxuxux/CPPProcess.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxux_ttxuxux/CPPProcess.h
index 4e53fa1250..c2ca443c0e 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxux_ttxuxux/CPPProcess.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxux_ttxuxux/CPPProcess.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -25,7 +25,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -110,7 +110,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   computeDependentCouplings( const fptype* allgs,    // input: Gs[nevt]
                              fptype* allcouplings ); // output: couplings[nevt*ndcoup*2]
@@ -123,7 +123,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin_getGoodHel( const fptype* allmomenta,   // input: momenta[nevt*npar*4]
                        const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
@@ -153,7 +153,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   __global__ void
   sigmaKin( const fptype* allmomenta,      // input: momenta[nevt*npar*4]
             const fptype* allcouplings,    // input: couplings[nevt*ndcoup*2]
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxux_ttxuxux/CudaRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxux_ttxuxux/CudaRuntime.h
deleted file mode 120000
index ce9e1a487a..0000000000
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxux_ttxuxux/CudaRuntime.h
+++ /dev/null
@@ -1 +0,0 @@
-../CudaRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxux_ttxuxux/GpuAbstraction.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxux_ttxuxux/GpuAbstraction.h
new file mode 120000
index 0000000000..72054e19ba
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxux_ttxuxux/GpuAbstraction.h
@@ -0,0 +1 @@
+../GpuAbstraction.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxux_ttxuxux/GpuRuntime.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxux_ttxuxux/GpuRuntime.h
new file mode 120000
index 0000000000..3920e83be4
--- /dev/null
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxux_ttxuxux/GpuRuntime.h
@@ -0,0 +1 @@
+../GpuRuntime.h
\ No newline at end of file
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxux_ttxuxux/check_sa.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxux_ttxuxux/check_sa.cc
index 3fbf0ffbee..bde384c69e 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxux_ttxuxux/check_sa.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/P2_uxux_ttxuxux/check_sa.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: O. Mattelaer (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -12,6 +12,7 @@
 #include "BridgeKernels.h"
 #include "CPPProcess.h"
 #include "CrossSectionKernels.h"
+#include "GpuRuntime.h"
 #include "MatrixElementKernels.h"
 #include "MemoryAccessMatrixElements.h"
 #include "MemoryAccessMomenta.h"
@@ -65,7 +66,7 @@ usage( char* argv0, int ret = 1 )
   std::cout << std::endl;
   std::cout << "Summary stats are always computed: '-p' and '-j' only control their printout" << std::endl;
   std::cout << "The '-d' flag only enables NaN/abnormal warnings and OMP debugging" << std::endl;
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   std::cout << std::endl;
   std::cout << "Use the OMP_NUM_THREADS environment variable to control OMP multi-threading" << std::endl;
@@ -75,7 +76,7 @@ usage( char* argv0, int ret = 1 )
   return ret;
 }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +84,7 @@ namespace mg5amcCpu
 {
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU)" << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU)" << std::endl;
@@ -96,7 +97,7 @@ int
 main( int argc, char** argv )
 {
   // Namespaces for CUDA and C++ (FIXME - eventually use the same namespace everywhere...)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -134,9 +135,11 @@ main( int argc, char** argv )
     CurandDevice = 2
   };
 #ifdef MGONGPU_HAS_NO_CURAND
-  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784)
+  RandomNumberMode rndgen = RandomNumberMode::CommonRandom; // this is the only supported mode if build has no curand (PR #784 and #785)
+#elif defined __HIPCC__
+#error Internal error: MGONGPU_HAS_NO_CURAND should have been set for __HIPCC__ // default on AMD GPUs should be common random
 #elif defined __CUDACC__
-  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on GPU if build has curand
+  RandomNumberMode rndgen = RandomNumberMode::CurandDevice; // default on NVidia GPU if build has curand
 #else
   RandomNumberMode rndgen = RandomNumberMode::CurandHost; // default on CPU if build has curand
 #endif
@@ -146,10 +149,10 @@ main( int argc, char** argv )
     RamboHost = 1,
     RamboDevice = 2
   };
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
 #else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;  // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
 #endif
   // Bridge emulation mode (NB Bridge implies RamboHost!)
   bool bridge = false;
@@ -177,7 +180,7 @@ main( int argc, char** argv )
     else if( arg == "--curdev" )
     {
 #ifndef __CUDACC__
-      throw std::runtime_error( "CurandDevice is not supported on CPUs" );
+      throw std::runtime_error( "CurandDevice is not supported on CPUs or non-NVidia GPUs" );
 #elif defined MGONGPU_HAS_NO_CURAND
       throw std::runtime_error( "CurandDevice is not supported because this application was built without Curand support" );
 #else
@@ -198,7 +201,7 @@ main( int argc, char** argv )
     }
     else if( arg == "--rmbdev" )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       rmbsmp = RamboSamplingMode::RamboDevice;
 #else
       throw std::runtime_error( "RamboDevice is not supported on CPUs" );
@@ -272,13 +275,13 @@ main( int argc, char** argv )
     return usage( argv[0] );
   }
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
   ompnumthreadsNotSetMeansOneThread( debug ? 1 : 0 ); // quiet(-1), info(0), debug(1)
 #endif
 #endif
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   // Fail gently and avoid "Illegal instruction (core dumped)" if the host does not support the SIMD used in the ME calculation
   // Note: this prevents a crash on pmpe04 but not on some github CI nodes?
   // [NB: SIMD vectorization in mg5amc C++ code is only used in the ME calculation below MatrixElementKernelHost!]
@@ -296,14 +299,14 @@ main( int argc, char** argv )
 
   // === STEP 0 - INITIALISE
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
-  // --- 00. Initialise cuda
-  // Instantiate a CudaRuntime at the beginnining of the application's main to
-  // invoke cudaSetDevice(0) in the constructor and book a cudaDeviceReset() call in the destructor
-  const std::string cdinKey = "00 CudaInit";
+  // --- 00. Initialise GPU
+  // Instantiate a GpuRuntime at the beginnining of the application's main.
+  // For CUDA this invokes cudaSetDevice(0) in the constructor and books a cudaDeviceReset() call in the destructor.
+  const std::string cdinKey = "00 GpuInit";
   timermap.start( cdinKey );
-  CudaRuntime cudaRuntime( debug );
+  GpuRuntime GpuRuntime( debug );
 #endif
 
   // --- 0a. Initialise physics process
@@ -325,7 +328,7 @@ main( int argc, char** argv )
   timermap.start( alloKey );
 
   // Memory buffers for random numbers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumMomenta hstRndmom( nevt );
 #else
   PinnedHostBufferRndNumMomenta hstRndmom( nevt );
@@ -333,7 +336,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for sampling weights
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferWeights hstWeights( nevt );
 #else
   PinnedHostBufferWeights hstWeights( nevt );
@@ -341,7 +344,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for momenta
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMomenta hstMomenta( nevt );
 #else
   PinnedHostBufferMomenta hstMomenta( nevt );
@@ -349,7 +352,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for Gs
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferGs hstGs( nevt );
 #else
   PinnedHostBufferGs hstGs( nevt );
@@ -366,7 +369,7 @@ main( int argc, char** argv )
   }
 
   // Memory buffers for matrix elements
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferMatrixElements hstMatrixElements( nevt );
 #else
   PinnedHostBufferMatrixElements hstMatrixElements( nevt );
@@ -375,7 +378,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for helicity selection
   // *** NB #403 these buffers always remain initialised at 0: no need for helicity choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumHelicity hstRndHel( nevt );
 #else
   PinnedHostBufferRndNumHelicity hstRndHel( nevt );
@@ -384,7 +387,7 @@ main( int argc, char** argv )
 
   // Memory buffers for random numbers for color selection
   // *** NB #402 these buffers always remain initialised at 0: no need for color choice in gcheck/check (no LHE produced) ***
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferRndNumColor hstRndCol( nevt );
 #else
   PinnedHostBufferRndNumColor hstRndCol( nevt );
@@ -392,7 +395,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for helicity selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedHelicity hstSelHel( nevt );
 #else
   PinnedHostBufferSelectedHelicity hstSelHel( nevt );
@@ -400,7 +403,7 @@ main( int argc, char** argv )
 #endif
 
   // Memory buffers for color selection
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   HostBufferSelectedColor hstSelCol( nevt );
 #else
   PinnedHostBufferSelectedColor hstSelCol( nevt );
@@ -438,7 +441,7 @@ main( int argc, char** argv )
     const bool onDevice = true;
     prnk.reset( new CurandRandomNumberKernel( devRndmom, onDevice ) );
 #else
-    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
+    throw std::logic_error( "INTERNAL ERROR! CurandDevice is not supported on CPUs or non-NVidia GPUs" ); // INTERNAL ERROR (no path to this statement)
 #endif
   }
 
@@ -450,7 +453,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     prsk.reset( new RamboSamplingKernelDevice( energy, devRndmom, devMomenta, devWeights, gpublocks, gputhreads ) );
 #else
     throw std::logic_error( "RamboDevice is not supported on CPUs" ); // INTERNAL ERROR (no path to this statement)
@@ -461,7 +464,7 @@ main( int argc, char** argv )
   std::unique_ptr<MatrixElementKernelBase> pmek;
   if( !bridge )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new MatrixElementKernelDevice( devMomenta, devGs, devRndHel, devRndCol, devMatrixElements, devSelHel, devSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new MatrixElementKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -469,7 +472,7 @@ main( int argc, char** argv )
   }
   else
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     pmek.reset( new BridgeKernelDevice( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, gpublocks, gputhreads ) );
 #else
     pmek.reset( new BridgeKernelHost( hstMomenta, hstGs, hstRndHel, hstRndCol, hstMatrixElements, hstSelHel, hstSelCol, nevt ) );
@@ -511,7 +514,7 @@ main( int argc, char** argv )
     prnk->generateRnarray();
     //std::cout << "Got random numbers" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rndgen != RandomNumberMode::CurandDevice && rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 1c. Copy rndmom from host to device
@@ -543,7 +546,7 @@ main( int argc, char** argv )
     prsk->getMomentaFinal();
     //std::cout << "Got final momenta" << std::endl;
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( rmbsmp == RamboSamplingMode::RamboDevice )
     {
       // --- 2c. CopyDToH Weights
@@ -588,7 +591,7 @@ main( int argc, char** argv )
       dynamic_cast<BridgeKernelBase*>( pmek.get() )->transposeInputMomentaC2F();
     }
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // --- 2d. CopyHToD Momenta
     const std::string gKey = "0.. CpHTDg";
     rambtime += timermap.start( gKey ); // FIXME! NOT A RAMBO TIMER!
@@ -617,7 +620,7 @@ main( int argc, char** argv )
     wv3atime += timermap.stop(); // calc only
     wavetime += wv3atime;        // calc plus copy
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     if( !bridge )
     {
       // --- 3b. CopyDToH MEs
@@ -760,18 +763,22 @@ main( int argc, char** argv )
     rndgentxt = "CURAND DEVICE";
 #ifdef __CUDACC__
   rndgentxt += " (CUDA code)";
+#elif defined __HIPCC__
+  rndgentxt += " (HIP code)";
 #else
   rndgentxt += " (C++ code)";
 #endif
 
   // Workflow description summary
   std::string wrkflwtxt;
-  // -- CUDA or C++?
+  // -- CUDA or HIP or C++?
 #ifdef __CUDACC__
   wrkflwtxt += "CUD:";
+#elif defined __HIPCC__
+  wrkflwtxt += "HIP:";
 #else
   wrkflwtxt += "CPP:";
-#endif
+#endif /* clang-format off */
   // -- DOUBLE or FLOAT?
 #if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
   wrkflwtxt += "MIX+"; // mixed fptypes (single precision color algebra #537)
@@ -781,8 +788,8 @@ main( int argc, char** argv )
   wrkflwtxt += "FLT+";
 #else
   wrkflwtxt += "???+"; // no path to this statement
-#endif
-  // -- CUCOMPLEX or THRUST or STD complex numbers?
+#endif /* clang-format on */
+  // -- CUCOMPLEX or THRUST or STD or CXSIMPLE complex numbers?
 #ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
   wrkflwtxt += "CUX:";
@@ -793,6 +800,12 @@ main( int argc, char** argv )
 #else
   wrkflwtxt += "???:"; // no path to this statement
 #endif
+#elif defined __HIPCC__
+#if defined MGONGPU_HIPCXTYPE_CXSMPL
+  wrkflwtxt += "CXS:";
+#else
+  wrkflwtxt += "???:"; // no path to this statement
+#endif
 #else
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
   wrkflwtxt += "STX:";
@@ -818,7 +831,7 @@ main( int argc, char** argv )
     wrkflwtxt += "RMBDEV+";
   else
     wrkflwtxt += "??????+"; // no path to this statement
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // -- HOST or DEVICE matrix elements? Standalone MEs or BRIDGE?
   if( !bridge )
     wrkflwtxt += "MESDEV";
@@ -850,7 +863,7 @@ main( int argc, char** argv )
   wrkflwtxt += "/sse4";
 #endif
 #else
-  wrkflwtxt += "/????"; // no path to this statement
+  wrkflwtxt += "/????";                                           // no path to this statement
 #endif
   // -- Has cxtype_v::operator[] bracket with non-const reference?
 #if defined MGONGPU_CPPSIMD
@@ -874,7 +887,7 @@ main( int argc, char** argv )
 
   if( perf )
   {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
     // Get the output of "nproc --all" (https://stackoverflow.com/a/478960)
     std::string nprocall;
@@ -895,6 +908,8 @@ main( int argc, char** argv )
     std::cout << std::string( SEP79, '*' ) << std::endl
 #ifdef __CUDACC__
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CUDA"
+#elif defined __HIPCC__
+              << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_HIP"
 #else
               << "Process                     = " << XSTRINGIFY( MG_EPOCH_PROCESS_ID ) << "_CPP"
 #endif
@@ -921,21 +936,23 @@ main( int argc, char** argv )
 #elif defined MGONGPU_FPTYPE_FLOAT
               << "FP precision                = FLOAT (NaN/abnormal=" << nabn << ", zero=" << nzero << ")" << std::endl
 #endif
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
               << "Complex type                = CUCOMPLEX" << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
               << "Complex type                = THRUST::COMPLEX" << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+              << "Complex type                = CXSIMPLE" << std::endl
+#elif defined MGONGPU_CPPCXTYPE_STDCOMPLEX
               << "Complex type                = STD::COMPLEX" << std::endl
+#else
+              << "Complex type                = ???" << std::endl // no path to this statement...
 #endif
               << "RanNumb memory layout       = AOSOA[" << neppR << "]"
               << ( neppR == 1 ? " == AOS" : "" )
               << " [HARDCODED FOR REPRODUCIBILITY]" << std::endl
               << "Momenta memory layout       = AOSOA[" << neppM << "]"
               << ( neppM == 1 ? " == AOS" : "" ) << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "Wavefunction GPU memory     = LOCAL" << std::endl
 #else
 #if !defined MGONGPU_CPPSIMD
@@ -966,7 +983,7 @@ main( int argc, char** argv )
 #endif
 #endif
               << "Random number generation    = " << rndgentxt << std::endl
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 #ifdef _OPENMP
               << "OMP threads / `nproc --all` = " << omp_get_max_threads() << " / " << nprocall // includes a newline
 #endif
@@ -1062,14 +1079,16 @@ main( int argc, char** argv )
              << "\"FLOAT (NaN/abnormal=" << nabn << ")\"," << std::endl
 #endif
              << "\"Complex type\": "
-#ifdef __CUDACC__
 #if defined MGONGPU_CUCXTYPE_CUCOMPLEX
              << "\"CUCOMPLEX\"," << std::endl
 #elif defined MGONGPU_CUCXTYPE_THRUST
              << "\"THRUST::COMPLEX\"," << std::endl
-#endif
-#else
+#elif defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+             << "\"CXSIMPLE\"," << std::endl
+#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
              << "\"STD::COMPLEX\"," << std::endl
+#else
+             << "\"???\"," << std::endl                           // no path to this statement...
 #endif
              << "\"RanNumb memory layout\": "
              << "\"AOSOA[" << neppR << "]\""
@@ -1077,7 +1096,7 @@ main( int argc, char** argv )
              << "\"Momenta memory layout\": "
              << "\"AOSOA[" << neppM << "]\""
              << ( neppM == 1 ? " == AOS" : "" ) << ", " << std::endl
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     //<< "\"Wavefunction GPU memory\": " << "\"LOCAL\"," << std::endl
 #endif
              << "\"Curand generation\": "
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/RamboSamplingKernels.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/RamboSamplingKernels.cc
index da68aa9255..79abbcc4f8 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/RamboSamplingKernels.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/RamboSamplingKernels.cc
@@ -1,11 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "RamboSamplingKernels.h"
 
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 #include "MemoryAccessMomenta.h"
 #include "MemoryAccessRandomNumbers.h"
 #include "MemoryAccessWeights.h"
@@ -14,7 +14,7 @@
 
 #include <sstream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -92,7 +92,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   RamboSamplingKernelDevice::RamboSamplingKernelDevice( const fptype energy,               // input: energy
                                                         const BufferRndNumMomenta& rndmom, // input: random numbers in [0,1]
                                                         BufferMomenta& momenta,            // output: momenta
@@ -135,7 +135,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   getMomentaInitialDevice( const fptype energy,
                            fptype* momenta )
@@ -147,17 +147,17 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   void
   RamboSamplingKernelDevice::getMomentaInitial()
   {
-    getMomentaInitialDevice<<<m_gpublocks, m_gputhreads>>>( m_energy, m_momenta.data() );
+    gpuLaunchKernel( getMomentaInitialDevice, m_gpublocks, m_gputhreads, m_energy, m_momenta.data() );
   }
 #endif
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   __global__ void
   getMomentaFinalDevice( const fptype energy,
                          const fptype* rndmom,
@@ -171,11 +171,11 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   void
   RamboSamplingKernelDevice::getMomentaFinal()
   {
-    getMomentaFinalDevice<<<m_gpublocks, m_gputhreads>>>( m_energy, m_rndmom.data(), m_momenta.data(), m_weights.data() );
+    gpuLaunchKernel( getMomentaFinalDevice, m_gpublocks, m_gputhreads, m_energy, m_rndmom.data(), m_momenta.data(), m_weights.data() );
   }
 #endif
 
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/RamboSamplingKernels.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/RamboSamplingKernels.h
index 184089efd7..7c214cd74b 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/RamboSamplingKernels.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/RamboSamplingKernels.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef RAMBOSAMPLINGKERNELS_H
 #define RAMBOSAMPLINGKERNELS_H 1
@@ -10,7 +10,7 @@
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -93,7 +93,7 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   // A class encapsulating RAMBO phase space sampling on a GPU device
   class RamboSamplingKernelDevice final : public SamplingKernelBase, public NumberOfEvents
   {
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/RandomNumberKernels.h b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/RandomNumberKernels.h
index 188a72c2c9..21d63beeac 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/RandomNumberKernels.h
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/RandomNumberKernels.h
@@ -1,14 +1,14 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Dec 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef RANDOMNUMBERKERNELS_H
 #define RANDOMNUMBERKERNELS_H 1
 
 #include "mgOnGpuConfig.h"
 
-// NB This must come AFTER mgOnGpuConfig.h which contains our definition of __global__ when __CUDACC__ is not defined
+// NB This must come AFTER mgOnGpuConfig.h which contains our definition of __global__ when MGONGPUCPP_GPUIMPL is not defined
 #ifndef MGONGPU_HAS_NO_CURAND
 //#include "curand.h"
 struct curandGenerator_st; // forward definition from curand.h
@@ -16,7 +16,7 @@ struct curandGenerator_st; // forward definition from curand.h
 
 #include "MemoryBuffers.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/cudacpp.mk b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/cudacpp.mk
index 509307506b..df74dfc284 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/cudacpp.mk
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/cudacpp.mk
@@ -1,7 +1,7 @@
 # Copyright (C) 2020-2023 CERN and UCLouvain.
 # Licensed under the GNU Lesser General Public License (version 3 or later).
 # Created by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-# Further modified by: O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+# Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #=== Determine the name of this makefile (https://ftp.gnu.org/old-gnu/Manuals/make-3.80/html_node/make_17.html)
 #=== NB: use ':=' to ensure that the value of CUDACPP_MAKEFILE is not modified further down after including make_opts
@@ -42,10 +42,10 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure common compiler flags for C++ and CUDA
+#=== Configure common compiler flags for C++ and CUDA/HIP
 
 INCFLAGS = -I.
-OPTFLAGS = -O3 # this ends up in CUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
+OPTFLAGS = -O3 # this ends up in GPUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
 
 # Dependency on src directory
 MG5AMC_COMMONLIB = mg5amc_common
@@ -121,24 +121,45 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure the CUDA compiler
+#=== Configure the GPU compiler (CUDA or HIP)
+
+# FIXME! (AV 24.01.2024)
+# In the current implementation (without separate builds for C++ and CUDA/HIP), we first check for cudacc and hipcc in CUDA_HOME and HIP_HOME.
+# If CUDA_HOME or HIP_HOME are not set, try to determine them from the path to cudacc and hipcc.
+# While convoluted, this is currently necessary to allow disabling CUDA/HIP builds by setting CUDA_HOME or HIP_HOME to invalid paths.
+# This will (probably?) be fixed when separate C++ and CUDA/HIP builds are implemented (PR #775).
 
 # If CXX is not a single word (example "clang++ --gcc-toolchain...") then disable CUDA builds (issue #505)
-# This is because it is impossible to pass this to "CUFLAGS += -ccbin <host-compiler>" below
+# This is because it is impossible to pass this to "GPUFLAGS += -ccbin <host-compiler>" below
 ifneq ($(words $(subst ccache ,,$(CXX))),1) # allow at most "CXX=ccache <host-compiler>" from outside
   $(warning CUDA builds are not supported for multi-word CXX "$(CXX)")
   override CUDA_HOME=disabled
 endif
 
-# If CUDA_HOME is not set, try to set it from the location of nvcc
+# If CUDA_HOME is not set, try to set it from the path to nvcc
 ifndef CUDA_HOME
-  CUDA_HOME = $(patsubst %bin/nvcc,%,$(shell which nvcc 2>/dev/null))
+  CUDA_HOME = $(patsubst %/bin/nvcc,%,$(shell which nvcc 2>/dev/null))
   $(warning CUDA_HOME was not set: using "$(CUDA_HOME)")
 endif
 
-# Set NVCC as $(CUDA_HOME)/bin/nvcc if it exists
+# If HIP_HOME is not set, try to set it from the path to hipcc
+ifndef HIP_HOME
+  HIP_HOME = $(patsubst %/bin/hipcc,%,$(shell which hipcc 2>/dev/null))
+  $(warning HIP_HOME was not set: using "$(HIP_HOME)")
+endif
+
+# FIXME! (AV 24.01.2024)
+# In the current implementation (without separate builds for C++ and CUDA/HIP),
+# builds are performed for HIP only if CUDA is not found in the path.
+# If both CUDA and HIP are installed, HIP builds can be triggered by unsetting CUDA_HOME.
+# This will be fixed when separate C++ and CUDA/HIP builds are implemented (PR #775).
+
+#--- Option 1: CUDA exists -> use CUDA
+
+# Set GPUCC as $(CUDA_HOME)/bin/nvcc if it exists
 ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
-  NVCC = $(CUDA_HOME)/bin/nvcc
+
+  GPUCC = $(CUDA_HOME)/bin/nvcc
   USE_NVTX ?=-DUSE_NVTX
   # See https://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/index.html
   # See https://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/
@@ -148,7 +169,7 @@ ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
   # Examples: use 60 for P100 (Piz Daint), 80 for A100 (Juwels Booster, NVidia raplab/Curiosity).
   MADGRAPH_CUDA_ARCHITECTURE ?= 70
   ###CUARCHFLAGS = -gencode arch=compute_$(MADGRAPH_CUDA_ARCHITECTURE),code=compute_$(MADGRAPH_CUDA_ARCHITECTURE) -gencode arch=compute_$(MADGRAPH_CUDA_ARCHITECTURE),code=sm_$(MADGRAPH_CUDA_ARCHITECTURE) # Older implementation (AV): go back to this one for multi-GPU support #533
-  ###CUARCHFLAGS = --gpu-architecture=compute_$(MADGRAPH_CUDA_ARCHITECTURE) --gpu-code=sm_$(MADGRAPH_CUDA_ARCHITECTURE),compute_$(MADGRAPH_CUDA_ARCHITECTURE) # Newer implementation (SH): cannot use this as-is for multi-GPU support #533
+  ###CUARCHFLAGS = --gpu-architecture=compute_$(MADGRAPH_CUDA_ARCHITECTURE) --gpu-code=sm_$(MADGRAPH_CUDA_ARCHITECTURE),compute_$(MADGRAPH_CUDA_ARCHITECTURE)  # Newer implementation (SH): cannot use this as-is for multi-GPU support #533
   comma:=,
   CUARCHFLAGS = $(foreach arch,$(subst $(comma), ,$(MADGRAPH_CUDA_ARCHITECTURE)),-gencode arch=compute_$(arch),code=compute_$(arch) -gencode arch=compute_$(arch),code=sm_$(arch))
   CUINC = -I$(CUDA_HOME)/include/
@@ -158,41 +179,79 @@ ifneq ($(wildcard $(CUDA_HOME)/bin/nvcc),)
     CURANDLIBFLAGS = -L$(CUDA_HOME)/lib64/ -lcurand # NB: -lcuda is not needed here!
   endif
   CUOPTFLAGS = -lineinfo
-  CUFLAGS = $(foreach opt, $(OPTFLAGS), -Xcompiler $(opt)) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
-  ###CUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
-  ###NVCC_VERSION = $(shell $(NVCC) --version | grep 'Cuda compilation tools' | cut -d' ' -f5 | cut -d, -f1)
-  CUFLAGS += -std=c++17 # need CUDA >= 11.2 (see #333): this is enforced in mgOnGpuConfig.h
+  ###GPUFLAGS = $(OPTFLAGS) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
+  GPUFLAGS = $(foreach opt, $(OPTFLAGS), -Xcompiler $(opt)) $(CUOPTFLAGS) $(INCFLAGS) $(CUINC) $(USE_NVTX) $(CUARCHFLAGS) -use_fast_math
+  ###GPUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
+  ###GPUCC_VERSION = $(shell $(GPUCC) --version | grep 'Cuda compilation tools' | cut -d' ' -f5 | cut -d, -f1)
+  GPUFLAGS += -std=c++17 # need CUDA >= 11.2 (see #333): this is enforced in mgOnGpuConfig.h
   # Without -maxrregcount: baseline throughput: 6.5E8 (16384 32 12) up to 7.3E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 160 # improves throughput: 6.9E8 (16384 32 12) up to 7.7E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 128 # improves throughput: 7.3E8 (16384 32 12) up to 7.6E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 96 # degrades throughput: 4.1E8 (16384 32 12) up to 4.5E8 (65536 128 12)
-  ###CUFLAGS+= --maxrregcount 64 # degrades throughput: 1.7E8 (16384 32 12) flat at 1.7E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 160 # improves throughput: 6.9E8 (16384 32 12) up to 7.7E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 128 # improves throughput: 7.3E8 (16384 32 12) up to 7.6E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 96 # degrades throughput: 4.1E8 (16384 32 12) up to 4.5E8 (65536 128 12)
+  ###GPUFLAGS+= --maxrregcount 64 # degrades throughput: 1.7E8 (16384 32 12) flat at 1.7E8 (65536 128 12)
+  CUBUILDRULEFLAGS = -Xcompiler -fPIC -c
+  CCBUILDRULEFLAGS = -Xcompiler -fPIC -c -x cu
+  CUDATESTFLAGS = -lcuda
+
+  # Set the host C++ compiler for GPUCC via "-ccbin <host-compiler>"
+  # (NB issue #505: this must be a single word, "clang++ --gcc-toolchain..." is not supported)
+  GPUFLAGS += -ccbin $(shell which $(subst ccache ,,$(CXX)))
+
+  # Allow newer (unsupported) C++ compilers with older versions of CUDA if ALLOW_UNSUPPORTED_COMPILER_IN_CUDA is set (#504)
+  ifneq ($(origin ALLOW_UNSUPPORTED_COMPILER_IN_CUDA),undefined)
+  GPUFLAGS += -allow-unsupported-compiler
+  endif
+
 else ifneq ($(origin REQUIRE_CUDA),undefined)
+
   # If REQUIRE_CUDA is set but no cuda is found, stop here (e.g. for CI tests on GPU #443)
-  $(error No cuda installation found (set CUDA_HOME or make nvcc visible in PATH))
+  $(error No cuda installation found (set CUDA_HOME or make GPUCC visible in PATH))
+
+#--- Option 2: CUDA does not exist, HIP exists -> use HIP
+
+# Set GPUCC as $(HIP_HOME)/bin/hipcc if it exists
+else ifneq ($(wildcard $(HIP_HOME)/bin/hipcc),)
+
+  GPUCC = $(HIP_HOME)/bin/hipcc
+  #USE_NVTX ?=-DUSE_NVTX # should maybe find something equivalent to this in HIP?
+  HIPARCHFLAGS = -target x86_64-linux-gnu --offload-arch=gfx90a
+  HIPINC = -I$(HIP_HOME)/include/
+  # Note: -DHIP_FAST_MATH is equivalent to -use_fast_math in HIP 
+  # (but only for single precision line 208: https://rocm-developer-tools.github.io/HIP/hcc__detail_2math__functions_8h_source.html)
+  # Note: CUOPTFLAGS should not be used for HIP, it had been added here but was then removed (#808)
+  GPUFLAGS = $(OPTFLAGS) $(INCFLAGS) $(HIPINC) $(HIPARCHFLAGS) -DHIP_FAST_MATH -DHIP_PLATFORM=amd -fPIC
+  ###GPUFLAGS += -Xcompiler -Wall -Xcompiler -Wextra -Xcompiler -Wshadow
+  GPUFLAGS += -std=c++17
+  ###GPUFLAGS+= --maxrregcount 255 # (AV: is this option valid on HIP and meaningful on AMD GPUs?)
+  CUBUILDRULEFLAGS = -fPIC -c
+  CCBUILDRULEFLAGS = -fPIC -c -x hip
+
+else ifneq ($(origin REQUIRE_HIP),undefined)
+
+  # If REQUIRE_HIP is set but no HIP is found, stop here (e.g. for CI tests on GPU #443)
+  $(error No hip installation found (set HIP_HOME or make GPUCC visible in PATH))
+
+#--- Option 3: CUDA does not exist, HIP does not exist -> switch off both CUDA and HIP
+
 else
-  # No cuda. Switch cuda compilation off and go to common random numbers in C++
+
+  # No cudacc and no hipcc: switch CUDA and HIP compilation off and go to common random numbers in C++
   $(warning CUDA_HOME is not set or is invalid: export CUDA_HOME to compile with cuda)
-  override NVCC=
+  $(warning HIP_HOME is not set or is invalid: export HIP_HOME to compile with hip)
+  override GPUCC=
   override USE_NVTX=
   override CUINC=
   override CURANDLIBFLAGS=
-endif
-export NVCC
-export CUFLAGS
 
-# Set the host C++ compiler for nvcc via "-ccbin <host-compiler>"
-# (NB issue #505: this must be a single word, "clang++ --gcc-toolchain..." is not supported)
-CUFLAGS += -ccbin $(shell which $(subst ccache ,,$(CXX)))
-
-# Allow newer (unsupported) C++ compilers with older versions of CUDA if ALLOW_UNSUPPORTED_COMPILER_IN_CUDA is set (#504)
-ifneq ($(origin ALLOW_UNSUPPORTED_COMPILER_IN_CUDA),undefined)
-CUFLAGS += -allow-unsupported-compiler
 endif
 
+# Export GPUCC (so that it can also be used in cudacpp_src.mk?)
+export GPUCC
+export GPUFLAGS
+
 #-------------------------------------------------------------------------------
 
-#=== Configure ccache for C++ and CUDA builds
+#=== Configure ccache for C++ and CUDA/HIP builds
 
 # Enable ccache if USECCACHE=1
 ifeq ($(USECCACHE)$(shell echo $(CXX) | grep ccache),1)
@@ -201,15 +260,15 @@ endif
 #ifeq ($(USECCACHE)$(shell echo $(AR) | grep ccache),1)
 #  override AR:=ccache $(AR)
 #endif
-ifneq ($(NVCC),)
-  ifeq ($(USECCACHE)$(shell echo $(NVCC) | grep ccache),1)
-    override NVCC:=ccache $(NVCC)
+ifneq ($(GPUCC),)
+  ifeq ($(USECCACHE)$(shell echo $(GPUCC) | grep ccache),1)
+    override GPUCC:=ccache $(GPUCC)
   endif
 endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure PowerPC-specific compiler flags for C++ and CUDA
+#=== Configure PowerPC-specific compiler flags for C++ and CUDA/HIP
 
 # PowerPC-specific CXX compiler flags (being reviewed)
 ifeq ($(UNAME_P),ppc64le)
@@ -225,9 +284,9 @@ else
   ######CXXFLAGS+= -fno-semantic-interposition # no benefit (neither alone, nor combined with -flto)
 endif
 
-# PowerPC-specific CUDA compiler flags (to be reviewed!)
+# PowerPC-specific CUDA/HIP compiler flags (to be reviewed!)
 ifeq ($(UNAME_P),ppc64le)
-  CUFLAGS+= -Xcompiler -mno-float128
+  GPUFLAGS+= -Xcompiler -mno-float128
 endif
 
 #-------------------------------------------------------------------------------
@@ -235,9 +294,11 @@ endif
 #=== Configure defaults and check if user-defined choices exist for OMPFLAGS, AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
 
 # Set the default OMPFLAGS choice
-ifneq ($(shell $(CXX) --version | egrep '^Intel'),)
+ifneq ($(findstring hipcc,$(GPUCC)),)
+override OMPFLAGS = # disable OpenMP MT when using hipcc #802
+else ifneq ($(shell $(CXX) --version | egrep '^Intel'),)
 override OMPFLAGS = -fopenmp
-###override OMPFLAGS = # disable OpenMP MT on Intel (was ok without nvcc but not ok with nvcc before #578)
+###override OMPFLAGS = # disable OpenMP MT on Intel (was ok without GPUCC but not ok with GPUCC before #578)
 else ifneq ($(shell $(CXX) --version | egrep '^(clang)'),)
 override OMPFLAGS = -fopenmp
 ###override OMPFLAGS = # disable OpenMP MT on clang (was not ok without or with nvcc before #578)
@@ -293,7 +354,10 @@ endif
 
 # Set the default RNDGEN (random number generator) choice
 ifeq ($(RNDGEN),)
-  ifeq ($(NVCC),)
+  ifeq ($(GPUCC),)
+    override RNDGEN = hasNoCurand
+  # Edgecase for HIP compilation
+  else ifeq ($(findstring hipcc,$(GPUCC)),hipcc)
     override RNDGEN = hasNoCurand
   else ifeq ($(RNDGEN),)
     override RNDGEN = hasCurand
@@ -310,7 +374,7 @@ export OMPFLAGS
 
 #-------------------------------------------------------------------------------
 
-#=== Set the CUDA/C++ compiler flags appropriate to user-defined choices of AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
+#=== Set the CUDA/HIP/C++ compiler flags appropriate to user-defined choices of AVX, FPTYPE, HELINL, HRDCOD, RNDGEN
 
 # Set the build flags appropriate to OMPFLAGS
 $(info OMPFLAGS=$(OMPFLAGS))
@@ -368,13 +432,13 @@ CXXFLAGS+= $(AVXFLAGS)
 $(info FPTYPE=$(FPTYPE))
 ifeq ($(FPTYPE),d)
   CXXFLAGS += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
-  CUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
+  GPUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE
 else ifeq ($(FPTYPE),f)
   CXXFLAGS += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
-  CUFLAGS  += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
+  GPUFLAGS  += -DMGONGPU_FPTYPE_FLOAT -DMGONGPU_FPTYPE2_FLOAT
 else ifeq ($(FPTYPE),m)
   CXXFLAGS += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
-  CUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
+  GPUFLAGS  += -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_FLOAT
 else
   $(error Unknown FPTYPE='$(FPTYPE)': only 'd', 'f' and 'm' are supported)
 endif
@@ -383,7 +447,7 @@ endif
 $(info HELINL=$(HELINL))
 ifeq ($(HELINL),1)
   CXXFLAGS += -DMGONGPU_INLINE_HELAMPS
-  CUFLAGS  += -DMGONGPU_INLINE_HELAMPS
+  GPUFLAGS  += -DMGONGPU_INLINE_HELAMPS
 else ifneq ($(HELINL),0)
   $(error Unknown HELINL='$(HELINL)': only '0' and '1' are supported)
 endif
@@ -392,7 +456,7 @@ endif
 $(info HRDCOD=$(HRDCOD))
 ifeq ($(HRDCOD),1)
   CXXFLAGS += -DMGONGPU_HARDCODE_PARAM
-  CUFLAGS  += -DMGONGPU_HARDCODE_PARAM
+  GPUFLAGS  += -DMGONGPU_HARDCODE_PARAM
 else ifneq ($(HRDCOD),0)
   $(error Unknown HRDCOD='$(HRDCOD)': only '0' and '1' are supported)
 endif
@@ -444,11 +508,11 @@ ifeq ($(UNAME_S),Darwin)
   override CULIBFLAGSRPATH2 =
 else
   # RPATH to cuda/cpp libs when linking executables
-  override CXXLIBFLAGSRPATH = -Wl,-rpath,$(LIBDIRRPATH)
-  override CULIBFLAGSRPATH = -Xlinker -rpath,$(LIBDIRRPATH)
+  override CXXLIBFLAGSRPATH = -Wl,-rpath=$(LIBDIRRPATH)
+  override CULIBFLAGSRPATH = -Xlinker -rpath=$(LIBDIRRPATH)
   # RPATH to common lib when linking cuda/cpp libs
-  override CXXLIBFLAGSRPATH2 = -Wl,-rpath,'$$ORIGIN'
-  override CULIBFLAGSRPATH2 = -Xlinker -rpath,'$$ORIGIN'
+  override CXXLIBFLAGSRPATH2 = -Wl,-rpath='$$ORIGIN'
+  override CULIBFLAGSRPATH2 = -Xlinker -rpath='$$ORIGIN'
 endif
 
 # Setting LD_LIBRARY_PATH or DYLD_LIBRARY_PATH in the RUNTIME is no longer necessary (neither on Linux nor on Mac)
@@ -461,7 +525,7 @@ override RUNTIME =
 cxx_main=$(BUILDDIR)/check.exe
 fcxx_main=$(BUILDDIR)/fcheck.exe
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 cu_main=$(BUILDDIR)/gcheck.exe
 fcu_main=$(BUILDDIR)/fgcheck.exe
 else
@@ -492,14 +556,15 @@ $(BUILDDIR)/.build.$(TAG):
 	@touch $(BUILDDIR)/.build.$(TAG)
 
 # Generic target and build rules: objects from CUDA compilation
-ifneq ($(NVCC),)
+# NB: CCBUILDRULEFLAGS includes "-x cu" for nvcc and "-x hip" for hipcc (#810)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/%.o : %.cu *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $(CUBUILDRULEFLAGS) $< -o $@
 
 $(BUILDDIR)/%_cu.o : %.cc *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c -x cu $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $(CCBUILDRULEFLAGS) $< -o $@
 endif
 
 # Generic target and build rules: objects from C++ compilation
@@ -509,11 +574,14 @@ $(BUILDDIR)/%.o : %.cc *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	$(CXX) $(CPPFLAGS) $(CXXFLAGS) -fPIC -c $< -o $@
 
 # Apply special build flags only to CrossSectionKernel.cc and gCrossSectionKernel.cu (no fast math, see #117 and #516)
+# Added edgecase for HIP compilation
 ifeq ($(shell $(CXX) --version | grep ^nvc++),)
 $(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS := $(filter-out -ffast-math,$(CXXFLAGS))
 $(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS += -fno-fast-math
-ifneq ($(NVCC),)
-$(BUILDDIR)/gCrossSectionKernels.o: CUFLAGS += -Xcompiler -fno-fast-math
+ifeq ($(findstring nvcc,$(GPUCC)),nvcc)
+  $(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -Xcompiler -fno-fast-math
+else
+  $(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -fno-fast-math
 endif
 endif
 
@@ -530,10 +598,10 @@ ifeq ($(RNDGEN),hasCurand)
 $(BUILDDIR)/CurandRandomNumberKernel.o: CXXFLAGS += $(CUINC)
 endif
 
-# Avoid "warning: builtin __has_trivial_... is deprecated; use __is_trivially_... instead" in nvcc with icx2023 (#592)
+# Avoid "warning: builtin __has_trivial_... is deprecated; use __is_trivially_... instead" in GPUCC with icx2023 (#592)
 ifneq ($(shell $(CXX) --version | egrep '^(Intel)'),)
-ifneq ($(NVCC),)
-CUFLAGS += -Xcompiler -Wno-deprecated-builtins
+ifneq ($(GPUCC),)
+GPUFLAGS += -Wno-deprecated-builtins
 endif
 endif
 
@@ -541,8 +609,8 @@ endif
 # This patch does remove the warning, but I prefer to keep it disabled for the moment...
 ###ifneq ($(shell $(CXX) --version | egrep '^(clang|Apple clang|Intel)'),)
 ###$(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS += -Wno-overriding-t-option
-###ifneq ($(NVCC),)
-###$(BUILDDIR)/gCrossSectionKernels.o: CUFLAGS += -Xcompiler -Wno-overriding-t-option
+###ifneq ($(GPUCC),)
+###$(BUILDDIR)/gCrossSectionKernels.o: GPUFLAGS += -Xcompiler -Wno-overriding-t-option
 ###endif
 ###endif
 
@@ -569,7 +637,7 @@ MG5AMC_CXXLIB = mg5amc_$(processid_short)_cpp
 cxx_objects_lib=$(BUILDDIR)/CPPProcess.o $(BUILDDIR)/MatrixElementKernels.o $(BUILDDIR)/BridgeKernels.o $(BUILDDIR)/CrossSectionKernels.o
 cxx_objects_exe=$(BUILDDIR)/CommonRandomNumberKernel.o $(BUILDDIR)/RamboSamplingKernels.o
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 MG5AMC_CULIB = mg5amc_$(processid_short)_cuda
 cu_objects_lib=$(BUILDDIR)/gCPPProcess.o $(BUILDDIR)/gMatrixElementKernels.o $(BUILDDIR)/gBridgeKernels.o $(BUILDDIR)/gCrossSectionKernels.o
 cu_objects_exe=$(BUILDDIR)/gCommonRandomNumberKernel.o $(BUILDDIR)/gRamboSamplingKernels.o
@@ -581,11 +649,17 @@ $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: cxx_objects_lib += $(BUILDDIR)/fbridge.o
 $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib)
 	$(CXX) -shared -o $@ $(cxx_objects_lib) $(CXXLIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: $(BUILDDIR)/fbridge_cu.o
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: cu_objects_lib += $(BUILDDIR)/fbridge_cu.o
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cu_objects_lib)
-	$(NVCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+# Bypass std::filesystem completely to ease portability on LUMI #803
+#ifneq ($(findstring hipcc,$(GPUCC)),)
+#	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB) -lstdc++fs
+#else
+#	$(GPUCC) --shared -o $@ $(cu_objects_lib) $(CULIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+#endif
 endif
 
 #-------------------------------------------------------------------------------
@@ -602,16 +676,16 @@ $(cxx_main): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PAT
 $(cxx_main): $(BUILDDIR)/check_sa.o $(LIBDIR)/lib$(MG5AMC_CXXLIB).so $(cxx_objects_exe) $(BUILDDIR)/CurandRandomNumberKernel.o
 	$(CXX) -o $@ $(BUILDDIR)/check_sa.o $(OMPFLAGS) -ldl -pthread $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe) $(BUILDDIR)/CurandRandomNumberKernel.o $(CURANDLIBFLAGS)
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(cu_main): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(cu_main): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(cu_main): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(cu_main): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 else ifneq ($(shell $(CXX) --version | grep ^nvc++),) # support nvc++ #531
 $(cu_main): LIBFLAGS += -L$(patsubst %bin/nvc++,%lib,$(subst ccache ,,$(CXX))) -lnvhpcatm -lnvcpumath -lnvc
 endif
 $(cu_main): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(cu_main): $(BUILDDIR)/gcheck_sa.o $(LIBDIR)/lib$(MG5AMC_CULIB).so $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o
-	$(NVCC) -o $@ $(BUILDDIR)/gcheck_sa.o $(CUARCHFLAGS) $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o $(CURANDLIBFLAGS)
+	$(GPUCC) -o $@ $(BUILDDIR)/gcheck_sa.o $(CUARCHFLAGS) $(LIBFLAGS) -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) $(BUILDDIR)/gCurandRandomNumberKernel.o $(CURANDLIBFLAGS)
 endif
 
 #-------------------------------------------------------------------------------
@@ -635,19 +709,27 @@ $(fcxx_main): LIBFLAGS += -L$(shell dirname $(shell $(FC) --print-file-name libg
 endif
 $(fcxx_main): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(fcxx_main): $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler.o $(LIBDIR)/lib$(MG5AMC_CXXLIB).so $(cxx_objects_exe)
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(BUILDDIR)/fcheck_sa.o $(OMPFLAGS) $(BUILDDIR)/fsampler.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe) -lstdc++
+else
 	$(CXX) -o $@ $(BUILDDIR)/fcheck_sa.o $(OMPFLAGS) $(BUILDDIR)/fsampler.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CXXLIB) $(cxx_objects_exe)
+endif
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(fcu_main): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(fcu_main): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(fcu_main): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(fcu_main): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 endif
 ifeq ($(UNAME_S),Darwin)
 $(fcu_main): LIBFLAGS += -L$(shell dirname $(shell $(FC) --print-file-name libgfortran.dylib)) # add path to libgfortran on Mac #375
 endif
 $(fcu_main): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(fcu_main): $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBDIR)/lib$(MG5AMC_CULIB).so $(cu_objects_exe)
-	$(NVCC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe)
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe) -lstdc++ -L$(shell dirname $(shell $(GPUCC) -print-prog-name=clang))/../../lib -lamdhip64
+else
+	$(GPUCC) -o $@ $(BUILDDIR)/fcheck_sa.o $(BUILDDIR)/fsampler_cu.o $(LIBFLAGS) -lgfortran -L$(LIBDIR) -l$(MG5AMC_CULIB) $(cu_objects_exe)
+endif
 endif
 
 #-------------------------------------------------------------------------------
@@ -659,7 +741,7 @@ $(BUILDDIR)/testxxx.o: testxxx_cc_ref.txt
 $(testmain): $(BUILDDIR)/testxxx.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/testxxx.o # Comment out this line to skip the C++ test of xxx functions
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/testxxx_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/testxxx_cu.o: INCFLAGS += $(GTESTINC)
 $(BUILDDIR)/testxxx_cu.o: testxxx_cc_ref.txt
@@ -672,7 +754,7 @@ $(BUILDDIR)/testmisc.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/testmisc.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/testmisc.o # Comment out this line to skip the C++ miscellaneous tests
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/testmisc_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/testmisc_cu.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/testmisc_cu.o
@@ -684,12 +766,12 @@ $(BUILDDIR)/runTest.o: INCFLAGS += $(GTESTINC)
 $(testmain): $(BUILDDIR)/runTest.o
 $(testmain): cxx_objects_exe += $(BUILDDIR)/runTest.o
 
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(BUILDDIR)/runTest_cu.o: $(GTESTLIBS)
 $(BUILDDIR)/runTest_cu.o: INCFLAGS += $(GTESTINC)
 ifneq ($(shell $(CXX) --version | grep ^Intel),)
-$(testmain): LIBFLAGS += -lintlc # compile with icpx and link with nvcc (undefined reference to `_intel_fast_memcpy')
-$(testmain): LIBFLAGS += -lsvml # compile with icpx and link with nvcc (undefined reference to `__svml_cos4_l9')
+$(testmain): LIBFLAGS += -lintlc # compile with icpx and link with GPUCC (undefined reference to `_intel_fast_memcpy')
+$(testmain): LIBFLAGS += -lsvml # compile with icpx and link with GPUCC (undefined reference to `__svml_cos4_l9')
 else ifneq ($(shell $(CXX) --version | grep ^nvc++),) # support nvc++ #531
 $(testmain): LIBFLAGS += -L$(patsubst %bin/nvc++,%lib,$(subst ccache ,,$(CXX))) -lnvhpcatm -lnvcpumath -lnvc
 endif
@@ -713,14 +795,23 @@ $(testmain): LIBFLAGS += -lgomp
 endif
 endif
 
-ifeq ($(NVCC),) # link only runTest.o
+# Bypass std::filesystem completely to ease portability on LUMI #803
+#ifneq ($(findstring hipcc,$(GPUCC)),)
+#$(testmain): LIBFLAGS += -lstdc++fs
+#endif
+
+ifeq ($(GPUCC),) # link only runTest.o
 $(testmain): LIBFLAGS += $(CXXLIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(testmain): $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib) $(cxx_objects_exe) $(GTESTLIBS)
 	$(CXX) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) -ldl -pthread $(LIBFLAGS)
 else # link both runTest.o and runTest_cu.o
 $(testmain): LIBFLAGS += $(CULIBFLAGSRPATH) # avoid the need for LD_LIBRARY_PATH
 $(testmain): $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) $(GTESTLIBS)
-	$(NVCC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) -lcuda
+ifneq ($(findstring hipcc,$(GPUCC)),) # link fortran/c++/hip using $FC when hipcc is used #802
+	$(FC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) $(CUDATESTFLAGS) -lstdc++ -lpthread  -L$(shell dirname $(shell $(GPUCC) -print-prog-name=clang))/../../lib -lamdhip64
+else
+	$(GPUCC) -o $@ $(cxx_objects_lib) $(cxx_objects_exe) $(cu_objects_lib) $(cu_objects_exe) -ldl $(LIBFLAGS) $(CUDATESTFLAGS)
+endif
 endif
 
 # Use target gtestlibs to build only googletest
@@ -829,9 +920,9 @@ ifeq ($(USECCACHE),1)
 	ccache --version | head -1
 endif
 	@echo ""
-	@echo NVCC=$(NVCC)
-ifneq ($(NVCC),)
-	$(NVCC) --version
+	@echo GPUCC=$(GPUCC)
+ifneq ($(GPUCC),)
+	$(GPUCC) --version
 endif
 	@echo ""
 	@echo CXX=$(CXX)
@@ -850,7 +941,7 @@ endif
 
 # Target: check (run the C++ test executable)
 # [NB THIS IS WHAT IS USED IN THE GITHUB CI!]
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 check: runTest cmpFcheck cmpFGcheck
 else
 check: runTest cmpFcheck
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/fbridge.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/fbridge.cc
index 2d2b36d560..22ce3f5115 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/fbridge.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/fbridge.cc
@@ -1,11 +1,11 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Roiser (Oct 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 
 #include "Bridge.h"
 #include "CPPProcess.h"
-#include "CudaRuntime.h"
+#include "GpuRuntime.h"
 
 extern "C"
 {
@@ -22,7 +22,7 @@ extern "C"
    * Using the same Fortran MadEvent code, linking to the hetrerogeneous library would allow access to both CPU and GPU implementations.
    * The specific heterogeneous configuration (how many GPUs, how many threads on each CPU, etc) could be loaded in CUDA/C++ from a data file.
    */
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -46,8 +46,8 @@ extern "C"
    */
   void fbridgecreate_( CppObjectInFortran** ppbridge, const int* pnevtF, const int* pnparF, const int* pnp4F )
   {
-#ifdef __CUDACC__
-    CudaRuntime::setUp();
+#ifdef MGONGPUCPP_GPUIMPL
+    GpuRuntime::setUp();
 #endif
     // (NB: CPPProcess::initProc no longer needs to be executed here because it is called in the Bridge constructor)
     // FIXME: disable OMP in Bridge when called from Fortran
@@ -65,8 +65,8 @@ extern "C"
     Bridge<FORTRANFPTYPE>* pbridge = dynamic_cast<Bridge<FORTRANFPTYPE>*>( *ppbridge );
     if( pbridge == 0 ) throw std::runtime_error( "fbridgedelete_: invalid Bridge address" );
     delete pbridge;
-#ifdef __CUDACC__
-    CudaRuntime::tearDown();
+#ifdef MGONGPUCPP_GPUIMPL
+    GpuRuntime::tearDown();
 #endif
   }
 
@@ -96,7 +96,7 @@ extern "C"
   {
     Bridge<FORTRANFPTYPE>* pbridge = dynamic_cast<Bridge<FORTRANFPTYPE>*>( *ppbridge );
     if( pbridge == 0 ) throw std::runtime_error( "fbridgesequence_: invalid Bridge address" );
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     // Use the device/GPU implementation in the CUDA library
     // (there is also a host implementation in this library)
     pbridge->gpu_sequence( momenta, gs, rndhel, rndcol, *pchannelId, mes, selhel, selcol );
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/fsampler.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/fsampler.cc
index 2fb445372d..3743934f41 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/fsampler.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/fsampler.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Feb 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #include "mgOnGpuConfig.h"
 
@@ -13,7 +13,7 @@
 
 //--------------------------------------------------------------------------
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -40,7 +40,7 @@ namespace mg5amcCpu
   private:
     const int m_nevt; // The number of events in each iteration
     int m_iiter;      // The iteration counter (for random number seeding)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     HostBufferRndNumMomenta m_hstRndmom; // Memory buffers for random numbers
     HostBufferMomenta m_hstMomenta;      // Memory buffers for momenta
     HostBufferWeights m_hstWeights;      // Memory buffers for sampling weights
@@ -105,7 +105,7 @@ namespace mg5amcCpu
 
 extern "C"
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/runTest.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/runTest.cc
index d4a760a71b..de327f2321 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/runTest.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/runTest.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Hageboeck (Nov 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*xxx to run only testxxx.cc tests
 //----------------------------------------------------------------------------
@@ -18,7 +18,7 @@
 #include "RandomNumberKernels.h"
 #include "epoch_process_id.h"
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using namespace mg5amcGpu;
 #else
 using namespace mg5amcCpu;
@@ -35,7 +35,7 @@ struct CUDA_CPU_TestBase : public TestDriverBase
     : TestDriverBase( npar, refFileName ) {}
 };
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 struct CPUTest : public CUDA_CPU_TestBase
 {
   // Struct data members (process, and memory structures for random numbers, momenta, matrix elements and weights on host and device)
@@ -119,7 +119,7 @@ struct CPUTest : public CUDA_CPU_TestBase
 };
 #endif
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 struct CUDATest : public CUDA_CPU_TestBase
 {
   // Reset the device when our test goes out of scope. Note that this should happen after
@@ -128,7 +128,7 @@ struct CUDATest : public CUDA_CPU_TestBase
   {
     ~DeviceReset()
     {
-      checkCuda( cudaDeviceReset() ); // this is needed by cuda-memcheck --leak-check full
+      checkGpu( gpuDeviceReset() ); // this is needed by cuda-memcheck --leak-check full
     }
   } deviceResetter;
 
@@ -256,7 +256,7 @@ INSTANTIATE_TEST_SUITE_P( prefix, \
                           test_suite_name, \
                           testing::Values( new CUDATest( MG_EPOCH_REFERENCE_FILE_NAME ) ) );
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 MG_INSTANTIATE_TEST_SUITE_GPU( XTESTID_GPU( MG_EPOCH_PROCESS_ID ), MadgraphTest );
 #else
 MG_INSTANTIATE_TEST_SUITE_CPU( XTESTID_CPU( MG_EPOCH_PROCESS_ID ), MadgraphTest );
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/testmisc.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/testmisc.cc
index 895d6eeb56..ba9e59a8a3 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/testmisc.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/testmisc.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*misc to run only testmisc.cc tests
 //----------------------------------------------------------------------------
@@ -17,7 +17,7 @@
 #include <sstream>
 #include <typeinfo>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #define TESTID( s ) s##_GPU_MISC
 #else
 #define TESTID( s ) s##_CPU_MISC
@@ -26,7 +26,7 @@
 #define XTESTID( s ) TESTID( s )
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -59,7 +59,7 @@ namespace mg5amcCpu
 
 TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testmisc )
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/testxxx.cc b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/testxxx.cc
index 3361fe5aa9..e5167de00c 100644
--- a/epochX/cudacpp/pp_tt012j.mad/SubProcesses/testxxx.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/SubProcesses/testxxx.cc
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Apr 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //----------------------------------------------------------------------------
 // Use ./runTest.exe --gtest_filter=*xxx to run only testxxx.cc tests
 //----------------------------------------------------------------------------
@@ -24,7 +24,7 @@
 #include <iomanip>
 #include <iostream>
 #include <vector>
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #define TESTID( s ) s##_GPU_XXX
 #else
 #define TESTID( s ) s##_CPU_XXX
@@ -32,7 +32,7 @@
 
 #define XTESTID( s ) TESTID( s )
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -42,7 +42,7 @@ namespace mg5amcCpu
   int FPEhandlerIevt = -1;
   inline void FPEhandler( int sig )
   {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
     std::cerr << "Floating Point Exception (GPU): '" << FPEhandlerMessage << "' ievt=" << FPEhandlerIevt << std::endl;
 #else
     std::cerr << "Floating Point Exception (CPU neppV=" << neppV << "): '" << FPEhandlerMessage << "' ievt=" << FPEhandlerIevt << std::endl;
@@ -53,7 +53,7 @@ namespace mg5amcCpu
 
 TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
 {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   using namespace mg5amcGpu;
 #else
   using namespace mg5amcCpu;
@@ -77,7 +77,7 @@ TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
   assert( nevt % neppM == 0 ); // nevt must be a multiple of neppM
   assert( nevt % neppV == 0 ); // nevt must be a multiple of neppV
   // Fill in the input momenta
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   mg5amcGpu::PinnedHostBufferMomenta hstMomenta( nevt ); // AOSOA[npagM][npar=4][np4=4][neppM]
 #else
   mg5amcCpu::HostBufferMomenta hstMomenta( nevt ); // AOSOA[npagM][npar=4][np4=4][neppM]
@@ -322,7 +322,7 @@ TEST( XTESTID( MG_EPOCH_PROCESS_ID ), testxxx )
   {
     for( int ievt = 0; ievt < nevt; ievt++ )
     {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
       using namespace mg5amcGpu;
 #else
       using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/pp_tt012j.mad/bin/internal/launch_plugin.py b/epochX/cudacpp/pp_tt012j.mad/bin/internal/launch_plugin.py
index 3b09713e12..90af6b7053 100644
--- a/epochX/cudacpp/pp_tt012j.mad/bin/internal/launch_plugin.py
+++ b/epochX/cudacpp/pp_tt012j.mad/bin/internal/launch_plugin.py
@@ -103,7 +103,7 @@ def default_setup(self):
                        fct_mod=(self.reset_makeopts,(),{}),
                        allowed=['auto', 'none', 'sse4', 'avx2','512y','512z'])
         self.add_param('cudacpp_backend', 'CPP', include=False, hidden=False,
-                       allowed=['Fortan', 'CPP', 'CUDA'])
+                       allowed=['Fortran', 'CPP', 'CUDA'])
         self['vector_size'] = 16 # already setup in default class (just change value)
         self['aloha_flag'] = '--fast-math'
         self['matrix_flag'] = '-O3'
diff --git a/epochX/cudacpp/pp_tt012j.mad/src/HelAmps_sm.h b/epochX/cudacpp/pp_tt012j.mad/src/HelAmps_sm.h
index 8df465ad6d..8b4ad719be 100644
--- a/epochX/cudacpp/pp_tt012j.mad/src/HelAmps_sm.h
+++ b/epochX/cudacpp/pp_tt012j.mad/src/HelAmps_sm.h
@@ -5,7 +5,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -28,7 +28,7 @@
 //#include <iomanip>
 //#include <iostream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/pp_tt012j.mad/src/Parameters_sm.cc b/epochX/cudacpp/pp_tt012j.mad/src/Parameters_sm.cc
index 64fc3fea62..067445b198 100644
--- a/epochX/cudacpp/pp_tt012j.mad/src/Parameters_sm.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/src/Parameters_sm.cc
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: A. Valassi (Sep 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2021-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 // This file has been automatically generated for CUDA/C++ standalone by
 // MadGraph5_aMC@NLO v. 3.5.2_lo_vect, 2023-11-08
@@ -17,7 +17,7 @@
 #include <iomanip>
 #include <iostream>
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 using namespace mg5amcGpu;
 #else
 using namespace mg5amcCpu;
diff --git a/epochX/cudacpp/pp_tt012j.mad/src/Parameters_sm.h b/epochX/cudacpp/pp_tt012j.mad/src/Parameters_sm.h
index b6568d3761..9581d66e0e 100644
--- a/epochX/cudacpp/pp_tt012j.mad/src/Parameters_sm.h
+++ b/epochX/cudacpp/pp_tt012j.mad/src/Parameters_sm.h
@@ -27,7 +27,7 @@
 #include "read_slha.h"
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -93,7 +93,7 @@ namespace mg5amcCpu
 #include <limits>
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -218,7 +218,7 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -239,7 +239,7 @@ namespace mg5amcCpu
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wunused-variable"  // e.g. <<warning: unused variable ‘mdl_G__exp__2’ [-Wunused-variable]>>
 #pragma GCC diagnostic ignored "-Wunused-parameter" // e.g. <<warning: unused parameter ‘G’ [-Wunused-parameter]>>
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #pragma nv_diagnostic push
 #pragma nv_diag_suppress 177 // e.g. <<warning #177-D: variable "mdl_G__exp__2" was declared but never referenced>>
 #endif
@@ -267,7 +267,7 @@ namespace mg5amcCpu
       // End SM implementation - no special handling of vectors of floats as in EFT (#439)
       return out;
     }
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 #pragma GCC diagnostic pop
 #pragma nv_diagnostic pop
 #endif
diff --git a/epochX/cudacpp/pp_tt012j.mad/src/cudacpp_src.mk b/epochX/cudacpp/pp_tt012j.mad/src/cudacpp_src.mk
index d4cc628aec..b2b9da5288 100644
--- a/epochX/cudacpp/pp_tt012j.mad/src/cudacpp_src.mk
+++ b/epochX/cudacpp/pp_tt012j.mad/src/cudacpp_src.mk
@@ -19,7 +19,7 @@ SHELL := /bin/bash
 #=== Configure common compiler flags for CUDA and C++
 
 INCFLAGS = -I.
-OPTFLAGS = -O3 # this ends up in CUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
+OPTFLAGS = -O3 # this ends up in GPUFLAGS too (should it?), cannot add -Ofast or -ffast-math here
 
 #-------------------------------------------------------------------------------
 
@@ -45,13 +45,13 @@ endif
 
 #-------------------------------------------------------------------------------
 
-#=== Configure the CUDA compiler (note: NVCC is already exported including ccache)
+#=== Configure the CUDA compiler (note: GPUCC is already exported including ccache)
 
-###$(info NVCC=$(NVCC))
+###$(info GPUCC=$(GPUCC))
 
 #-------------------------------------------------------------------------------
 
-#=== Configure ccache for C++ builds (note: NVCC is already exported including ccache)
+#=== Configure ccache for C++ builds (note: GPUCC is already exported including ccache)
 
 # Enable ccache if USECCACHE=1
 ifeq ($(USECCACHE)$(shell echo $(CXX) | grep ccache),1)
@@ -92,6 +92,13 @@ endif
 ###$(info OMPFLAGS=$(OMPFLAGS))
 CXXFLAGS += $(OMPFLAGS)
 
+# Add correct flags for nvcc (-x cu) and hipcc (-x hip) for GPU code (see #810)
+ifeq ($(findstring nvcc,$(GPUCC)),nvcc)
+  GPUFLAGS += -Xcompiler -fPIC -c -x cu
+else ifeq ($(findstring hipcc,$(GPUCC)),hipcc)
+  GPUFLAGS += -fPIC -c -x hip
+endif
+
 # Set the build flags appropriate to each AVX choice (example: "make AVX=none")
 # [NB MGONGPU_PVW512 is needed because "-mprefer-vector-width=256" is not exposed in a macro]
 # [See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96476]
@@ -253,20 +260,20 @@ $(BUILDDIR)/%.o : %.cc *.h $(BUILDDIR)/.build.$(TAG)
 # Generic target and build rules: objects from CUDA compilation
 $(BUILDDIR)/%_cu.o : %.cc *.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
-	$(NVCC) $(CPPFLAGS) $(CUFLAGS) -Xcompiler -fPIC -c -x cu $< -o $@
+	$(GPUCC) $(CPPFLAGS) $(GPUFLAGS) $< -o $@
 
 #-------------------------------------------------------------------------------
 
 cxx_objects=$(addprefix $(BUILDDIR)/, Parameters_sm.o read_slha.o)
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 cu_objects=$(addprefix $(BUILDDIR)/, Parameters_sm_cu.o)
 endif
 
 # Target (and build rules): common (src) library
-ifneq ($(NVCC),)
+ifneq ($(GPUCC),)
 $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so : $(cxx_objects) $(cu_objects)
 	@if [ ! -d $(LIBDIR) ]; then echo "mkdir -p $(LIBDIR)"; mkdir -p $(LIBDIR); fi
-	$(NVCC) -shared -o $@ $(cxx_objects) $(cu_objects) $(LDFLAGS)
+	$(GPUCC) -shared -o $@ $(cxx_objects) $(cu_objects) $(LDFLAGS)
 else
 $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so : $(cxx_objects)
 	@if [ ! -d $(LIBDIR) ]; then echo "mkdir -p $(LIBDIR)"; mkdir -p $(LIBDIR); fi
diff --git a/epochX/cudacpp/pp_tt012j.mad/src/mgOnGpuConfig.h b/epochX/cudacpp/pp_tt012j.mad/src/mgOnGpuConfig.h
index 80032e528b..6bde4466d0 100644
--- a/epochX/cudacpp/pp_tt012j.mad/src/mgOnGpuConfig.h
+++ b/epochX/cudacpp/pp_tt012j.mad/src/mgOnGpuConfig.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jul 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUCONFIG_H
 #define MGONGPUCONFIG_H 1
@@ -10,12 +10,26 @@
 // There are two different code bases for standalone_cudacpp (without multichannel) and madevent+cudacpp (with multichannel)
 #define MGONGPU_SUPPORTS_MULTICHANNEL 1
 
+// Is this a GPU (CUDA, HIP) or CPU implementation?
+#ifdef __CUDACC__
+#define MGONGPUCPP_GPUIMPL cuda
+#elif defined __HIPCC__
+#define MGONGPUCPP_GPUIMPL hip
+#include "hip/hip_runtime.h" // needed for blockDim, blockIdx, threadIdx: better in mgOnGpuConfig.h than in GpuAbstraction.h
+#else
+#undef MGONGPUCPP_GPUIMPL
+#endif
+
 // ** NB1 Throughputs (e.g. 6.8E8) are events/sec for "./gcheck.exe -p 65536 128 12"
 // ** NB2 Baseline on b7g47n0004 fluctuates (probably depends on load on other VMs)
 
 // Choose if curand is supported for generating random numbers
+// For HIP, by default, do not use curand (common random numbers will be used instead)
 // For both CUDA and C++, by default, do not inline, but allow this macro to be set from outside with e.g. -DMGONGPU_HAS_NO_CURAND
-// (there exist CUDA installations, e.g. using the HPC package, which do not include curand - see PR #784)
+// (there exist CUDA installations, e.g. using the HPC package, which do not include curand - see PR #784 and #785)
+#if defined __HIPCC__
+#define MGONGPU_HAS_NO_CURAND 1
+#else
 //#ifdef __CUDACC__
 //#undef MGONGPU_HAS_NO_CURAND // default
 ////#define MGONGPU_HAS_NO_CURAND 1
@@ -23,6 +37,7 @@
 //#undef MGONGPU_HAS_NO_CURAND // default
 ////#define MGONGPU_HAS_NO_CURAND 1
 //#endif
+#endif
 
 // Choose floating point precision (for everything but color algebra #537)
 // If one of these macros has been set from outside with e.g. -DMGONGPU_FPTYPE_FLOAT, nothing happens (issue #167)
@@ -54,23 +69,31 @@
 //#undef MGONGPU_HARDCODE_PARAM // default
 ////#define MGONGPU_HARDCODE_PARAM 1
 
-// Complex type in c++: std::complex or cxsmpl (CHOOSE ONLY ONE)
-#ifndef __CUDACC__
-//#define MGONGPU_CPPCXTYPE_STDCOMPLEX 1 // ~8 percent slower on float, same on double (5.1E6/double, 9.4E6/float)
-#define MGONGPU_CPPCXTYPE_CXSMPL 1 // new default (5.1E6/double, 10.2E6/float)
-#endif
-
-// Complex type in cuda: thrust or cucomplex or cxsmpl (CHOOSE ONLY ONE)
+// Complex type in CUDA: thrust or cucomplex or cxsmpl (CHOOSE ONLY ONE)
+// (NB THIS IS MGONGPU_*CU*CXTYPE_xxx)
 #ifdef __CUDACC__
 #define MGONGPU_CUCXTYPE_THRUST 1 // default (~1.15E9/double, ~3.2E9/float)
 //#define MGONGPU_CUCXTYPE_CUCOMPLEX 1 // ~10 percent slower (1.03E9/double, ~2.8E9/float)
 //#define MGONGPU_CUCXTYPE_CXSMPL 1 // ~10 percent slower (1.00E9/double, ~2.9E9/float)
+
+// Complex type in HIP: cxsmpl (ONLY ONE OPTION POSSIBLE? #810)
+// (NB THIS IS MGONGPU_*HIP*CXTYPE_xxx)
+#elif defined __HIPCC__
+#define MGONGPU_HIPCXTYPE_CXSMPL 1 // default for HIP
+
+// Complex type in C++: std::complex or cxsmpl (CHOOSE ONLY ONE)
+// (NB THIS IS MGONGPU_*CPP*CXTYPE_xxx)
+#else
+//#define MGONGPU_CPPCXTYPE_STDCOMPLEX 1 // ~8 percent slower on float, same on double (5.1E6/double, 9.4E6/float)
+#define MGONGPU_CPPCXTYPE_CXSMPL 1 // new default (5.1E6/double, 10.2E6/float)
 #endif
 
-// Cuda nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
+// CUDA nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
 #ifdef __CUDACC__
-#undef MGONGPU_NSIGHT_DEBUG // default
+#undef MGONGPU_NSIGHT_DEBUG // default in CUDA
 //#define MGONGPU_NSIGHT_DEBUG 1
+#else
+#undef MGONGPU_NSIGHT_DEBUG // only option in HIP or C++
 #endif
 
 // SANITY CHECKS (floating point precision for everything but color algebra #537)
@@ -86,17 +109,21 @@
 #error You cannot use double precision for color algebra and single precision elsewhere
 #endif
 
-// SANITY CHECKS (c++ complex number implementation)
-#ifndef __CUDACC__
-#if defined MGONGPU_CPPCXTYPE_STDCOMPLEX and defined MGONGPU_CPPCXTYPE_CXSMPL
-#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CPPCXTYPE_STDCOMPLEX or MGONGPU_CPPCXTYPE_CXSMPL
+// SANITY CHECKS (CUDA complex number implementation)
+#ifdef __CUDACC__
+#if defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CUCOMPLEX
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX for CUDA
+#elif defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CXSMPL for CUDA
+#elif defined MGONGPU_CUCXTYPE_CUCOMPLEX and defined MGONGPU_CUCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE OF MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL for CUDA
 #endif
 #endif
 
-// SANITY CHECKS (cuda complex number implementation)
-#ifdef __CUDACC__
-#if defined MGONGPU_CUCXTYPE_THRUST and defined MGONGPU_CUCXTYPE_CUCOMPLEX and defined MGONGPU_CUCXTYPE_CXSMPL
-#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL
+// SANITY CHECKS (C++ complex number implementation)
+#ifndef MGONGPUCPP_GPUIMPL
+#if defined MGONGPU_CPPCXTYPE_STDCOMPLEX and defined MGONGPU_CPPCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CPPCXTYPE_STDCOMPLEX or MGONGPU_CPPCXTYPE_CXSMPL for C++
 #endif
 #endif
 
@@ -134,7 +161,7 @@ namespace mgOnGpu
   // Alignment requirement for using reinterpret_cast with SIMD vectorized code
   // (using reinterpret_cast with non aligned memory may lead to segmentation faults!)
   // Only needed for C++ code but can be enforced also in NVCC builds of C++ code using CUDA>=11.2 and C++17 (#318, #319, #333)
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
   constexpr int cppAlign = 64; // alignment requirement for SIMD vectorization (64-byte i.e. 512-bit)
 #endif
 
@@ -145,7 +172,7 @@ using mgOnGpu::fptype;
 using mgOnGpu::fptype2;
 
 // C++ SIMD vectorization width (this will be used to set neppV)
-#ifdef __CUDACC__ // CUDA implementation has no SIMD
+#ifdef MGONGPUCPP_GPUIMPL // CUDA and HIP implementations have no SIMD
 #undef MGONGPU_CPPSIMD
 #elif defined __AVX512VL__ && defined MGONGPU_PVW512 // C++ "512z" AVX512 with 512 width (512-bit ie 64-byte): 8 (DOUBLE) or 16 (FLOAT)
 #ifdef MGONGPU_FPTYPE_DOUBLE
@@ -175,9 +202,9 @@ using mgOnGpu::fptype2;
 #undef MGONGPU_CPPSIMD
 #endif
 
-// Cuda nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
+// CUDA nsight compute (ncu) debug: add dummy lines to ease SASS program flow navigation
 // Arguments (not used so far): text is __FUNCTION__, code is 0 (start) or 1 (end)
-#if defined __CUDACC__ && defined MGONGPU_NSIGHT_DEBUG /* clang-format off */
+#if defined __CUDA__ && defined MGONGPU_NSIGHT_DEBUG /* clang-format off */
 #define mgDebugDeclare() __shared__ float mgDebugCounter[mgOnGpu::ntpbMAX];
 #define mgDebugInitialise() { mgDebugCounter[threadIdx.x] = 0; }
 #define mgDebug( code, text ) { mgDebugCounter[threadIdx.x] += 1; }
@@ -189,8 +216,8 @@ using mgOnGpu::fptype2;
 #define mgDebugFinalise() { /*noop*/ }
 #endif /* clang-format on */
 
-// Define empty CUDA declaration specifiers for C++
-#ifndef __CUDACC__
+// Define empty CUDA/HIP declaration specifiers for C++
+#ifndef MGONGPUCPP_GPUIMPL
 #define __global__
 #define __host__
 #define __device__
diff --git a/epochX/cudacpp/pp_tt012j.mad/src/mgOnGpuCxtypes.h b/epochX/cudacpp/pp_tt012j.mad/src/mgOnGpuCxtypes.h
index ca9a9f00c0..7ede1dbfae 100644
--- a/epochX/cudacpp/pp_tt012j.mad/src/mgOnGpuCxtypes.h
+++ b/epochX/cudacpp/pp_tt012j.mad/src/mgOnGpuCxtypes.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022, based on earlier work by D. Smith) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUCXTYPES_H
 #define MGONGPUCXTYPES_H 1
@@ -19,7 +19,7 @@
 #include <complex>
 
 // Complex type in cuda: thrust or cucomplex or cxsmpl
-#ifdef __CUDACC__
+#ifdef __CUDACC__ // this must be __CUDAC__ (not MGONGPUCPP_GPUIMPL)
 #if defined MGONGPU_CUCXTYPE_THRUST
 #pragma clang diagnostic push
 #pragma clang diagnostic ignored "-Wtautological-compare" // for icpx2021/clang13 (https://stackoverflow.com/a/15864661)
@@ -30,8 +30,13 @@
 #elif not defined MGONGPU_CUCXTYPE_CXSMPL
 #error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_CUCXTYPE_THRUST or MGONGPU_CUCXTYPE_CUCOMPLEX or MGONGPU_CUCXTYPE_CXSMPL
 #endif
+// Complex type in HIP: cxsmpl
+#elif defined __HIPCC__
+#if not defined MGONGPU_HIPCXTYPE_CXSMPL
+#error You must CHOOSE (ONE AND) ONLY ONE of MGONGPU_HIPCXTYPE_CXSMPL
+#endif
 #else
-// Complex type in c++: std::complex or cxsmpl
+// Complex type in c++ or HIP: std::complex or cxsmpl
 #if defined MGONGPU_CPPCXTYPE_STDCOMPLEX
 #include <cmath>
 #elif not defined MGONGPU_CPPCXTYPE_CXSMPL
@@ -82,7 +87,7 @@ namespace mgOnGpu /* clang-format off */
 using mgOnGpu::cxsmpl;
 
 // Printout to stream for user defined types
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -92,7 +97,7 @@ namespace mg5amcCpu
   inline __host__ std::ostream&
   operator<<( std::ostream& out, const cxsmpl<FP>& c )
   {
-    out << std::complex( c.real(), c.imag() );
+    out << std::complex<FP>( c.real(), c.imag() );
     return out;
   }
 
@@ -215,14 +220,14 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
   // --- Type definitions (complex type: cxtype)
-#ifdef __CUDACC__ // cuda
+#ifdef __CUDACC__ // this must be __CUDAC__ (not MGONGPUCPP_GPUIMPL)
 #if defined MGONGPU_CUCXTYPE_THRUST
   typedef thrust::complex<fptype> cxtype;
 #elif defined MGONGPU_CUCXTYPE_CUCOMPLEX
@@ -255,13 +260,13 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
-#if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+#if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
 
   //------------------------------
   // CUDA or C++ - using cxsmpl
@@ -303,11 +308,11 @@ namespace mg5amcCpu
     return cxmake( c.real(), c.imag() );
   }
 
-#endif // #if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
+#endif // #if defined MGONGPU_CUCXTYPE_CXSMPL or defined MGONGPU_HIPCXTYPE_CXSMPL or defined MGONGPU_CPPCXTYPE_CXSMPL
 
   //==========================================================================
 
-#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_THRUST // cuda + thrust
+#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_THRUST // cuda + thrust (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // CUDA - using thrust::complex
@@ -347,7 +352,7 @@ namespace mg5amcCpu
 
   //==========================================================================
 
-#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_CUCOMPLEX // cuda + cucomplex
+#if defined __CUDACC__ and defined MGONGPU_CUCXTYPE_CUCOMPLEX // cuda + cucomplex (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // CUDA - using cuComplex
@@ -566,7 +571,7 @@ namespace mg5amcCpu
 
   //==========================================================================
 
-#if not defined __CUDACC__ and defined MGONGPU_CPPCXTYPE_STDCOMPLEX // c++ + stdcomplex
+#if not defined __CUDACC__ and defined MGONGPU_CPPCXTYPE_STDCOMPLEX // c++/hip + stdcomplex (this must be __CUDAC__ and not MGONGPUCPP_GPUIMPL)
 
   //------------------------------
   // C++ - using std::complex
@@ -633,7 +638,7 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
diff --git a/epochX/cudacpp/pp_tt012j.mad/src/mgOnGpuFptypes.h b/epochX/cudacpp/pp_tt012j.mad/src/mgOnGpuFptypes.h
index 905c97d700..fa3a02664b 100644
--- a/epochX/cudacpp/pp_tt012j.mad/src/mgOnGpuFptypes.h
+++ b/epochX/cudacpp/pp_tt012j.mad/src/mgOnGpuFptypes.h
@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi (2022-2023) for the MG5aMC CUDACPP plugin.
 
 #ifndef MGONGPUFPTYPES_H
 #define MGONGPUFPTYPES_H 1
@@ -12,7 +12,7 @@
 #include <cmath>
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL // cuda
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -20,7 +20,7 @@ namespace mg5amcCpu
 {
   //==========================================================================
 
-#ifdef __CUDACC__ // cuda
+#ifdef MGONGPUCPP_GPUIMPL // cuda
 
   //------------------------------
   // Floating point types - Cuda
@@ -64,11 +64,11 @@ namespace mg5amcCpu
 #endif
   }
 
-#endif // #ifdef __CUDACC__
+#endif // #ifdef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
   //------------------------------
   // Floating point types - C++
@@ -92,7 +92,7 @@ namespace mg5amcCpu
     return std::sqrt( f );
   }
 
-#endif // #ifndef __CUDACC__
+#endif // #ifndef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
diff --git a/epochX/cudacpp/pp_tt012j.mad/src/mgOnGpuVectors.h b/epochX/cudacpp/pp_tt012j.mad/src/mgOnGpuVectors.h
index e1299ba81e..cdae04326b 100644
--- a/epochX/cudacpp/pp_tt012j.mad/src/mgOnGpuVectors.h
+++ b/epochX/cudacpp/pp_tt012j.mad/src/mgOnGpuVectors.h
@@ -32,7 +32,7 @@
 #endif
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -131,7 +131,7 @@ namespace mg5amcCpu
 #endif
 #endif
 
-#else // i.e #ifndef MGONGPU_CPPSIMD (this includes #ifdef __CUDACC__)
+#else // i.e #ifndef MGONGPU_CPPSIMD (this includes #ifdef MGONGPUCPP_GPUIMPL)
 
   const int neppV = 1;
 
@@ -153,13 +153,13 @@ namespace mg5amcCpu
 //==========================================================================
 
 // NB: namespaces mg5amcGpu and mg5amcCpu includes types which are defined in different ways for CPU and GPU builds (see #318 and #725)
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
 #endif
 {
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
 
   // Printout to stream for user defined types
 
@@ -805,11 +805,11 @@ namespace mg5amcCpu
 
 #endif // #if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
 
-#endif // #ifndef __CUDACC__
+#endif // #ifndef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 
   //------------------------------
   // Vector types - CUDA
@@ -853,12 +853,12 @@ namespace mg5amcCpu
     return mask;
   }
 
-#endif // #ifdef __CUDACC__
+#endif // #ifdef MGONGPUCPP_GPUIMPL
 
   //==========================================================================
 
   // Scalar-or-vector types: scalar in CUDA, vector or scalar in C++
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
   typedef bool bool_sv;
   typedef fptype fptype_sv;
   typedef fptype2 fptype2_sv;
@@ -879,7 +879,7 @@ namespace mg5amcCpu
 #endif
 
   // Scalar-or-vector zeros: scalar in CUDA, vector or scalar in C++
-#ifdef __CUDACC__ /* clang-format off */
+#ifdef MGONGPUCPP_GPUIMPL /* clang-format off */
   inline __host__ __device__ cxtype cxzero_sv(){ return cxtype( 0, 0 ); }
 #elif defined MGONGPU_CPPSIMD
   inline cxtype_v cxzero_sv() { return cxtype_v(); } // RRRR=0000 IIII=0000
diff --git a/epochX/cudacpp/pp_tt012j.mad/src/rambo.h b/epochX/cudacpp/pp_tt012j.mad/src/rambo.h
index e02ea52496..cd7e1008ea 100644
--- a/epochX/cudacpp/pp_tt012j.mad/src/rambo.h
+++ b/epochX/cudacpp/pp_tt012j.mad/src/rambo.h
@@ -4,7 +4,7 @@
 // Copyright (C) 2020-2023 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Modified by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2023) for the MG5aMC CUDACPP plugin.
 //==========================================================================
 
 #include "mgOnGpuConfig.h"
@@ -18,7 +18,7 @@
 #include <iostream>
 
 // Simplified rambo version for 2 to N (with N>=2) processes with massless particles
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
 namespace mg5amcGpu
 #else
 namespace mg5amcCpu
@@ -83,7 +83,7 @@ namespace mg5amcCpu
       static bool first = true;
       if( first )
       {
-#ifdef __CUDACC__
+#ifdef MGONGPUCPP_GPUIMPL
         if constexpr( M_ACCESS::isOnDevice() ) // avoid
         {
           const int ievt0 = 0;
@@ -166,7 +166,7 @@ namespace mg5amcCpu
     wt = po2log;
     if( nparf != 2 ) wt = ( 2. * nparf - 4. ) * log( energy ) + z[nparf - 1];
 
-#ifndef __CUDACC__
+#ifndef MGONGPUCPP_GPUIMPL
     // issue warnings if weight is too small or too large
     static int iwarn[5] = { 0, 0, 0, 0, 0 };
     if( wt < -180. )
diff --git a/epochX/cudacpp/pp_tt012j.mad/src/read_slha.cc b/epochX/cudacpp/pp_tt012j.mad/src/read_slha.cc
index 055b19a779..f8e46f2e66 100644
--- a/epochX/cudacpp/pp_tt012j.mad/src/read_slha.cc
+++ b/epochX/cudacpp/pp_tt012j.mad/src/read_slha.cc
@@ -11,7 +11,11 @@
 
 #include <algorithm>
 #include <cstdlib>
-#include <filesystem>
+//#ifdef __HIPCC__
+//#include <experimental/filesystem> // see https://rocm.docs.amd.com/en/docs-5.4.3/CHANGELOG.html#id79
+//#else
+//#include <filesystem> // bypass this completely to ease portability on LUMI #803
+//#endif
 #include <fstream>
 #include <iostream>
 
@@ -60,7 +64,14 @@ SLHAReader::read_slha_file( std::string file_name, bool verbose )
     {
       std::cout << "WARNING! Card file '" << file_name << "' does not exist:"
                 << " look for the file in directory $" << envpath << "='" << getenv( envpath ) << "'" << std::endl;
+      /*
+#ifdef __HIPCC__
+      const std::string file_name2 = std::experimental::filesystem::path( getenv( envpath ) ) / std::experimental::filesystem::path( file_name ).filename();
+#else
       const std::string file_name2 = std::filesystem::path( getenv( envpath ) ) / std::filesystem::path( file_name ).filename();
+#endif
+      */
+      const std::string file_name2 = std::string( getenv( envpath ) ) + "/" + file_name; // bypass std::filesystem #803
       param_card.open( file_name2.c_str(), std::ifstream::in );
       if( param_card.good() )
       {
diff --git a/epochX/cudacpp/tmad/logs_eemumu_mad/log_eemumu_mad_d_inl0_hrd0.txt b/epochX/cudacpp/tmad/logs_eemumu_mad/log_eemumu_mad_d_inl0_hrd0.txt
index 67b7aa5182..459e70d382 100644
--- a/epochX/cudacpp/tmad/logs_eemumu_mad/log_eemumu_mad_d_inl0_hrd0.txt
+++ b/epochX/cudacpp/tmad/logs_eemumu_mad/log_eemumu_mad_d_inl0_hrd0.txt
@@ -3,9 +3,9 @@ CUDACPP_BUILDDIR='.'
 
 make USEBUILDDIR=1 AVX=none
 
-
 make USEBUILDDIR=1 AVX=sse4
 
+
 make USEBUILDDIR=1 AVX=avx2
 make USEBUILDDIR=1 AVX=512y
 
@@ -15,15 +15,15 @@ make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum'
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum'
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum'
-CUDACPP_BUILDDIR='build.sse4_d_inl0_hrd0'
 CUDACPP_BUILDDIR='build.avx2_d_inl0_hrd0'
-CUDACPP_BUILDDIR='build.512y_d_inl0_hrd0'
-CUDACPP_BUILDDIR='build.none_d_inl0_hrd0'
+CUDACPP_BUILDDIR='build.512z_d_inl0_hrd0'
+CUDACPP_BUILDDIR='build.sse4_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum'
-CUDACPP_BUILDDIR='build.512z_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum'
+CUDACPP_BUILDDIR='build.none_d_inl0_hrd0'
+CUDACPP_BUILDDIR='build.512y_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum'
 make[1]: Nothing to be done for 'all'.
@@ -33,7 +33,7 @@ make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/
 
 OMP_NUM_THREADS=
 
-DATE: 2023-11-24_15:35:02
+DATE: 2024-01-30_06:09:28
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 Working directory (run): /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum
@@ -59,9 +59,9 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_eemumu_x1_fortran > /tmp/av
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.2175 [0.21747169064681776] fbridge_mode=0
  [UNWEIGHT] Wrote 3893 events (found 7395 events)
- [COUNTERS] PROGRAM TOTAL          :    0.6482s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.6399s
- [COUNTERS] Fortran MEs      ( 1 ) :    0.0083s for     8192 events => throughput is 9.86E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.6491s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.6403s
+ [COUNTERS] Fortran MEs      ( 1 ) :    0.0088s for     8192 events => throughput is 9.31E+05 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x1 (create events.lhe) ***
 --------------------
@@ -84,9 +84,9 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_eemumu_x1_fortran > /tmp/av
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.2175 [0.21747169064681776] fbridge_mode=0
  [UNWEIGHT] Wrote 1611 events (found 1616 events)
- [COUNTERS] PROGRAM TOTAL          :    0.1914s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.1828s
- [COUNTERS] Fortran MEs      ( 1 ) :    0.0087s for     8192 events => throughput is 9.47E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.1850s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.1762s
+ [COUNTERS] Fortran MEs      ( 1 ) :    0.0089s for     8192 events => throughput is 9.22E+05 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x10 (create events.lhe) ***
 --------------------
@@ -107,11 +107,11 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_eemumu_x10_fortran > /tmp/a
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.0915 [9.1501919904813656E-002] fbridge_mode=0
+ [XSECTION] Cross section = 0.0915 [9.1501919904813669E-002] fbridge_mode=0
  [UNWEIGHT] Wrote 1803 events (found 1808 events)
- [COUNTERS] PROGRAM TOTAL          :    0.4504s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3577s
- [COUNTERS] Fortran MEs      ( 1 ) :    0.0927s for    90112 events => throughput is 9.72E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.4417s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3482s
+ [COUNTERS] Fortran MEs      ( 1 ) :    0.0935s for    90112 events => throughput is 9.64E+05 events/s
 
 *** (2-none) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -132,15 +132,15 @@ Executing ' ./build.none_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_eemumu_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.2175 [0.21747169064681776] fbridge_mode=1
+ [XSECTION] Cross section = 0.2175 [0.21747169064681779] fbridge_mode=1
  [UNWEIGHT] Wrote 1611 events (found 1616 events)
- [COUNTERS] PROGRAM TOTAL          :    0.2021s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.1952s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0069s for     8192 events => throughput is 1.19E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.1896s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.1825s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0071s for     8192 events => throughput is 1.15E+06 events/s
 
 *** (2-none) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.21747169064681776) and cpp (0.21747169064681776) differ by less than 2E-14 (0.0)
+OK! xsec from fortran (0.21747169064681776) and cpp (0.21747169064681779) differ by less than 2E-14 (2.220446049250313e-16)
 
 *** (2-none) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -165,29 +165,29 @@ Executing ' ./build.none_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_eemumu_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.0915 [9.1501919904813656E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.0915 [9.1501919904813669E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 1803 events (found 1808 events)
- [COUNTERS] PROGRAM TOTAL          :    0.4477s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3711s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0766s for    90112 events => throughput is 1.18E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.4377s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3566s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0811s for    90112 events => throughput is 1.11E+06 events/s
 
 *** (2-none) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (9.1501919904813656E-002) and cpp (9.1501919904813656E-002) differ by less than 2E-14 (0.0)
+OK! xsec from fortran (9.1501919904813669E-002) and cpp (9.1501919904813669E-002) differ by less than 2E-14 (0.0)
 
 *** (2-none) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.222047e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.115404e+06                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.231835e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.135181e+06                 )  sec^-1
 
 *** (2-sse4) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -208,15 +208,15 @@ Executing ' ./build.sse4_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_eemumu_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.2175 [0.21747169064681776] fbridge_mode=1
+ [XSECTION] Cross section = 0.2175 [0.21747169064681779] fbridge_mode=1
  [UNWEIGHT] Wrote 1611 events (found 1616 events)
- [COUNTERS] PROGRAM TOTAL          :    0.1890s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.1847s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0042s for     8192 events => throughput is 1.93E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.1854s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.1812s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0042s for     8192 events => throughput is 1.96E+06 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.21747169064681776) and cpp (0.21747169064681776) differ by less than 2E-14 (0.0)
+OK! xsec from fortran (0.21747169064681776) and cpp (0.21747169064681779) differ by less than 2E-14 (2.220446049250313e-16)
 
 *** (2-sse4) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -241,29 +241,29 @@ Executing ' ./build.sse4_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_eemumu_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.0915 [9.1501919904813628E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.0915 [9.1501919904813656E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 1803 events (found 1808 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3996s
+ [COUNTERS] PROGRAM TOTAL          :    0.4018s
  [COUNTERS] Fortran Overhead ( 0 ) :    0.3542s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0454s for    90112 events => throughput is 1.98E+06 events/s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0476s for    90112 events => throughput is 1.89E+06 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (9.1501919904813656E-002) and cpp (9.1501919904813628E-002) differ by less than 2E-14 (3.3306690738754696e-16)
+OK! xsec from fortran (9.1501919904813669E-002) and cpp (9.1501919904813656E-002) differ by less than 2E-14 (1.1102230246251565e-16)
 
 *** (2-sse4) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.953284e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.873422e+06                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.976393e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.997339e+06                 )  sec^-1
 
 *** (2-avx2) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -286,9 +286,9 @@ Executing ' ./build.avx2_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_eemumu_x
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.2175 [0.21747169064681776] fbridge_mode=1
  [UNWEIGHT] Wrote 1611 events (found 1616 events)
- [COUNTERS] PROGRAM TOTAL          :    0.1867s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.1835s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0032s for     8192 events => throughput is 2.58E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.1841s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.1809s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0031s for     8192 events => throughput is 2.62E+06 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
@@ -317,29 +317,29 @@ Executing ' ./build.avx2_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_eemumu_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.0915 [9.1501919904813656E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.0915 [9.1501919904813669E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 1803 events (found 1808 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3874s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3527s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0347s for    90112 events => throughput is 2.60E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3915s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3558s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0357s for    90112 events => throughput is 2.53E+06 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (9.1501919904813656E-002) and cpp (9.1501919904813656E-002) differ by less than 2E-14 (0.0)
+OK! xsec from fortran (9.1501919904813669E-002) and cpp (9.1501919904813669E-002) differ by less than 2E-14 (0.0)
 
 *** (2-avx2) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.554304e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.563581e+06                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.675247e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.701936e+06                 )  sec^-1
 
 *** (2-512y) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -362,9 +362,9 @@ Executing ' ./build.512y_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_eemumu_x
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.2175 [0.21747169064681776] fbridge_mode=1
  [UNWEIGHT] Wrote 1611 events (found 1616 events)
- [COUNTERS] PROGRAM TOTAL          :    0.1863s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.1833s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0030s for     8192 events => throughput is 2.70E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.1820s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.1791s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0030s for     8192 events => throughput is 2.77E+06 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
@@ -393,29 +393,29 @@ Executing ' ./build.512y_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_eemumu_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.0915 [9.1501919904813656E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.0915 [9.1501919904813669E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 1803 events (found 1808 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3840s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3511s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0329s for    90112 events => throughput is 2.74E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3886s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3550s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0336s for    90112 events => throughput is 2.68E+06 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (9.1501919904813656E-002) and cpp (9.1501919904813656E-002) differ by less than 2E-14 (0.0)
+OK! xsec from fortran (9.1501919904813669E-002) and cpp (9.1501919904813669E-002) differ by less than 2E-14 (0.0)
 
 *** (2-512y) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.686020e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.733041e+06                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.793515e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.837193e+06                 )  sec^-1
 
 *** (2-512z) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -438,9 +438,9 @@ Executing ' ./build.512z_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_eemumu_x
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.2175 [0.21747169064681776] fbridge_mode=1
  [UNWEIGHT] Wrote 1611 events (found 1616 events)
- [COUNTERS] PROGRAM TOTAL          :    0.1887s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.1850s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0037s for     8192 events => throughput is 2.20E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.1852s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.1819s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0033s for     8192 events => throughput is 2.50E+06 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
@@ -469,29 +469,29 @@ Executing ' ./build.512z_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_eemumu_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.0915 [9.1501919904813656E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.0915 [9.1501919904813669E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 1803 events (found 1808 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3954s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3543s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0411s for    90112 events => throughput is 2.19E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3986s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3589s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0397s for    90112 events => throughput is 2.27E+06 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (9.1501919904813656E-002) and cpp (9.1501919904813656E-002) differ by less than 2E-14 (0.0)
+OK! xsec from fortran (9.1501919904813669E-002) and cpp (9.1501919904813669E-002) differ by less than 2E-14 (0.0)
 
 *** (2-512z) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.045622e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.277912e+06                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.176762e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.388817e+06                 )  sec^-1
 
 *** (3) EXECUTE MADEVENT_CUDA x1 (create events.lhe) ***
 --------------------
@@ -514,9 +514,9 @@ Executing ' ./build.none_d_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_eemumu_
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.2175 [0.21747169064681776] fbridge_mode=1
  [UNWEIGHT] Wrote 1611 events (found 1616 events)
- [COUNTERS] PROGRAM TOTAL          :    0.6011s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.6006s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0005s for     8192 events => throughput is 1.61E+07 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.6311s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.6306s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0005s for     8192 events => throughput is 1.64E+07 events/s
 
 *** (3) Compare MADEVENT_CUDA x1 xsec to MADEVENT_FORTRAN xsec ***
 
@@ -545,58 +545,58 @@ Executing ' ./build.none_d_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_eemumu_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.0915 [9.1501919904813628E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.0915 [9.1501919904813656E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 1803 events (found 1808 events)
- [COUNTERS] PROGRAM TOTAL          :    0.7869s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.7816s
+ [COUNTERS] PROGRAM TOTAL          :    0.7893s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.7841s
  [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0052s for    90112 events => throughput is 1.73E+07 events/s
 
 *** (3) Compare MADEVENT_CUDA x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (9.1501919904813656E-002) and cpp (9.1501919904813628E-002) differ by less than 2E-14 (3.3306690738754696e-16)
+OK! xsec from fortran (9.1501919904813669E-002) and cpp (9.1501919904813656E-002) differ by less than 2E-14 (1.1102230246251565e-16)
 
 *** (3) Compare MADEVENT_CUDA x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cuda.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.145579e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.924722e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.960559e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.934037e+08                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 16384 32 1 --bridge ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.712501e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.691813e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 16384 32 1 ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 7.408619e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.449601e+08                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 4096 128 1 --bridge ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.718171e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.683159e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 4096 128 1 ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 8.071930e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.033957e+08                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 65536 8 1 --bridge ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.709644e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.708340e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 65536 8 1 ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.143512e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.130631e+08                 )  sec^-1
 
 TEST COMPLETED
diff --git a/epochX/cudacpp/tmad/logs_eemumu_mad/log_eemumu_mad_f_inl0_hrd0.txt b/epochX/cudacpp/tmad/logs_eemumu_mad/log_eemumu_mad_f_inl0_hrd0.txt
index 79af27bb3b..161c62cc9b 100644
--- a/epochX/cudacpp/tmad/logs_eemumu_mad/log_eemumu_mad_f_inl0_hrd0.txt
+++ b/epochX/cudacpp/tmad/logs_eemumu_mad/log_eemumu_mad_f_inl0_hrd0.txt
@@ -2,9 +2,9 @@ Working directory (build): /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/e
 CUDACPP_BUILDDIR='.'
 
 
-
 make USEBUILDDIR=1 AVX=none
 
+
 make USEBUILDDIR=1 AVX=sse4
 make USEBUILDDIR=1 AVX=avx2
 make USEBUILDDIR=1 AVX=512y
@@ -15,13 +15,13 @@ make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum'
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum'
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum'
-CUDACPP_BUILDDIR='build.avx2_f_inl0_hrd0'
-CUDACPP_BUILDDIR='build.none_f_inl0_hrd0'
 CUDACPP_BUILDDIR='build.sse4_f_inl0_hrd0'
+CUDACPP_BUILDDIR='build.avx2_f_inl0_hrd0'
 CUDACPP_BUILDDIR='build.512y_f_inl0_hrd0'
 CUDACPP_BUILDDIR='build.512z_f_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum'
+CUDACPP_BUILDDIR='build.none_f_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum'
 make[1]: Nothing to be done for 'all'.
@@ -33,7 +33,7 @@ make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/
 
 OMP_NUM_THREADS=
 
-DATE: 2023-11-24_15:35:20
+DATE: 2024-01-30_06:09:46
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 Working directory (run): /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum
@@ -59,9 +59,9 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_eemumu_x1_fortran > /tmp/av
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.2175 [0.21747169064681776] fbridge_mode=0
  [UNWEIGHT] Wrote 3893 events (found 7395 events)
- [COUNTERS] PROGRAM TOTAL          :    0.6532s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.6452s
- [COUNTERS] Fortran MEs      ( 1 ) :    0.0080s for     8192 events => throughput is 1.02E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.6495s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.6408s
+ [COUNTERS] Fortran MEs      ( 1 ) :    0.0086s for     8192 events => throughput is 9.47E+05 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x1 (create events.lhe) ***
 --------------------
@@ -84,9 +84,9 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_eemumu_x1_fortran > /tmp/av
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.2175 [0.21747169064681776] fbridge_mode=0
  [UNWEIGHT] Wrote 1611 events (found 1616 events)
- [COUNTERS] PROGRAM TOTAL          :    0.1876s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.1795s
- [COUNTERS] Fortran MEs      ( 1 ) :    0.0081s for     8192 events => throughput is 1.01E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.1848s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.1762s
+ [COUNTERS] Fortran MEs      ( 1 ) :    0.0085s for     8192 events => throughput is 9.59E+05 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x10 (create events.lhe) ***
 --------------------
@@ -107,11 +107,11 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_eemumu_x10_fortran > /tmp/a
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.0915 [9.1501919904813656E-002] fbridge_mode=0
+ [XSECTION] Cross section = 0.0915 [9.1501919904813669E-002] fbridge_mode=0
  [UNWEIGHT] Wrote 1803 events (found 1808 events)
- [COUNTERS] PROGRAM TOTAL          :    0.4462s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3548s
- [COUNTERS] Fortran MEs      ( 1 ) :    0.0914s for    90112 events => throughput is 9.85E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.4421s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3487s
+ [COUNTERS] Fortran MEs      ( 1 ) :    0.0934s for    90112 events => throughput is 9.65E+05 events/s
 
 *** (2-none) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -132,15 +132,15 @@ Executing ' ./build.none_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_eemumu_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.2175 [0.21747166087172673] fbridge_mode=1
+ [XSECTION] Cross section = 0.2175 [0.21747165492032638] fbridge_mode=1
  [UNWEIGHT] Wrote 1611 events (found 1616 events)
- [COUNTERS] PROGRAM TOTAL          :    0.1980s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.1915s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0065s for     8192 events => throughput is 1.26E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.1909s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.1840s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0069s for     8192 events => throughput is 1.19E+06 events/s
 
 *** (2-none) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.21747169064681776) and cpp (0.21747166087172673) differ by less than 4E-4 (1.369147908381052e-07)
+OK! xsec from fortran (0.21747169064681776) and cpp (0.21747165492032638) differ by less than 4E-4 (1.6428111293542713e-07)
 
 *** (2-none) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -165,29 +165,29 @@ Executing ' ./build.none_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_eemumu_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.0915 [9.1501907796603360E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.0915 [9.1501905274264717E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 1803 events (found 1808 events)
- [COUNTERS] PROGRAM TOTAL          :    0.4361s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3625s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0735s for    90112 events => throughput is 1.23E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.4341s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3569s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0773s for    90112 events => throughput is 1.17E+06 events/s
 
 *** (2-none) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (9.1501919904813656E-002) and cpp (9.1501907796603360E-002) differ by less than 4E-4 (1.3232739060065057e-07)
+OK! xsec from fortran (9.1501919904813669E-002) and cpp (9.1501905274264717E-002) differ by less than 4E-4 (1.5989335488963974e-07)
 
 *** (2-none) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.253065e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.185144e+06                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.252116e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.215006e+06                 )  sec^-1
 
 *** (2-sse4) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -210,9 +210,9 @@ Executing ' ./build.sse4_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_eemumu_x
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.2175 [0.21747165570339780] fbridge_mode=1
  [UNWEIGHT] Wrote 1611 events (found 1616 events)
- [COUNTERS] PROGRAM TOTAL          :    0.1859s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.1834s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0025s for     8192 events => throughput is 3.23E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.1811s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.1786s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0025s for     8192 events => throughput is 3.30E+06 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
@@ -243,27 +243,27 @@ Executing ' ./build.sse4_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_eemumu_x
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.0915 [9.1501905322826635E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 1803 events (found 1808 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3765s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3483s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0282s for    90112 events => throughput is 3.20E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3824s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3531s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0293s for    90112 events => throughput is 3.08E+06 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (9.1501919904813656E-002) and cpp (9.1501905322826635E-002) differ by less than 4E-4 (1.5936263453308896e-07)
+OK! xsec from fortran (9.1501919904813669E-002) and cpp (9.1501905322826635E-002) differ by less than 4E-4 (1.5936263464411127e-07)
 
 *** (2-sse4) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.183638e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.133589e+06                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.347573e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.329458e+06                 )  sec^-1
 
 *** (2-avx2) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -286,9 +286,9 @@ Executing ' ./build.avx2_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_eemumu_x
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.2175 [0.21747165593922979] fbridge_mode=1
  [UNWEIGHT] Wrote 1611 events (found 1616 events)
- [COUNTERS] PROGRAM TOTAL          :    0.1896s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.1874s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0022s for     8192 events => throughput is 3.64E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.1817s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.1793s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0024s for     8192 events => throughput is 3.43E+06 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
@@ -319,27 +319,27 @@ Executing ' ./build.avx2_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_eemumu_x
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.0915 [9.1501905316084181E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 1803 events (found 1808 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3808s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3552s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0256s for    90112 events => throughput is 3.52E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3770s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3509s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0262s for    90112 events => throughput is 3.44E+06 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (9.1501919904813656E-002) and cpp (9.1501905316084181E-002) differ by less than 4E-4 (1.5943632103443406e-07)
+OK! xsec from fortran (9.1501919904813669E-002) and cpp (9.1501905316084181E-002) differ by less than 4E-4 (1.5943632114545636e-07)
 
 *** (2-avx2) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.438480e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.621496e+06                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.639168e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.716307e+06                 )  sec^-1
 
 *** (2-512y) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -362,9 +362,9 @@ Executing ' ./build.512y_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_eemumu_x
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.2175 [0.21747165593922979] fbridge_mode=1
  [UNWEIGHT] Wrote 1611 events (found 1616 events)
- [COUNTERS] PROGRAM TOTAL          :    0.1913s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.1889s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0023s for     8192 events => throughput is 3.50E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.1846s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.1824s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0023s for     8192 events => throughput is 3.63E+06 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
@@ -395,27 +395,27 @@ Executing ' ./build.512y_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_eemumu_x
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.0915 [9.1501905316084181E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 1803 events (found 1808 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3804s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3553s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0250s for    90112 events => throughput is 3.60E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3795s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3546s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0248s for    90112 events => throughput is 3.63E+06 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (9.1501919904813656E-002) and cpp (9.1501905316084181E-002) differ by less than 4E-4 (1.5943632103443406e-07)
+OK! xsec from fortran (9.1501919904813669E-002) and cpp (9.1501905316084181E-002) differ by less than 4E-4 (1.5943632114545636e-07)
 
 *** (2-512y) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.503584e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.744053e+06                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.715865e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.971296e+06                 )  sec^-1
 
 *** (2-512z) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -436,15 +436,15 @@ Executing ' ./build.512z_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_eemumu_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.2175 [0.21747166440400542] fbridge_mode=1
+ [XSECTION] Cross section = 0.2175 [0.21747166446533123] fbridge_mode=1
  [UNWEIGHT] Wrote 1611 events (found 1616 events)
- [COUNTERS] PROGRAM TOTAL          :    0.1893s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.1871s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0023s for     8192 events => throughput is 3.61E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.1822s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.1798s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0024s for     8192 events => throughput is 3.36E+06 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.21747169064681776) and cpp (0.21747166440400542) differ by less than 4E-4 (1.20672314918302e-07)
+OK! xsec from fortran (0.21747169064681776) and cpp (0.21747166446533123) differ by less than 4E-4 (1.2039032049049325e-07)
 
 *** (2-512z) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -469,29 +469,29 @@ Executing ' ./build.512z_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_eemumu_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.0915 [9.1501908978565555E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.0915 [9.1501908990866423E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 1803 events (found 1808 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3848s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3584s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0264s for    90112 events => throughput is 3.42E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3805s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3541s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0263s for    90112 events => throughput is 3.42E+06 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (9.1501919904813656E-002) and cpp (9.1501908978565555E-002) differ by less than 4E-4 (1.194100419654731e-07)
+OK! xsec from fortran (9.1501919904813669E-002) and cpp (9.1501908990866423E-002) differ by less than 4E-4 (1.1927560927826875e-07)
 
 *** (2-512z) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.231359e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.615840e+06                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.611631e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.903927e+06                 )  sec^-1
 
 *** (3) EXECUTE MADEVENT_CUDA x1 (create events.lhe) ***
 --------------------
@@ -514,9 +514,9 @@ Executing ' ./build.none_f_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_eemumu_
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.2175 [0.21747166823487174] fbridge_mode=1
  [UNWEIGHT] Wrote 1611 events (found 1616 events)
- [COUNTERS] PROGRAM TOTAL          :    0.6022s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.6017s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0005s for     8192 events => throughput is 1.67E+07 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.6073s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.6068s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0005s for     8192 events => throughput is 1.61E+07 events/s
 
 *** (3) Compare MADEVENT_CUDA x1 xsec to MADEVENT_FORTRAN xsec ***
 
@@ -547,56 +547,56 @@ Executing ' ./build.none_f_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_eemumu_
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.0915 [9.1501910542849674E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 1803 events (found 1808 events)
- [COUNTERS] PROGRAM TOTAL          :    0.7749s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.7703s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0046s for    90112 events => throughput is 1.95E+07 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.7841s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.7794s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0048s for    90112 events => throughput is 1.89E+07 events/s
 
 *** (3) Compare MADEVENT_CUDA x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (9.1501919904813656E-002) and cpp (9.1501910542849674E-002) differ by less than 4E-4 (1.0231439939722975e-07)
+OK! xsec from fortran (9.1501919904813669E-002) and cpp (9.1501910542849674E-002) differ by less than 4E-4 (1.0231439961927435e-07)
 
 *** (3) Compare MADEVENT_CUDA x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cuda.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 4.534234e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.032746e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.793140e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.810870e+08                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 16384 32 1 --bridge ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 6.972505e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.874936e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 16384 32 1 ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.054646e+09                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.028452e+09                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 4096 128 1 --bridge ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 6.870406e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.891915e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 4096 128 1 ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.242224e+09                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.234607e+09                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 65536 8 1 --bridge ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 6.358520e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.256002e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 65536 8 1 ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 6.400117e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.441320e+08                 )  sec^-1
 
 TEST COMPLETED
diff --git a/epochX/cudacpp/tmad/logs_eemumu_mad/log_eemumu_mad_m_inl0_hrd0.txt b/epochX/cudacpp/tmad/logs_eemumu_mad/log_eemumu_mad_m_inl0_hrd0.txt
index 52cbc87cca..f51b70af46 100644
--- a/epochX/cudacpp/tmad/logs_eemumu_mad/log_eemumu_mad_m_inl0_hrd0.txt
+++ b/epochX/cudacpp/tmad/logs_eemumu_mad/log_eemumu_mad_m_inl0_hrd0.txt
@@ -1,11 +1,11 @@
 Working directory (build): /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum
 CUDACPP_BUILDDIR='.'
 
-make USEBUILDDIR=1 AVX=none
 
 
-make USEBUILDDIR=1 AVX=sse4
+make USEBUILDDIR=1 AVX=none
 
+make USEBUILDDIR=1 AVX=sse4
 make USEBUILDDIR=1 AVX=avx2
 make USEBUILDDIR=1 AVX=512y
 
@@ -15,11 +15,11 @@ make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum'
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum'
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum'
-CUDACPP_BUILDDIR='build.sse4_m_inl0_hrd0'
-CUDACPP_BUILDDIR='build.512y_m_inl0_hrd0'
 CUDACPP_BUILDDIR='build.512z_m_inl0_hrd0'
-CUDACPP_BUILDDIR='build.avx2_m_inl0_hrd0'
+CUDACPP_BUILDDIR='build.512y_m_inl0_hrd0'
+CUDACPP_BUILDDIR='build.sse4_m_inl0_hrd0'
 CUDACPP_BUILDDIR='build.none_m_inl0_hrd0'
+CUDACPP_BUILDDIR='build.avx2_m_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum'
 make[1]: Nothing to be done for 'all'.
@@ -33,7 +33,7 @@ make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/
 
 OMP_NUM_THREADS=
 
-DATE: 2023-11-24_15:35:37
+DATE: 2024-01-30_06:10:03
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 Working directory (run): /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum
@@ -59,9 +59,9 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_eemumu_x1_fortran > /tmp/av
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.2175 [0.21747169064681776] fbridge_mode=0
  [UNWEIGHT] Wrote 3893 events (found 7395 events)
- [COUNTERS] PROGRAM TOTAL          :    0.6492s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.6412s
- [COUNTERS] Fortran MEs      ( 1 ) :    0.0081s for     8192 events => throughput is 1.02E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.6504s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.6416s
+ [COUNTERS] Fortran MEs      ( 1 ) :    0.0088s for     8192 events => throughput is 9.35E+05 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x1 (create events.lhe) ***
 --------------------
@@ -84,9 +84,9 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_eemumu_x1_fortran > /tmp/av
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.2175 [0.21747169064681776] fbridge_mode=0
  [UNWEIGHT] Wrote 1611 events (found 1616 events)
- [COUNTERS] PROGRAM TOTAL          :    0.1853s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.1770s
- [COUNTERS] Fortran MEs      ( 1 ) :    0.0083s for     8192 events => throughput is 9.84E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.1846s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.1761s
+ [COUNTERS] Fortran MEs      ( 1 ) :    0.0086s for     8192 events => throughput is 9.55E+05 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x10 (create events.lhe) ***
 --------------------
@@ -107,11 +107,11 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_eemumu_x10_fortran > /tmp/a
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.0915 [9.1501919904813656E-002] fbridge_mode=0
+ [XSECTION] Cross section = 0.0915 [9.1501919904813669E-002] fbridge_mode=0
  [UNWEIGHT] Wrote 1803 events (found 1808 events)
- [COUNTERS] PROGRAM TOTAL          :    0.4357s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3463s
- [COUNTERS] Fortran MEs      ( 1 ) :    0.0895s for    90112 events => throughput is 1.01E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.4424s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3488s
+ [COUNTERS] Fortran MEs      ( 1 ) :    0.0936s for    90112 events => throughput is 9.63E+05 events/s
 
 *** (2-none) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -132,15 +132,15 @@ Executing ' ./build.none_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_eemumu_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.2175 [0.21747169074211734] fbridge_mode=1
+ [XSECTION] Cross section = 0.2175 [0.21747169074211736] fbridge_mode=1
  [UNWEIGHT] Wrote 1611 events (found 1616 events)
- [COUNTERS] PROGRAM TOTAL          :    0.1959s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.1891s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0069s for     8192 events => throughput is 1.19E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.1932s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.1857s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0075s for     8192 events => throughput is 1.09E+06 events/s
 
 *** (2-none) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.21747169064681776) and cpp (0.21747169074211734) differ by less than 2E-4 (4.382159080051906e-10)
+OK! xsec from fortran (0.21747169064681776) and cpp (0.21747169074211736) differ by less than 2E-4 (4.3821613004979554e-10)
 
 *** (2-none) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -167,27 +167,27 @@ Executing ' ./build.none_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_eemumu_x
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.0915 [9.1501919915927155E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 1803 events (found 1808 events)
- [COUNTERS] PROGRAM TOTAL          :    0.4287s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3535s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0751s for    90112 events => throughput is 1.20E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.4421s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3600s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0820s for    90112 events => throughput is 1.10E+06 events/s
 
 *** (2-none) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (9.1501919904813656E-002) and cpp (9.1501919915927155E-002) differ by less than 2E-4 (1.214564004925478e-10)
+OK! xsec from fortran (9.1501919904813669E-002) and cpp (9.1501919915927155E-002) differ by less than 2E-4 (1.214564004925478e-10)
 
 *** (2-none) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+BRDHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.181180e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.101615e+06                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.190793e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.125271e+06                 )  sec^-1
 
 *** (2-sse4) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -208,15 +208,15 @@ Executing ' ./build.sse4_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_eemumu_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.2175 [0.21747169074211728] fbridge_mode=1
+ [XSECTION] Cross section = 0.2175 [0.21747169074211734] fbridge_mode=1
  [UNWEIGHT] Wrote 1611 events (found 1616 events)
- [COUNTERS] PROGRAM TOTAL          :    0.1904s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.1863s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0041s for     8192 events => throughput is 1.99E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.1841s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.1799s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0042s for     8192 events => throughput is 1.94E+06 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.21747169064681776) and cpp (0.21747169074211728) differ by less than 2E-4 (4.382156859605857e-10)
+OK! xsec from fortran (0.21747169064681776) and cpp (0.21747169074211734) differ by less than 2E-4 (4.382159080051906e-10)
 
 *** (2-sse4) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -243,27 +243,27 @@ Executing ' ./build.sse4_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_eemumu_x
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.0915 [9.1501919915927155E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 1803 events (found 1808 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3979s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3536s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0443s for    90112 events => throughput is 2.03E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.4024s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3561s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0464s for    90112 events => throughput is 1.94E+06 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (9.1501919904813656E-002) and cpp (9.1501919915927155E-002) differ by less than 2E-4 (1.214564004925478e-10)
+OK! xsec from fortran (9.1501919904813669E-002) and cpp (9.1501919915927155E-002) differ by less than 2E-4 (1.214564004925478e-10)
 
 *** (2-sse4) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+BRDHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.044048e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.982524e+06                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.969679e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.072335e+06                 )  sec^-1
 
 *** (2-avx2) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -286,9 +286,9 @@ Executing ' ./build.avx2_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_eemumu_x
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.2175 [0.21747169063975949] fbridge_mode=1
  [UNWEIGHT] Wrote 1611 events (found 1616 events)
- [COUNTERS] PROGRAM TOTAL          :    0.1875s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.1844s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0031s for     8192 events => throughput is 2.61E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.1864s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.1831s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0033s for     8192 events => throughput is 2.46E+06 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
@@ -319,27 +319,27 @@ Executing ' ./build.avx2_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_eemumu_x
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.0915 [9.1501919908700741E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 1803 events (found 1808 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3864s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3521s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0343s for    90112 events => throughput is 2.63E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3912s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3545s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0368s for    90112 events => throughput is 2.45E+06 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (9.1501919904813656E-002) and cpp (9.1501919908700741E-002) differ by less than 2E-4 (4.2480907680442215e-11)
+OK! xsec from fortran (9.1501919904813669E-002) and cpp (9.1501919908700741E-002) differ by less than 2E-4 (4.248068563583729e-11)
 
 *** (2-avx2) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+BRDHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.597342e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.431608e+06                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.692233e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.633191e+06                 )  sec^-1
 
 *** (2-512y) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -362,9 +362,9 @@ Executing ' ./build.512y_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_eemumu_x
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.2175 [0.21747169063975949] fbridge_mode=1
  [UNWEIGHT] Wrote 1611 events (found 1616 events)
- [COUNTERS] PROGRAM TOTAL          :    0.1873s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.1842s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0030s for     8192 events => throughput is 2.69E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.1842s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.1811s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0031s for     8192 events => throughput is 2.65E+06 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
@@ -395,27 +395,27 @@ Executing ' ./build.512y_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_eemumu_x
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.0915 [9.1501919908700741E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 1803 events (found 1808 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3865s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3532s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0333s for    90112 events => throughput is 2.71E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3888s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3546s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0343s for    90112 events => throughput is 2.63E+06 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (9.1501919904813656E-002) and cpp (9.1501919908700741E-002) differ by less than 2E-4 (4.2480907680442215e-11)
+OK! xsec from fortran (9.1501919904813669E-002) and cpp (9.1501919908700741E-002) differ by less than 2E-4 (4.248068563583729e-11)
 
 *** (2-512y) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+BRDHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.726998e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.754832e+06                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.851400e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.841477e+06                 )  sec^-1
 
 *** (2-512z) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -438,9 +438,9 @@ Executing ' ./build.512z_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_eemumu_x
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.2175 [0.21747169063975949] fbridge_mode=1
  [UNWEIGHT] Wrote 1611 events (found 1616 events)
- [COUNTERS] PROGRAM TOTAL          :    0.1875s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.1837s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0037s for     8192 events => throughput is 2.19E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.1840s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.1805s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0035s for     8192 events => throughput is 2.33E+06 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
@@ -471,27 +471,27 @@ Executing ' ./build.512z_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_eemumu_x
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.0915 [9.1501919908700741E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 1803 events (found 1808 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3941s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3550s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0391s for    90112 events => throughput is 2.31E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3972s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3574s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0398s for    90112 events => throughput is 2.26E+06 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (9.1501919904813656E-002) and cpp (9.1501919908700741E-002) differ by less than 2E-4 (4.2480907680442215e-11)
+OK! xsec from fortran (9.1501919904813669E-002) and cpp (9.1501919908700741E-002) differ by less than 2E-4 (4.248068563583729e-11)
 
 *** (2-512z) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+BRDHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.197385e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.333320e+06                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.341038e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.456277e+06                 )  sec^-1
 
 *** (3) EXECUTE MADEVENT_CUDA x1 (create events.lhe) ***
 --------------------
@@ -514,9 +514,9 @@ Executing ' ./build.none_m_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_eemumu_
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.2175 [0.21747169066587257] fbridge_mode=1
  [UNWEIGHT] Wrote 1611 events (found 1616 events)
- [COUNTERS] PROGRAM TOTAL          :    0.6091s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.6086s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0005s for     8192 events => throughput is 1.65E+07 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.6100s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.6095s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0005s for     8192 events => throughput is 1.62E+07 events/s
 
 *** (3) Compare MADEVENT_CUDA x1 xsec to MADEVENT_FORTRAN xsec ***
 
@@ -547,56 +547,56 @@ Executing ' ./build.none_m_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_eemumu_
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.0915 [9.1501919911173610E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 1803 events (found 1808 events)
- [COUNTERS] PROGRAM TOTAL          :    0.7703s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.7654s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0049s for    90112 events => throughput is 1.84E+07 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.7889s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.7838s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0052s for    90112 events => throughput is 1.74E+07 events/s
 
 *** (3) Compare MADEVENT_CUDA x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (9.1501919904813656E-002) and cpp (9.1501919911173610E-002) differ by less than 2E-4 (6.95061785904727e-11)
+OK! xsec from fortran (9.1501919904813669E-002) and cpp (9.1501919911173610E-002) differ by less than 2E-4 (6.95061785904727e-11)
 
 *** (3) Compare MADEVENT_CUDA x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cuda.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.964035e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.926990e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.893973e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.883911e+08                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 16384 32 1 --bridge ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.636276e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.714682e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 16384 32 1 ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 7.478323e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.463238e+08                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 4096 128 1 --bridge ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.697383e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.709935e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 4096 128 1 ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 7.585042e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.999719e+08                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 65536 8 1 --bridge ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.701325e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.716961e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 65536 8 1 ***
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.163459e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.154425e+08                 )  sec^-1
 
 TEST COMPLETED
diff --git a/epochX/cudacpp/tmad/logs_ggtt_mad/log_ggtt_mad_d_inl0_hrd0.txt b/epochX/cudacpp/tmad/logs_ggtt_mad/log_ggtt_mad_d_inl0_hrd0.txt
index 99bc7401b1..6a2d60f404 100644
--- a/epochX/cudacpp/tmad/logs_ggtt_mad/log_ggtt_mad_d_inl0_hrd0.txt
+++ b/epochX/cudacpp/tmad/logs_ggtt_mad/log_ggtt_mad_d_inl0_hrd0.txt
@@ -2,38 +2,38 @@ Working directory (build): /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/g
 CUDACPP_BUILDDIR='.'
 
 
-make USEBUILDDIR=1 AVX=none
-make USEBUILDDIR=1 AVX=sse4
 
+make USEBUILDDIR=1 AVX=none
 
+make USEBUILDDIR=1 AVX=sse4
 make USEBUILDDIR=1 AVX=avx2
-make USEBUILDDIR=1 AVX=512y
 
+make USEBUILDDIR=1 AVX=512y
 make USEBUILDDIR=1 AVX=512z
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
+CUDACPP_BUILDDIR='build.sse4_d_inl0_hrd0'
 CUDACPP_BUILDDIR='build.none_d_inl0_hrd0'
 CUDACPP_BUILDDIR='build.avx2_d_inl0_hrd0'
 CUDACPP_BUILDDIR='build.512y_d_inl0_hrd0'
-CUDACPP_BUILDDIR='build.sse4_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
+CUDACPP_BUILDDIR='build.512z_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
-CUDACPP_BUILDDIR='build.512z_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
 
 OMP_NUM_THREADS=
 
-DATE: 2023-11-24_15:35:54
+DATE: 2024-01-30_06:10:22
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 Working directory (run): /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx
@@ -59,9 +59,9 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggtt_x1_fortran > /tmp/aval
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 47.69 [47.690708277600116] fbridge_mode=0
  [UNWEIGHT] Wrote 420 events (found 1577 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3685s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3268s
- [COUNTERS] Fortran MEs      ( 1 ) :    0.0417s for     8192 events => throughput is 1.97E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.4078s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3635s
+ [COUNTERS] Fortran MEs      ( 1 ) :    0.0443s for     8192 events => throughput is 1.85E+05 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x1 (create events.lhe) ***
 --------------------
@@ -84,9 +84,9 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggtt_x1_fortran > /tmp/aval
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 47.69 [47.690708277600116] fbridge_mode=0
  [UNWEIGHT] Wrote 434 events (found 1125 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3181s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2769s
- [COUNTERS] Fortran MEs      ( 1 ) :    0.0412s for     8192 events => throughput is 1.99E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3357s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2918s
+ [COUNTERS] Fortran MEs      ( 1 ) :    0.0439s for     8192 events => throughput is 1.87E+05 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x10 (create events.lhe) ***
 --------------------
@@ -107,11 +107,11 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggtt_x10_fortran > /tmp/ava
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 46.22 [46.223782291775365] fbridge_mode=0
+ [XSECTION] Cross section = 46.22 [46.223782291775372] fbridge_mode=0
  [UNWEIGHT] Wrote 1727 events (found 1732 events)
- [COUNTERS] PROGRAM TOTAL          :    1.7190s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.2578s
- [COUNTERS] Fortran MEs      ( 1 ) :    0.4612s for    90112 events => throughput is 1.95E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    1.8736s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.3937s
+ [COUNTERS] Fortran MEs      ( 1 ) :    0.4799s for    90112 events => throughput is 1.88E+05 events/s
 
 *** (2-none) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -134,9 +134,9 @@ Executing ' ./build.none_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggtt_x1_
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 47.69 [47.690708277600102] fbridge_mode=1
  [UNWEIGHT] Wrote 434 events (found 1125 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3623s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3233s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0391s for     8192 events => throughput is 2.10E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3701s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3307s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0394s for     8192 events => throughput is 2.08E+05 events/s
 
 *** (2-none) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
@@ -165,29 +165,29 @@ Executing ' ./build.none_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggtt_x10
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 46.22 [46.223782291775372] fbridge_mode=1
+ [XSECTION] Cross section = 46.22 [46.223782291775379] fbridge_mode=1
  [UNWEIGHT] Wrote 1727 events (found 1732 events)
- [COUNTERS] PROGRAM TOTAL          :    1.7507s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.3257s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.4250s for    90112 events => throughput is 2.12E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    1.8532s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.4213s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.4319s for    90112 events => throughput is 2.09E+05 events/s
 
 *** (2-none) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (46.223782291775365) and cpp (46.223782291775372) differ by less than 2E-14 (2.220446049250313e-16)
+OK! xsec from fortran (46.223782291775372) and cpp (46.223782291775379) differ by less than 2E-14 (2.220446049250313e-16)
 
 *** (2-none) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.132082e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.120291e+05                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.162141e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.118287e+05                 )  sec^-1
 
 *** (2-sse4) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -208,15 +208,15 @@ Executing ' ./build.sse4_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggtt_x1_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 47.69 [47.690708277600102] fbridge_mode=1
+ [XSECTION] Cross section = 47.69 [47.690708277600109] fbridge_mode=1
  [UNWEIGHT] Wrote 434 events (found 1125 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3276s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3055s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0221s for     8192 events => throughput is 3.70E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3362s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3135s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0226s for     8192 events => throughput is 3.62E+05 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (47.690708277600116) and cpp (47.690708277600102) differ by less than 2E-14 (3.3306690738754696e-16)
+OK! xsec from fortran (47.690708277600116) and cpp (47.690708277600109) differ by less than 2E-14 (1.1102230246251565e-16)
 
 *** (2-sse4) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -243,27 +243,27 @@ Executing ' ./build.sse4_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggtt_x10
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 46.22 [46.223782291775379] fbridge_mode=1
  [UNWEIGHT] Wrote 1727 events (found 1732 events)
- [COUNTERS] PROGRAM TOTAL          :    1.5423s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.2971s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.2451s for    90112 events => throughput is 3.68E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    1.6510s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.4023s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.2487s for    90112 events => throughput is 3.62E+05 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (46.223782291775365) and cpp (46.223782291775379) differ by less than 2E-14 (2.220446049250313e-16)
+OK! xsec from fortran (46.223782291775372) and cpp (46.223782291775379) differ by less than 2E-14 (2.220446049250313e-16)
 
 *** (2-sse4) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.538888e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.657266e+05                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.694418e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.745669e+05                 )  sec^-1
 
 *** (2-avx2) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -286,9 +286,9 @@ Executing ' ./build.avx2_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggtt_x1_
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 47.69 [47.690708277600109] fbridge_mode=1
  [UNWEIGHT] Wrote 434 events (found 1125 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3041s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2904s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0136s for     8192 events => throughput is 6.02E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3190s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3046s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0144s for     8192 events => throughput is 5.69E+05 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
@@ -317,29 +317,29 @@ Executing ' ./build.avx2_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggtt_x10
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 46.22 [46.223782291775386] fbridge_mode=1
+ [XSECTION] Cross section = 46.22 [46.223782291775393] fbridge_mode=1
  [UNWEIGHT] Wrote 1727 events (found 1732 events)
- [COUNTERS] PROGRAM TOTAL          :    1.4320s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.2832s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.1488s for    90112 events => throughput is 6.05E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    1.5517s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.3928s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.1589s for    90112 events => throughput is 5.67E+05 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (46.223782291775365) and cpp (46.223782291775386) differ by less than 2E-14 (4.440892098500626e-16)
+OK! xsec from fortran (46.223782291775372) and cpp (46.223782291775393) differ by less than 2E-14 (4.440892098500626e-16)
 
 *** (2-avx2) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 6.020066e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.690369e+05                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 5.978626e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.998122e+05                 )  sec^-1
 
 *** (2-512y) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -362,9 +362,9 @@ Executing ' ./build.512y_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggtt_x1_
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 47.69 [47.690708277600109] fbridge_mode=1
  [UNWEIGHT] Wrote 434 events (found 1125 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3050s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2930s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0119s for     8192 events => throughput is 6.86E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3210s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3086s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0123s for     8192 events => throughput is 6.64E+05 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
@@ -393,29 +393,29 @@ Executing ' ./build.512y_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggtt_x10
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 46.22 [46.223782291775386] fbridge_mode=1
+ [XSECTION] Cross section = 46.22 [46.223782291775393] fbridge_mode=1
  [UNWEIGHT] Wrote 1727 events (found 1732 events)
- [COUNTERS] PROGRAM TOTAL          :    1.4144s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.2795s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.1349s for    90112 events => throughput is 6.68E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    1.5235s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.3878s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.1357s for    90112 events => throughput is 6.64E+05 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (46.223782291775365) and cpp (46.223782291775386) differ by less than 2E-14 (4.440892098500626e-16)
+OK! xsec from fortran (46.223782291775372) and cpp (46.223782291775393) differ by less than 2E-14 (4.440892098500626e-16)
 
 *** (2-512y) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 6.660341e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.740357e+05                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 6.696167e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.934348e+05                 )  sec^-1
 
 *** (2-512z) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -438,9 +438,9 @@ Executing ' ./build.512z_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggtt_x1_
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 47.69 [47.690708277600109] fbridge_mode=1
  [UNWEIGHT] Wrote 434 events (found 1125 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3346s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3127s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0220s for     8192 events => throughput is 3.73E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3284s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3089s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0194s for     8192 events => throughput is 4.22E+05 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
@@ -469,29 +469,29 @@ Executing ' ./build.512z_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggtt_x10
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 46.22 [46.223782291775386] fbridge_mode=1
+ [XSECTION] Cross section = 46.22 [46.223782291775393] fbridge_mode=1
  [UNWEIGHT] Wrote 1727 events (found 1732 events)
- [COUNTERS] PROGRAM TOTAL          :    1.5228s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.2980s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.2249s for    90112 events => throughput is 4.01E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    1.6116s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.3942s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.2174s for    90112 events => throughput is 4.15E+05 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (46.223782291775365) and cpp (46.223782291775386) differ by less than 2E-14 (4.440892098500626e-16)
+OK! xsec from fortran (46.223782291775372) and cpp (46.223782291775393) differ by less than 2E-14 (4.440892098500626e-16)
 
 *** (2-512z) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.916207e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.190038e+05                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.910036e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.219873e+05                 )  sec^-1
 
 *** (3) EXECUTE MADEVENT_CUDA x1 (create events.lhe) ***
 --------------------
@@ -514,9 +514,9 @@ Executing ' ./build.none_d_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_ggtt_x1
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 47.69 [47.690708277600109] fbridge_mode=1
  [UNWEIGHT] Wrote 434 events (found 1125 events)
- [COUNTERS] PROGRAM TOTAL          :    0.7194s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.7188s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0006s for     8192 events => throughput is 1.41E+07 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.7212s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.7206s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0006s for     8192 events => throughput is 1.39E+07 events/s
 
 *** (3) Compare MADEVENT_CUDA x1 xsec to MADEVENT_FORTRAN xsec ***
 
@@ -545,58 +545,58 @@ Executing ' ./build.none_d_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_ggtt_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 46.22 [46.223782291775386] fbridge_mode=1
+ [XSECTION] Cross section = 46.22 [46.223782291775393] fbridge_mode=1
  [UNWEIGHT] Wrote 1727 events (found 1732 events)
- [COUNTERS] PROGRAM TOTAL          :    1.6992s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.6928s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0064s for    90112 events => throughput is 1.41E+07 events/s
+ [COUNTERS] PROGRAM TOTAL          :    1.8076s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.8005s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0071s for    90112 events => throughput is 1.27E+07 events/s
 
 *** (3) Compare MADEVENT_CUDA x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (46.223782291775365) and cpp (46.223782291775386) differ by less than 2E-14 (4.440892098500626e-16)
+OK! xsec from fortran (46.223782291775372) and cpp (46.223782291775393) differ by less than 2E-14 (4.440892098500626e-16)
 
 *** (3) Compare MADEVENT_CUDA x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cuda.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.057278e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.036838e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 5.698426e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.660860e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 16384 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.013072e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.989378e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 16384 32 1 ***
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.074632e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.069849e+08                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 4096 128 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.996049e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.996952e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 4096 128 1 ***
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.152448e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.150687e+08                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 65536 8 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.011908e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.991689e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 65536 8 1 ***
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.029442e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.999027e+07                 )  sec^-1
 
 TEST COMPLETED
diff --git a/epochX/cudacpp/tmad/logs_ggtt_mad/log_ggtt_mad_f_inl0_hrd0.txt b/epochX/cudacpp/tmad/logs_ggtt_mad/log_ggtt_mad_f_inl0_hrd0.txt
index 61d36152df..fe11b37e1c 100644
--- a/epochX/cudacpp/tmad/logs_ggtt_mad/log_ggtt_mad_f_inl0_hrd0.txt
+++ b/epochX/cudacpp/tmad/logs_ggtt_mad/log_ggtt_mad_f_inl0_hrd0.txt
@@ -1,9 +1,9 @@
 Working directory (build): /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx
 CUDACPP_BUILDDIR='.'
 
-make USEBUILDDIR=1 AVX=none
 
 
+make USEBUILDDIR=1 AVX=none
 
 make USEBUILDDIR=1 AVX=sse4
 make USEBUILDDIR=1 AVX=avx2
@@ -15,15 +15,15 @@ make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
-CUDACPP_BUILDDIR='build.512z_f_inl0_hrd0'
-CUDACPP_BUILDDIR='build.avx2_f_inl0_hrd0'
-CUDACPP_BUILDDIR='build.512y_f_inl0_hrd0'
 CUDACPP_BUILDDIR='build.sse4_f_inl0_hrd0'
 CUDACPP_BUILDDIR='build.none_f_inl0_hrd0'
+CUDACPP_BUILDDIR='build.avx2_f_inl0_hrd0'
+CUDACPP_BUILDDIR='build.512y_f_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
+CUDACPP_BUILDDIR='build.512z_f_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
 make[1]: Nothing to be done for 'all'.
@@ -33,7 +33,7 @@ make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/
 
 OMP_NUM_THREADS=
 
-DATE: 2023-11-24_15:36:21
+DATE: 2024-01-30_06:10:49
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 Working directory (run): /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx
@@ -59,9 +59,9 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggtt_x1_fortran > /tmp/aval
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 47.69 [47.690708277600116] fbridge_mode=0
  [UNWEIGHT] Wrote 420 events (found 1577 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3714s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3296s
- [COUNTERS] Fortran MEs      ( 1 ) :    0.0418s for     8192 events => throughput is 1.96E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3836s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3391s
+ [COUNTERS] Fortran MEs      ( 1 ) :    0.0445s for     8192 events => throughput is 1.84E+05 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x1 (create events.lhe) ***
 --------------------
@@ -84,9 +84,9 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggtt_x1_fortran > /tmp/aval
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 47.69 [47.690708277600116] fbridge_mode=0
  [UNWEIGHT] Wrote 434 events (found 1125 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3233s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2806s
- [COUNTERS] Fortran MEs      ( 1 ) :    0.0427s for     8192 events => throughput is 1.92E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3381s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2936s
+ [COUNTERS] Fortran MEs      ( 1 ) :    0.0445s for     8192 events => throughput is 1.84E+05 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x10 (create events.lhe) ***
 --------------------
@@ -107,11 +107,11 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggtt_x10_fortran > /tmp/ava
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 46.22 [46.223782291775365] fbridge_mode=0
+ [XSECTION] Cross section = 46.22 [46.223782291775372] fbridge_mode=0
  [UNWEIGHT] Wrote 1727 events (found 1732 events)
- [COUNTERS] PROGRAM TOTAL          :    1.7540s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.2823s
- [COUNTERS] Fortran MEs      ( 1 ) :    0.4717s for    90112 events => throughput is 1.91E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    1.8752s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.3952s
+ [COUNTERS] Fortran MEs      ( 1 ) :    0.4799s for    90112 events => throughput is 1.88E+05 events/s
 
 *** (2-none) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -132,15 +132,15 @@ Executing ' ./build.none_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggtt_x1_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 47.69 [47.690706767555099] fbridge_mode=1
+ [XSECTION] Cross section = 47.69 [47.690703999052587] fbridge_mode=1
  [UNWEIGHT] Wrote 434 events (found 1125 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3524s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3167s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0356s for     8192 events => throughput is 2.30E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3627s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3261s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0366s for     8192 events => throughput is 2.24E+05 events/s
 
 *** (2-none) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (47.690708277600116) and cpp (47.690706767555099) differ by less than 4E-4 (3.1663296096162696e-08)
+OK! xsec from fortran (47.690708277600116) and cpp (47.690703999052587) differ by less than 4E-4 (8.971448917094449e-08)
 
 *** (2-none) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -165,29 +165,29 @@ Executing ' ./build.none_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggtt_x10
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 46.22 [46.223782605295497] fbridge_mode=1
+ [XSECTION] Cross section = 46.22 [46.223780103711483] fbridge_mode=1
  [UNWEIGHT] Wrote 1727 events (found 1732 events)
- [COUNTERS] PROGRAM TOTAL          :    1.6953s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.3044s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.3909s for    90112 events => throughput is 2.31E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    1.8092s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.4081s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.4011s for    90112 events => throughput is 2.25E+05 events/s
 
 *** (2-none) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (46.223782291775365) and cpp (46.223782605295497) differ by less than 4E-4 (6.782658656945273e-09)
+OK! xsec from fortran (46.223782291775372) and cpp (46.223780103711483) differ by less than 4E-4 (4.733632297249102e-08)
 
 *** (2-none) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.343511e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.286506e+05                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.350968e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.292834e+05                 )  sec^-1
 
 *** (2-sse4) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -208,15 +208,15 @@ Executing ' ./build.sse4_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggtt_x1_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 47.69 [47.690702885183541] fbridge_mode=1
+ [XSECTION] Cross section = 47.69 [47.690699958440689] fbridge_mode=1
  [UNWEIGHT] Wrote 434 events (found 1125 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3092s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2943s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0149s for     8192 events => throughput is 5.49E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3194s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3043s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0151s for     8192 events => throughput is 5.42E+05 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (47.690708277600116) and cpp (47.690702885183541) differ by less than 4E-4 (1.1307059111231865e-07)
+OK! xsec from fortran (47.690708277600116) and cpp (47.690699958440689) differ by less than 4E-4 (1.744398380187917e-07)
 
 *** (2-sse4) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -241,29 +241,29 @@ Executing ' ./build.sse4_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggtt_x10
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 46.22 [46.223778858016772] fbridge_mode=1
+ [XSECTION] Cross section = 46.22 [46.223776162337749] fbridge_mode=1
  [UNWEIGHT] Wrote 1727 events (found 1732 events)
- [COUNTERS] PROGRAM TOTAL          :    1.4474s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.2803s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.1671s for    90112 events => throughput is 5.39E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    1.5555s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.3898s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.1657s for    90112 events => throughput is 5.44E+05 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (46.223782291775365) and cpp (46.223778858016772) differ by less than 4E-4 (7.428553927546488e-08)
+OK! xsec from fortran (46.223782291775372) and cpp (46.223776162337749) differ by less than 4E-4 (1.326035499182865e-07)
 
 *** (2-sse4) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 5.223130e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.487444e+05                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 5.234366e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.523247e+05                 )  sec^-1
 
 *** (2-avx2) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -284,15 +284,15 @@ Executing ' ./build.avx2_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggtt_x1_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 47.69 [47.690694374060818] fbridge_mode=1
+ [XSECTION] Cross section = 47.69 [47.690691653203835] fbridge_mode=1
  [UNWEIGHT] Wrote 434 events (found 1125 events)
- [COUNTERS] PROGRAM TOTAL          :    0.2951s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2868s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0083s for     8192 events => throughput is 9.86E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3066s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2985s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0082s for     8192 events => throughput is 1.00E+06 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (47.690708277600116) and cpp (47.690694374060818) differ by less than 4E-4 (2.9153560099359765e-07)
+OK! xsec from fortran (47.690708277600116) and cpp (47.690691653203835) differ by less than 4E-4 (3.48587741338946e-07)
 
 *** (2-avx2) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -317,29 +317,29 @@ Executing ' ./build.avx2_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggtt_x10
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 46.22 [46.223775951815753] fbridge_mode=1
+ [XSECTION] Cross section = 46.22 [46.223773576247488] fbridge_mode=1
  [UNWEIGHT] Wrote 1727 events (found 1732 events)
- [COUNTERS] PROGRAM TOTAL          :    1.3687s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.2793s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0894s for    90112 events => throughput is 1.01E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    1.4802s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.3878s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0925s for    90112 events => throughput is 9.74E+05 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (46.223782291775365) and cpp (46.223775951815753) differ by less than 4E-4 (1.3715795843527445e-07)
+OK! xsec from fortran (46.223782291775372) and cpp (46.223773576247488) differ by less than 4E-4 (1.885507298071687e-07)
 
 *** (2-avx2) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 9.987231e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.007163e+06                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.011425e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.012972e+06                 )  sec^-1
 
 *** (2-512y) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -360,15 +360,15 @@ Executing ' ./build.512y_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggtt_x1_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 47.69 [47.690694374060818] fbridge_mode=1
+ [XSECTION] Cross section = 47.69 [47.690691653203835] fbridge_mode=1
  [UNWEIGHT] Wrote 434 events (found 1125 events)
- [COUNTERS] PROGRAM TOTAL          :    0.2935s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2860s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0075s for     8192 events => throughput is 1.10E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3042s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2966s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0076s for     8192 events => throughput is 1.08E+06 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (47.690708277600116) and cpp (47.690694374060818) differ by less than 4E-4 (2.9153560099359765e-07)
+OK! xsec from fortran (47.690708277600116) and cpp (47.690691653203835) differ by less than 4E-4 (3.48587741338946e-07)
 
 *** (2-512y) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -393,29 +393,29 @@ Executing ' ./build.512y_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggtt_x10
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 46.22 [46.223775951815753] fbridge_mode=1
+ [XSECTION] Cross section = 46.22 [46.223773576247488] fbridge_mode=1
  [UNWEIGHT] Wrote 1727 events (found 1732 events)
- [COUNTERS] PROGRAM TOTAL          :    1.3676s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.2860s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0816s for    90112 events => throughput is 1.10E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    1.4699s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.3862s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0836s for    90112 events => throughput is 1.08E+06 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (46.223782291775365) and cpp (46.223775951815753) differ by less than 4E-4 (1.3715795843527445e-07)
+OK! xsec from fortran (46.223782291775372) and cpp (46.223773576247488) differ by less than 4E-4 (1.885507298071687e-07)
 
 *** (2-512y) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.067146e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.018834e+06                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.108316e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.101159e+06                 )  sec^-1
 
 *** (2-512z) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -436,15 +436,15 @@ Executing ' ./build.512z_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggtt_x1_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 47.69 [47.690698914467276] fbridge_mode=1
+ [XSECTION] Cross section = 47.69 [47.690698822141186] fbridge_mode=1
  [UNWEIGHT] Wrote 434 events (found 1125 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3028s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2920s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0108s for     8192 events => throughput is 7.58E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3134s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3023s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0112s for     8192 events => throughput is 7.33E+05 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (47.690708277600116) and cpp (47.690698914467276) differ by less than 4E-4 (1.9633033720989346e-07)
+OK! xsec from fortran (47.690708277600116) and cpp (47.690698822141186) differ by less than 4E-4 (1.982662718447159e-07)
 
 *** (2-512z) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -469,29 +469,29 @@ Executing ' ./build.512z_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggtt_x10
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 46.22 [46.223780273983500] fbridge_mode=1
+ [XSECTION] Cross section = 46.22 [46.223780266165058] fbridge_mode=1
  [UNWEIGHT] Wrote 1727 events (found 1732 events)
- [COUNTERS] PROGRAM TOTAL          :    1.4045s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.2883s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.1163s for    90112 events => throughput is 7.75E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    1.5088s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.3876s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.1212s for    90112 events => throughput is 7.43E+05 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (46.223782291775365) and cpp (46.223780273983500) differ by less than 4E-4 (4.3652677583772004e-08)
+OK! xsec from fortran (46.223782291775372) and cpp (46.223780266165058) differ by less than 4E-4 (4.382182106077437e-08)
 
 *** (2-512z) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 7.363750e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.590700e+05                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 7.388485e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.687831e+05                 )  sec^-1
 
 *** (3) EXECUTE MADEVENT_CUDA x1 (create events.lhe) ***
 --------------------
@@ -512,15 +512,15 @@ Executing ' ./build.none_f_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_ggtt_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 47.69 [47.690703397697980] fbridge_mode=1
+ [XSECTION] Cross section = 47.69 [47.690703397697987] fbridge_mode=1
  [UNWEIGHT] Wrote 434 events (found 1125 events)
- [COUNTERS] PROGRAM TOTAL          :    0.7047s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.7042s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0005s for     8192 events => throughput is 1.50E+07 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.7336s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.7331s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0006s for     8192 events => throughput is 1.45E+07 events/s
 
 *** (3) Compare MADEVENT_CUDA x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (47.690708277600116) and cpp (47.690703397697980) differ by less than 4E-4 (1.0232396019382861e-07)
+OK! xsec from fortran (47.690708277600116) and cpp (47.690703397697987) differ by less than 4E-4 (1.0232396008280631e-07)
 
 *** (3) Compare MADEVENT_CUDA x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -547,56 +547,56 @@ Executing ' ./build.none_f_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_ggtt_x1
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 46.22 [46.223786763175951] fbridge_mode=1
  [UNWEIGHT] Wrote 1727 events (found 1732 events)
- [COUNTERS] PROGRAM TOTAL          :    1.7059s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.7003s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0055s for    90112 events => throughput is 1.63E+07 events/s
+ [COUNTERS] PROGRAM TOTAL          :    1.8220s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.8158s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0061s for    90112 events => throughput is 1.47E+07 events/s
 
 *** (3) Compare MADEVENT_CUDA x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (46.223782291775365) and cpp (46.223786763175951) differ by less than 4E-4 (9.673376699659286e-08)
+OK! xsec from fortran (46.223782291775372) and cpp (46.223786763175951) differ by less than 4E-4 (9.673376677454826e-08)
 
 *** (3) Compare MADEVENT_CUDA x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cuda.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.180729e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.211265e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 7.252831e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.993599e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 16384 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 5.767543e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.733080e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 16384 32 1 ***
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.772780e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.767769e+08                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 4096 128 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 5.790178e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.726692e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 4096 128 1 ***
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.842637e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.882266e+08                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 65536 8 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 5.336409e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.370639e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 65536 8 1 ***
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 8.382414e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.407782e+07                 )  sec^-1
 
 TEST COMPLETED
diff --git a/epochX/cudacpp/tmad/logs_ggtt_mad/log_ggtt_mad_m_inl0_hrd0.txt b/epochX/cudacpp/tmad/logs_ggtt_mad/log_ggtt_mad_m_inl0_hrd0.txt
index 0673f7e59b..a855e5b8c2 100644
--- a/epochX/cudacpp/tmad/logs_ggtt_mad/log_ggtt_mad_m_inl0_hrd0.txt
+++ b/epochX/cudacpp/tmad/logs_ggtt_mad/log_ggtt_mad_m_inl0_hrd0.txt
@@ -1,11 +1,11 @@
 Working directory (build): /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx
 CUDACPP_BUILDDIR='.'
 
-make USEBUILDDIR=1 AVX=none
-
 
 
+make USEBUILDDIR=1 AVX=none
 make USEBUILDDIR=1 AVX=sse4
+
 make USEBUILDDIR=1 AVX=avx2
 make USEBUILDDIR=1 AVX=512y
 
@@ -15,15 +15,15 @@ make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
+CUDACPP_BUILDDIR='build.sse4_m_inl0_hrd0'
 CUDACPP_BUILDDIR='build.512y_m_inl0_hrd0'
+CUDACPP_BUILDDIR='build.512z_m_inl0_hrd0'
 CUDACPP_BUILDDIR='build.none_m_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
-CUDACPP_BUILDDIR='build.512z_m_inl0_hrd0'
+CUDACPP_BUILDDIR='build.avx2_m_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
-CUDACPP_BUILDDIR='build.sse4_m_inl0_hrd0'
-CUDACPP_BUILDDIR='build.avx2_m_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
 make[1]: Nothing to be done for 'all'.
@@ -33,7 +33,7 @@ make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/
 
 OMP_NUM_THREADS=
 
-DATE: 2023-11-24_15:36:46
+DATE: 2024-01-30_06:11:16
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 Working directory (run): /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx
@@ -59,9 +59,9 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggtt_x1_fortran > /tmp/aval
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 47.69 [47.690708277600116] fbridge_mode=0
  [UNWEIGHT] Wrote 420 events (found 1577 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3604s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3188s
- [COUNTERS] Fortran MEs      ( 1 ) :    0.0416s for     8192 events => throughput is 1.97E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3775s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3336s
+ [COUNTERS] Fortran MEs      ( 1 ) :    0.0439s for     8192 events => throughput is 1.87E+05 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x1 (create events.lhe) ***
 --------------------
@@ -84,9 +84,9 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggtt_x1_fortran > /tmp/aval
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 47.69 [47.690708277600116] fbridge_mode=0
  [UNWEIGHT] Wrote 434 events (found 1125 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3181s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2759s
- [COUNTERS] Fortran MEs      ( 1 ) :    0.0422s for     8192 events => throughput is 1.94E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3333s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2891s
+ [COUNTERS] Fortran MEs      ( 1 ) :    0.0441s for     8192 events => throughput is 1.86E+05 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x10 (create events.lhe) ***
 --------------------
@@ -107,11 +107,11 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggtt_x10_fortran > /tmp/ava
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 46.22 [46.223782291775365] fbridge_mode=0
+ [XSECTION] Cross section = 46.22 [46.223782291775372] fbridge_mode=0
  [UNWEIGHT] Wrote 1727 events (found 1732 events)
- [COUNTERS] PROGRAM TOTAL          :    1.7030s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.2436s
- [COUNTERS] Fortran MEs      ( 1 ) :    0.4593s for    90112 events => throughput is 1.96E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    1.8813s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.4017s
+ [COUNTERS] Fortran MEs      ( 1 ) :    0.4796s for    90112 events => throughput is 1.88E+05 events/s
 
 *** (2-none) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -132,15 +132,15 @@ Executing ' ./build.none_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggtt_x1_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 47.69 [47.690709601032026] fbridge_mode=1
+ [XSECTION] Cross section = 47.69 [47.690709601032019] fbridge_mode=1
  [UNWEIGHT] Wrote 434 events (found 1125 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3545s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3161s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0384s for     8192 events => throughput is 2.13E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3684s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3289s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0395s for     8192 events => throughput is 2.07E+05 events/s
 
 *** (2-none) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (47.690708277600116) and cpp (47.690709601032026) differ by less than 2E-4 (2.7750309383733907e-08)
+OK! xsec from fortran (47.690708277600116) and cpp (47.690709601032019) differ by less than 2E-4 (2.77503091616893e-08)
 
 *** (2-none) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -165,29 +165,29 @@ Executing ' ./build.none_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggtt_x10
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 46.22 [46.223783635280988] fbridge_mode=1
+ [XSECTION] Cross section = 46.22 [46.223783635280974] fbridge_mode=1
  [UNWEIGHT] Wrote 1727 events (found 1732 events)
- [COUNTERS] PROGRAM TOTAL          :    1.7327s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.3085s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.4242s for    90112 events => throughput is 2.12E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    1.8620s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.4222s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.4398s for    90112 events => throughput is 2.05E+05 events/s
 
 *** (2-none) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (46.223782291775365) and cpp (46.223783635280988) differ by less than 2E-4 (2.9065246431869696e-08)
+OK! xsec from fortran (46.223782291775372) and cpp (46.223783635280974) differ by less than 2E-4 (2.9065245987780486e-08)
 
 *** (2-none) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+BRDHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.135635e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.094601e+05                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.125893e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.081772e+05                 )  sec^-1
 
 *** (2-sse4) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -210,9 +210,9 @@ Executing ' ./build.sse4_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggtt_x1_
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 47.69 [47.690709601032026] fbridge_mode=1
  [UNWEIGHT] Wrote 434 events (found 1125 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3220s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3000s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0221s for     8192 events => throughput is 3.71E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3338s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3117s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0221s for     8192 events => throughput is 3.70E+05 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
@@ -241,29 +241,29 @@ Executing ' ./build.sse4_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggtt_x10
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 46.22 [46.223783635280988] fbridge_mode=1
+ [XSECTION] Cross section = 46.22 [46.223783635280974] fbridge_mode=1
  [UNWEIGHT] Wrote 1727 events (found 1732 events)
- [COUNTERS] PROGRAM TOTAL          :    1.5189s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.2816s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.2374s for    90112 events => throughput is 3.80E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    1.6431s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.3996s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.2435s for    90112 events => throughput is 3.70E+05 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (46.223782291775365) and cpp (46.223783635280988) differ by less than 2E-4 (2.9065246431869696e-08)
+OK! xsec from fortran (46.223782291775372) and cpp (46.223783635280974) differ by less than 2E-4 (2.9065245987780486e-08)
 
 *** (2-sse4) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+BRDHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.761231e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.699859e+05                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.742706e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.728578e+05                 )  sec^-1
 
 *** (2-avx2) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -284,15 +284,15 @@ Executing ' ./build.avx2_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggtt_x1_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 47.69 [47.690709681138244] fbridge_mode=1
+ [XSECTION] Cross section = 47.69 [47.690709643441508] fbridge_mode=1
  [UNWEIGHT] Wrote 434 events (found 1125 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3057s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2927s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0130s for     8192 events => throughput is 6.28E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3207s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3066s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0141s for     8192 events => throughput is 5.81E+05 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (47.690708277600116) and cpp (47.690709681138244) differ by less than 2E-4 (2.9430012205011735e-08)
+OK! xsec from fortran (47.690708277600116) and cpp (47.690709643441508) differ by less than 2E-4 (2.863957027088304e-08)
 
 *** (2-avx2) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -317,29 +317,29 @@ Executing ' ./build.avx2_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggtt_x10
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 46.22 [46.223783652032040] fbridge_mode=1
+ [XSECTION] Cross section = 46.22 [46.223783660238851] fbridge_mode=1
  [UNWEIGHT] Wrote 1727 events (found 1732 events)
- [COUNTERS] PROGRAM TOTAL          :    1.4221s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.2763s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.1458s for    90112 events => throughput is 6.18E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    1.5524s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.3953s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.1572s for    90112 events => throughput is 5.73E+05 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (46.223782291775365) and cpp (46.223783652032040) differ by less than 2E-4 (2.9427636771828247e-08)
+OK! xsec from fortran (46.223782291775372) and cpp (46.223783660238851) differ by less than 2E-4 (2.9605181861569463e-08)
 
 *** (2-avx2) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+BRDHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 6.050162e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.934952e+05                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 6.113895e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.894708e+05                 )  sec^-1
 
 *** (2-512y) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -360,15 +360,15 @@ Executing ' ./build.512y_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggtt_x1_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 47.69 [47.690709681138244] fbridge_mode=1
+ [XSECTION] Cross section = 47.69 [47.690709643441508] fbridge_mode=1
  [UNWEIGHT] Wrote 434 events (found 1125 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3016s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2899s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0118s for     8192 events => throughput is 6.95E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3189s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3068s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0122s for     8192 events => throughput is 6.74E+05 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (47.690708277600116) and cpp (47.690709681138244) differ by less than 2E-4 (2.9430012205011735e-08)
+OK! xsec from fortran (47.690708277600116) and cpp (47.690709643441508) differ by less than 2E-4 (2.863957027088304e-08)
 
 *** (2-512y) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -393,29 +393,29 @@ Executing ' ./build.512y_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggtt_x10
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 46.22 [46.223783652032040] fbridge_mode=1
+ [XSECTION] Cross section = 46.22 [46.223783660238851] fbridge_mode=1
  [UNWEIGHT] Wrote 1727 events (found 1732 events)
- [COUNTERS] PROGRAM TOTAL          :    1.4299s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.2973s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.1326s for    90112 events => throughput is 6.80E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    1.5411s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.4039s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.1372s for    90112 events => throughput is 6.57E+05 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (46.223782291775365) and cpp (46.223783652032040) differ by less than 2E-4 (2.9427636771828247e-08)
+OK! xsec from fortran (46.223782291775372) and cpp (46.223783660238851) differ by less than 2E-4 (2.9605181861569463e-08)
 
 *** (2-512y) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+BRDHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 6.791940e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.847513e+05                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 7.001376e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.835511e+05                 )  sec^-1
 
 *** (2-512z) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -436,15 +436,15 @@ Executing ' ./build.512z_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggtt_x1_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 47.69 [47.690709681138244] fbridge_mode=1
+ [XSECTION] Cross section = 47.69 [47.690709643441508] fbridge_mode=1
  [UNWEIGHT] Wrote 434 events (found 1125 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3188s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2994s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0195s for     8192 events => throughput is 4.21E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3300s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3106s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0194s for     8192 events => throughput is 4.22E+05 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (47.690708277600116) and cpp (47.690709681138244) differ by less than 2E-4 (2.9430012205011735e-08)
+OK! xsec from fortran (47.690708277600116) and cpp (47.690709643441508) differ by less than 2E-4 (2.863957027088304e-08)
 
 *** (2-512z) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -469,29 +469,29 @@ Executing ' ./build.512z_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggtt_x10
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 46.22 [46.223783652032040] fbridge_mode=1
+ [XSECTION] Cross section = 46.22 [46.223783660238851] fbridge_mode=1
  [UNWEIGHT] Wrote 1727 events (found 1732 events)
- [COUNTERS] PROGRAM TOTAL          :    1.5212s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.3007s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.2205s for    90112 events => throughput is 4.09E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    1.6128s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.4020s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.2108s for    90112 events => throughput is 4.27E+05 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (46.223782291775365) and cpp (46.223783652032040) differ by less than 2E-4 (2.9427636771828247e-08)
+OK! xsec from fortran (46.223782291775372) and cpp (46.223783660238851) differ by less than 2E-4 (2.9605181861569463e-08)
 
 *** (2-512z) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+BRDHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.985174e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.284904e+05                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 4.068643e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.343219e+05                 )  sec^-1
 
 *** (3) EXECUTE MADEVENT_CUDA x1 (create events.lhe) ***
 --------------------
@@ -512,15 +512,15 @@ Executing ' ./build.none_m_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_ggtt_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 47.69 [47.690708266690699] fbridge_mode=1
+ [XSECTION] Cross section = 47.69 [47.690708266690706] fbridge_mode=1
  [UNWEIGHT] Wrote 434 events (found 1125 events)
- [COUNTERS] PROGRAM TOTAL          :    0.7071s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.7066s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0006s for     8192 events => throughput is 1.40E+07 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.7257s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.7251s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0006s for     8192 events => throughput is 1.43E+07 events/s
 
 *** (3) Compare MADEVENT_CUDA x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (47.690708277600116) and cpp (47.690708266690699) differ by less than 2E-4 (2.2875357164053867e-10)
+OK! xsec from fortran (47.690708277600116) and cpp (47.690708266690706) differ by less than 2E-4 (2.2875334959593374e-10)
 
 *** (3) Compare MADEVENT_CUDA x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -547,56 +547,56 @@ Executing ' ./build.none_m_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_ggtt_x1
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 46.22 [46.223782303744791] fbridge_mode=1
  [UNWEIGHT] Wrote 1727 events (found 1732 events)
- [COUNTERS] PROGRAM TOTAL          :    1.7018s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.6953s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0065s for    90112 events => throughput is 1.39E+07 events/s
+ [COUNTERS] PROGRAM TOTAL          :    1.8142s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.8072s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0070s for    90112 events => throughput is 1.29E+07 events/s
 
 *** (3) Compare MADEVENT_CUDA x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (46.223782291775365) and cpp (46.223782303744791) differ by less than 2E-4 (2.5894508759449764e-10)
+OK! xsec from fortran (46.223782291775372) and cpp (46.223782303744791) differ by less than 2E-4 (2.5894508759449764e-10)
 
 *** (3) Compare MADEVENT_CUDA x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cuda.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.014529e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.009523e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 5.676393e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.569379e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 16384 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.016903e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.989648e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 16384 32 1 ***
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.060987e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.069002e+08                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 4096 128 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.006671e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.991680e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 4096 128 1 ***
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.140193e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.142771e+08                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 65536 8 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.002803e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.993358e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 65536 8 1 ***
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.981810e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.008591e+07                 )  sec^-1
 
 TEST COMPLETED
diff --git a/epochX/cudacpp/tmad/logs_ggttg_mad/log_ggttg_mad_d_inl0_hrd0.txt b/epochX/cudacpp/tmad/logs_ggttg_mad/log_ggttg_mad_d_inl0_hrd0.txt
index 1e8a82a6de..ad1d0f839b 100644
--- a/epochX/cudacpp/tmad/logs_ggttg_mad/log_ggttg_mad_d_inl0_hrd0.txt
+++ b/epochX/cudacpp/tmad/logs_ggttg_mad/log_ggttg_mad_d_inl0_hrd0.txt
@@ -1,12 +1,12 @@
 Working directory (build): /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg
 CUDACPP_BUILDDIR='.'
 
-make USEBUILDDIR=1 AVX=none
 
+make USEBUILDDIR=1 AVX=none
 
 
-make USEBUILDDIR=1 AVX=sse4
 make USEBUILDDIR=1 AVX=avx2
+make USEBUILDDIR=1 AVX=sse4
 make USEBUILDDIR=1 AVX=512y
 
 make USEBUILDDIR=1 AVX=512z
@@ -18,9 +18,9 @@ make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp
 CUDACPP_BUILDDIR='build.avx2_d_inl0_hrd0'
 CUDACPP_BUILDDIR='build.sse4_d_inl0_hrd0'
 CUDACPP_BUILDDIR='build.512z_d_inl0_hrd0'
-CUDACPP_BUILDDIR='build.none_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg'
+CUDACPP_BUILDDIR='build.none_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg'
 CUDACPP_BUILDDIR='build.512y_d_inl0_hrd0'
@@ -33,7 +33,7 @@ make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/
 
 OMP_NUM_THREADS=
 
-DATE: 2023-11-24_15:37:12
+DATE: 2024-01-30_06:11:43
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 Working directory (run): /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg
@@ -59,9 +59,9 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggttg_x1_fortran > /tmp/ava
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.0972 [9.7196357922470764E-002] fbridge_mode=0
  [UNWEIGHT] Wrote 42 events (found 469 events)
- [COUNTERS] PROGRAM TOTAL          :    0.5593s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2322s
- [COUNTERS] Fortran MEs      ( 1 ) :    0.3271s for     8192 events => throughput is 2.50E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.5934s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2449s
+ [COUNTERS] Fortran MEs      ( 1 ) :    0.3485s for     8192 events => throughput is 2.35E+04 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x1 (create events.lhe) ***
 --------------------
@@ -84,9 +84,9 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggttg_x1_fortran > /tmp/ava
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.0972 [9.7196357922470764E-002] fbridge_mode=0
  [UNWEIGHT] Wrote 41 events (found 467 events)
- [COUNTERS] PROGRAM TOTAL          :    0.5556s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2267s
- [COUNTERS] Fortran MEs      ( 1 ) :    0.3290s for     8192 events => throughput is 2.49E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.5871s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2379s
+ [COUNTERS] Fortran MEs      ( 1 ) :    0.3492s for     8192 events => throughput is 2.35E+04 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x10 (create events.lhe) ***
 --------------------
@@ -107,11 +107,11 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggttg_x10_fortran > /tmp/av
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.08131 [8.1310872077655555E-002] fbridge_mode=0
+ [XSECTION] Cross section = 0.08131 [8.1310872077655569E-002] fbridge_mode=0
  [UNWEIGHT] Wrote 679 events (found 1787 events)
- [COUNTERS] PROGRAM TOTAL          :    5.0132s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.4373s
- [COUNTERS] Fortran MEs      ( 1 ) :    3.5759s for    90112 events => throughput is 2.52E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    5.3849s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.5565s
+ [COUNTERS] Fortran MEs      ( 1 ) :    3.8284s for    90112 events => throughput is 2.35E+04 events/s
 
 *** (2-none) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -132,15 +132,15 @@ Executing ' ./build.none_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttg_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.0972 [9.7196357922470791E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.0972 [9.7196357922470777E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 41 events (found 467 events)
- [COUNTERS] PROGRAM TOTAL          :    0.8869s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.5534s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.3335s for     8192 events => throughput is 2.46E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.9218s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.5788s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.3431s for     8192 events => throughput is 2.39E+04 events/s
 
 *** (2-none) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (9.7196357922470764E-002) and cpp (9.7196357922470791E-002) differ by less than 2E-14 (2.220446049250313e-16)
+OK! xsec from fortran (9.7196357922470764E-002) and cpp (9.7196357922470777E-002) differ by less than 2E-14 (2.220446049250313e-16)
 
 *** (2-none) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -167,27 +167,27 @@ Executing ' ./build.none_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttg_x1
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.08131 [8.1310872077655597E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 679 events (found 1787 events)
- [COUNTERS] PROGRAM TOTAL          :    5.5033s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.7676s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    3.7357s for    90112 events => throughput is 2.41E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    5.7045s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.9313s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    3.7731s for    90112 events => throughput is 2.39E+04 events/s
 
 *** (2-none) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (8.1310872077655555E-002) and cpp (8.1310872077655597E-002) differ by less than 2E-14 (4.440892098500626e-16)
+OK! xsec from fortran (8.1310872077655569E-002) and cpp (8.1310872077655597E-002) differ by less than 2E-14 (4.440892098500626e-16)
 
 *** (2-none) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.519174e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.459637e+04                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.514864e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.457759e+04                 )  sec^-1
 
 *** (2-sse4) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -210,9 +210,9 @@ Executing ' ./build.sse4_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttg_x1
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.0972 [9.7196357922470777E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 41 events (found 467 events)
- [COUNTERS] PROGRAM TOTAL          :    0.5689s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3958s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.1731s for     8192 events => throughput is 4.73E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.5930s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.4166s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.1764s for     8192 events => throughput is 4.64E+04 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
@@ -243,27 +243,27 @@ Executing ' ./build.sse4_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttg_x1
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.08131 [8.1310872077655555E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 679 events (found 1787 events)
- [COUNTERS] PROGRAM TOTAL          :    3.5161s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.5979s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    1.9182s for    90112 events => throughput is 4.70E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    3.7266s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.7761s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    1.9506s for    90112 events => throughput is 4.62E+04 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (8.1310872077655555E-002) and cpp (8.1310872077655555E-002) differ by less than 2E-14 (0.0)
+OK! xsec from fortran (8.1310872077655569E-002) and cpp (8.1310872077655555E-002) differ by less than 2E-14 (2.220446049250313e-16)
 
 *** (2-sse4) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 4.818399e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.735934e+04                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 4.781347e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.731299e+04                 )  sec^-1
 
 *** (2-avx2) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -286,9 +286,9 @@ Executing ' ./build.avx2_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttg_x1
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.0972 [9.7196357922470750E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 41 events (found 467 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3971s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3094s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0877s for     8192 events => throughput is 9.34E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.4266s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3331s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0935s for     8192 events => throughput is 8.76E+04 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
@@ -317,29 +317,29 @@ Executing ' ./build.avx2_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttg_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.08131 [8.1310872077655541E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.08131 [8.1310872077655555E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 679 events (found 1787 events)
- [COUNTERS] PROGRAM TOTAL          :    2.4605s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.5052s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.9553s for    90112 events => throughput is 9.43E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.6908s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.6878s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    1.0029s for    90112 events => throughput is 8.98E+04 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (8.1310872077655555E-002) and cpp (8.1310872077655541E-002) differ by less than 2E-14 (2.220446049250313e-16)
+OK! xsec from fortran (8.1310872077655569E-002) and cpp (8.1310872077655555E-002) differ by less than 2E-14 (2.220446049250313e-16)
 
 *** (2-avx2) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 9.701055e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.260876e+04                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 9.597059e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.205229e+04                 )  sec^-1
 
 *** (2-512y) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -362,9 +362,9 @@ Executing ' ./build.512y_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttg_x1
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.0972 [9.7196357922470750E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 41 events (found 467 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3796s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3015s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0781s for     8192 events => throughput is 1.05E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.4089s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3259s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0830s for     8192 events => throughput is 9.87E+04 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
@@ -393,29 +393,29 @@ Executing ' ./build.512y_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttg_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.08131 [8.1310872077655541E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.08131 [8.1310872077655555E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 679 events (found 1787 events)
- [COUNTERS] PROGRAM TOTAL          :    2.3975s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.5231s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.8744s for    90112 events => throughput is 1.03E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.5356s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.6675s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.8681s for    90112 events => throughput is 1.04E+05 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (8.1310872077655555E-002) and cpp (8.1310872077655541E-002) differ by less than 2E-14 (2.220446049250313e-16)
+OK! xsec from fortran (8.1310872077655569E-002) and cpp (8.1310872077655555E-002) differ by less than 2E-14 (2.220446049250313e-16)
 
 *** (2-512y) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.088451e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.066040e+05                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.080860e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.071802e+05                 )  sec^-1
 
 *** (2-512z) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -438,9 +438,9 @@ Executing ' ./build.512z_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttg_x1
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.0972 [9.7196357922470750E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 41 events (found 467 events)
- [COUNTERS] PROGRAM TOTAL          :    0.4429s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3340s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.1088s for     8192 events => throughput is 7.53E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.4633s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3521s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.1112s for     8192 events => throughput is 7.37E+04 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
@@ -469,29 +469,29 @@ Executing ' ./build.512z_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttg_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.08131 [8.1310872077655541E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.08131 [8.1310872077655555E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 679 events (found 1787 events)
- [COUNTERS] PROGRAM TOTAL          :    2.7475s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.5382s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    1.2094s for    90112 events => throughput is 7.45E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.9267s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.7103s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    1.2164s for    90112 events => throughput is 7.41E+04 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (8.1310872077655555E-002) and cpp (8.1310872077655541E-002) differ by less than 2E-14 (2.220446049250313e-16)
+OK! xsec from fortran (8.1310872077655569E-002) and cpp (8.1310872077655555E-002) differ by less than 2E-14 (2.220446049250313e-16)
 
 *** (2-512z) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 7.536802e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.521586e+04                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 7.505191e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.533151e+04                 )  sec^-1
 
 *** (3) EXECUTE MADEVENT_CUDA x1 (create events.lhe) ***
 --------------------
@@ -514,8 +514,8 @@ Executing ' ./build.none_d_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_ggttg_x
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.0972 [9.7196357922470764E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 41 events (found 467 events)
- [COUNTERS] PROGRAM TOTAL          :    0.6621s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.6566s
+ [COUNTERS] PROGRAM TOTAL          :    0.6854s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.6800s
  [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0055s for     8192 events => throughput is 1.50E+06 events/s
 
 *** (3) Compare MADEVENT_CUDA x1 xsec to MADEVENT_FORTRAN xsec ***
@@ -545,58 +545,58 @@ Executing ' ./build.none_d_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_ggttg_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.08131 [8.1310872077655597E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.08131 [8.1310872077655610E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 679 events (found 1787 events)
- [COUNTERS] PROGRAM TOTAL          :    1.8669s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.8440s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0228s for    90112 events => throughput is 3.94E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.0439s
+ [COUNTERS] Fortran Overhead ( 0 ) :    2.0206s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0234s for    90112 events => throughput is 3.85E+06 events/s
 
 *** (3) Compare MADEVENT_CUDA x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (8.1310872077655555E-002) and cpp (8.1310872077655597E-002) differ by less than 2E-14 (4.440892098500626e-16)
+OK! xsec from fortran (8.1310872077655569E-002) and cpp (8.1310872077655610E-002) differ by less than 2E-14 (4.440892098500626e-16)
 
 *** (3) Compare MADEVENT_CUDA x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cuda.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.613696e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.630499e+06                 )  sec^-1
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 8.322649e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.083902e+06                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 16384 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 9.665904e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.662154e+06                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 16384 32 1 ***
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.242154e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.243596e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 4096 128 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 9.685072e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.668083e+06                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 4096 128 1 ***
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.252410e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.255740e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 65536 8 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 9.680051e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.633959e+06                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 65536 8 1 ***
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.765757e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.773665e+06                 )  sec^-1
 
 TEST COMPLETED
diff --git a/epochX/cudacpp/tmad/logs_ggttg_mad/log_ggttg_mad_f_inl0_hrd0.txt b/epochX/cudacpp/tmad/logs_ggttg_mad/log_ggttg_mad_f_inl0_hrd0.txt
index c73e05bddd..c17be1788d 100644
--- a/epochX/cudacpp/tmad/logs_ggttg_mad/log_ggttg_mad_f_inl0_hrd0.txt
+++ b/epochX/cudacpp/tmad/logs_ggttg_mad/log_ggttg_mad_f_inl0_hrd0.txt
@@ -2,26 +2,26 @@ Working directory (build): /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/g
 CUDACPP_BUILDDIR='.'
 
 
-
-
 make USEBUILDDIR=1 AVX=none
+
 make USEBUILDDIR=1 AVX=sse4
-make USEBUILDDIR=1 AVX=avx2
 
+make USEBUILDDIR=1 AVX=avx2
 make USEBUILDDIR=1 AVX=512y
+
 make USEBUILDDIR=1 AVX=512z
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg'
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg'
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg'
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg'
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg'
-CUDACPP_BUILDDIR='build.512z_f_inl0_hrd0'
 CUDACPP_BUILDDIR='build.512y_f_inl0_hrd0'
-CUDACPP_BUILDDIR='build.avx2_f_inl0_hrd0'
 CUDACPP_BUILDDIR='build.none_f_inl0_hrd0'
-CUDACPP_BUILDDIR='build.sse4_f_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg'
+CUDACPP_BUILDDIR='build.avx2_f_inl0_hrd0'
+CUDACPP_BUILDDIR='build.512z_f_inl0_hrd0'
+CUDACPP_BUILDDIR='build.sse4_f_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg'
 make[1]: Nothing to be done for 'all'.
@@ -33,7 +33,7 @@ make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/
 
 OMP_NUM_THREADS=
 
-DATE: 2023-11-24_15:37:54
+DATE: 2024-01-30_06:12:28
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 Working directory (run): /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg
@@ -59,9 +59,9 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggttg_x1_fortran > /tmp/ava
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.0972 [9.7196357922470764E-002] fbridge_mode=0
  [UNWEIGHT] Wrote 42 events (found 469 events)
- [COUNTERS] PROGRAM TOTAL          :    0.5542s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2274s
- [COUNTERS] Fortran MEs      ( 1 ) :    0.3268s for     8192 events => throughput is 2.51E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.5874s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2394s
+ [COUNTERS] Fortran MEs      ( 1 ) :    0.3480s for     8192 events => throughput is 2.35E+04 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x1 (create events.lhe) ***
 --------------------
@@ -84,9 +84,9 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggttg_x1_fortran > /tmp/ava
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.0972 [9.7196357922470764E-002] fbridge_mode=0
  [UNWEIGHT] Wrote 41 events (found 467 events)
- [COUNTERS] PROGRAM TOTAL          :    0.5511s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2267s
- [COUNTERS] Fortran MEs      ( 1 ) :    0.3244s for     8192 events => throughput is 2.53E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.5918s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2437s
+ [COUNTERS] Fortran MEs      ( 1 ) :    0.3481s for     8192 events => throughput is 2.35E+04 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x10 (create events.lhe) ***
 --------------------
@@ -107,11 +107,11 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggttg_x10_fortran > /tmp/av
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.08131 [8.1310872077655555E-002] fbridge_mode=0
+ [XSECTION] Cross section = 0.08131 [8.1310872077655569E-002] fbridge_mode=0
  [UNWEIGHT] Wrote 679 events (found 1787 events)
- [COUNTERS] PROGRAM TOTAL          :    5.0510s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.4446s
- [COUNTERS] Fortran MEs      ( 1 ) :    3.6064s for    90112 events => throughput is 2.50E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    5.3901s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.5607s
+ [COUNTERS] Fortran MEs      ( 1 ) :    3.8294s for    90112 events => throughput is 2.35E+04 events/s
 
 *** (2-none) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -132,15 +132,15 @@ Executing ' ./build.none_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttg_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.0972 [9.7196349765248158E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.0972 [9.7196347758884971E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 41 events (found 467 events)
- [COUNTERS] PROGRAM TOTAL          :    0.8665s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.5402s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.3263s for     8192 events => throughput is 2.51E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.8754s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.5552s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.3203s for     8192 events => throughput is 2.56E+04 events/s
 
 *** (2-none) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (9.7196357922470764E-002) and cpp (9.7196349765248158E-002) differ by less than 4E-4 (8.392518791033865e-08)
+OK! xsec from fortran (9.7196357922470764E-002) and cpp (9.7196347758884971E-002) differ by less than 4E-4 (1.0456755794585604e-07)
 
 *** (2-none) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -165,29 +165,29 @@ Executing ' ./build.none_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttg_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.08131 [8.1310860767768514E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.08131 [8.1310858119443913E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 679 events (found 1787 events)
- [COUNTERS] PROGRAM TOTAL          :    5.2983s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.7336s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    3.5648s for    90112 events => throughput is 2.53E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    5.4534s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.9123s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    3.5411s for    90112 events => throughput is 2.54E+04 events/s
 
 *** (2-none) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (8.1310872077655555E-002) and cpp (8.1310860767768514E-002) differ by less than 4E-4 (1.3909440088610836e-07)
+OK! xsec from fortran (8.1310872077655569E-002) and cpp (8.1310858119443913E-002) differ by less than 4E-4 (1.7166476384833373e-07)
 
 *** (2-none) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.499402e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.651171e+04                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.491631e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.640512e+04                 )  sec^-1
 
 *** (2-sse4) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -208,15 +208,15 @@ Executing ' ./build.sse4_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttg_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.0972 [9.7196334183509370E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.0972 [9.7196323434217816E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 41 events (found 467 events)
- [COUNTERS] PROGRAM TOTAL          :    0.4217s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3247s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0970s for     8192 events => throughput is 8.45E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.4354s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3378s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0976s for     8192 events => throughput is 8.39E+04 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (9.7196357922470764E-002) and cpp (9.7196334183509370E-002) differ by less than 4E-4 (2.4423714939381114e-07)
+OK! xsec from fortran (9.7196357922470764E-002) and cpp (9.7196323434217816E-002) differ by less than 4E-4 (3.548307125900152e-07)
 
 *** (2-sse4) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -241,29 +241,29 @@ Executing ' ./build.sse4_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttg_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.08131 [8.1310847547651041E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.08131 [8.1310842598054087E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 679 events (found 1787 events)
- [COUNTERS] PROGRAM TOTAL          :    2.6020s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.5094s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    1.0926s for    90112 events => throughput is 8.25E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.7743s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.6927s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    1.0816s for    90112 events => throughput is 8.33E+04 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (8.1310872077655555E-002) and cpp (8.1310847547651041E-002) differ by less than 4E-4 (3.0168172948652483e-07)
+OK! xsec from fortran (8.1310872077655569E-002) and cpp (8.1310842598054087E-002) differ by less than 4E-4 (3.625542406293647e-07)
 
 *** (2-sse4) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 8.590308e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.607319e+04                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 8.556597e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.623071e+04                 )  sec^-1
 
 *** (2-avx2) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -284,15 +284,15 @@ Executing ' ./build.avx2_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttg_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.0972 [9.7196330801117323E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.0972 [9.7196325695161859E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 41 events (found 467 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3155s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2706s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0449s for     8192 events => throughput is 1.82E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3345s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2882s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0463s for     8192 events => throughput is 1.77E+05 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (9.7196357922470764E-002) and cpp (9.7196330801117323E-002) differ by less than 4E-4 (2.790367255034454e-07)
+OK! xsec from fortran (9.7196357922470764E-002) and cpp (9.7196325695161859E-002) differ by less than 4E-4 (3.3156909984288774e-07)
 
 *** (2-avx2) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -317,29 +317,29 @@ Executing ' ./build.avx2_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttg_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.08131 [8.1310847326088065E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.08131 [8.1310842393515825E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 679 events (found 1787 events)
- [COUNTERS] PROGRAM TOTAL          :    1.9588s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.4633s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.4955s for    90112 events => throughput is 1.82E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.1519s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.6391s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.5128s for    90112 events => throughput is 1.76E+05 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (8.1310872077655555E-002) and cpp (8.1310847326088065E-002) differ by less than 4E-4 (3.0440661691333304e-07)
+OK! xsec from fortran (8.1310872077655569E-002) and cpp (8.1310842393515825E-002) differ by less than 4E-4 (3.650697499857358e-07)
 
 *** (2-avx2) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.839831e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.801449e+05                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.863334e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.824776e+05                 )  sec^-1
 
 *** (2-512y) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -360,15 +360,15 @@ Executing ' ./build.512y_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttg_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.0972 [9.7196330801117323E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.0972 [9.7196325695161859E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 41 events (found 467 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3060s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2651s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0409s for     8192 events => throughput is 2.00E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3234s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2821s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0413s for     8192 events => throughput is 1.98E+05 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (9.7196357922470764E-002) and cpp (9.7196330801117323E-002) differ by less than 4E-4 (2.790367255034454e-07)
+OK! xsec from fortran (9.7196357922470764E-002) and cpp (9.7196325695161859E-002) differ by less than 4E-4 (3.3156909984288774e-07)
 
 *** (2-512y) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -393,29 +393,29 @@ Executing ' ./build.512y_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttg_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.08131 [8.1310847326088065E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.08131 [8.1310842393515825E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 679 events (found 1787 events)
- [COUNTERS] PROGRAM TOTAL          :    1.8963s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.4517s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.4446s for    90112 events => throughput is 2.03E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.0864s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.6356s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.4508s for    90112 events => throughput is 2.00E+05 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (8.1310872077655555E-002) and cpp (8.1310847326088065E-002) differ by less than 4E-4 (3.0440661691333304e-07)
+OK! xsec from fortran (8.1310872077655569E-002) and cpp (8.1310842393515825E-002) differ by less than 4E-4 (3.650697499857358e-07)
 
 *** (2-512y) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.073522e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.066287e+05                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.038945e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.073001e+05                 )  sec^-1
 
 *** (2-512z) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -436,15 +436,15 @@ Executing ' ./build.512z_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttg_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.0972 [9.7196344079460428E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.0972 [9.7196344080460087E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 41 events (found 467 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3327s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2795s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0532s for     8192 events => throughput is 1.54E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3518s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2972s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0546s for     8192 events => throughput is 1.50E+05 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (9.7196357922470764E-002) and cpp (9.7196344079460428E-002) differ by less than 4E-4 (1.424231383939656e-07)
+OK! xsec from fortran (9.7196357922470764E-002) and cpp (9.7196344080460087E-002) differ by less than 4E-4 (1.4241285339888776e-07)
 
 *** (2-512z) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -469,29 +469,29 @@ Executing ' ./build.512z_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttg_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.08131 [8.1310857804286998E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.08131 [8.1310857813116089E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 679 events (found 1787 events)
- [COUNTERS] PROGRAM TOTAL          :    2.0553s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.4670s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.5882s for    90112 events => throughput is 1.53E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.2586s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.6507s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.6080s for    90112 events => throughput is 1.48E+05 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (8.1310872077655555E-002) and cpp (8.1310857804286998E-002) differ by less than 4E-4 (1.7554071418679484e-07)
+OK! xsec from fortran (8.1310872077655569E-002) and cpp (8.1310857813116089E-002) differ by less than 4E-4 (1.754321300451167e-07)
 
 *** (2-512z) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.535514e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.497722e+05                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.548771e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.492408e+05                 )  sec^-1
 
 *** (3) EXECUTE MADEVENT_CUDA x1 (create events.lhe) ***
 --------------------
@@ -512,15 +512,15 @@ Executing ' ./build.none_f_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_ggttg_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.0972 [9.7196349366365994E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.0972 [9.7196349366366022E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 41 events (found 467 events)
- [COUNTERS] PROGRAM TOTAL          :    0.6530s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.6521s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0008s for     8192 events => throughput is 9.69E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.6751s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.6742s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0008s for     8192 events => throughput is 9.66E+06 events/s
 
 *** (3) Compare MADEVENT_CUDA x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (9.7196357922470764E-002) and cpp (9.7196349366365994E-002) differ by less than 4E-4 (8.802906770188912e-08)
+OK! xsec from fortran (9.7196357922470764E-002) and cpp (9.7196349366366022E-002) differ by less than 4E-4 (8.802906736882221e-08)
 
 *** (3) Compare MADEVENT_CUDA x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -545,58 +545,58 @@ Executing ' ./build.none_f_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_ggttg_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.08131 [8.1310864949473968E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.08131 [8.1310864949473954E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 679 events (found 1787 events)
- [COUNTERS] PROGRAM TOTAL          :    1.8988s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.8888s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0101s for    90112 events => throughput is 8.95E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.0322s
+ [COUNTERS] Fortran Overhead ( 0 ) :    2.0221s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0102s for    90112 events => throughput is 8.88E+06 events/s
 
 *** (3) Compare MADEVENT_CUDA x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (8.1310872077655555E-002) and cpp (8.1310864949473968E-002) differ by less than 4E-4 (8.766578696306482e-08)
+OK! xsec from fortran (8.1310872077655569E-002) and cpp (8.1310864949473954E-002) differ by less than 4E-4 (8.766578729613173e-08)
 
 *** (3) Compare MADEVENT_CUDA x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cuda.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.308714e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.288695e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.861624e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.864373e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 16384 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.626812e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.630957e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 16384 32 1 ***
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 4.413554e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.365812e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 4096 128 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.631638e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.633946e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 4096 128 1 ***
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 4.496116e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.471906e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 65536 8 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.503568e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.509519e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 65536 8 1 ***
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.624454e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.624050e+07                 )  sec^-1
 
 TEST COMPLETED
diff --git a/epochX/cudacpp/tmad/logs_ggttg_mad/log_ggttg_mad_m_inl0_hrd0.txt b/epochX/cudacpp/tmad/logs_ggttg_mad/log_ggttg_mad_m_inl0_hrd0.txt
index b4daea9308..daa5ca9a3d 100644
--- a/epochX/cudacpp/tmad/logs_ggttg_mad/log_ggttg_mad_m_inl0_hrd0.txt
+++ b/epochX/cudacpp/tmad/logs_ggttg_mad/log_ggttg_mad_m_inl0_hrd0.txt
@@ -3,8 +3,8 @@ CUDACPP_BUILDDIR='.'
 
 
 
-
 make USEBUILDDIR=1 AVX=none
+
 make USEBUILDDIR=1 AVX=sse4
 make USEBUILDDIR=1 AVX=avx2
 make USEBUILDDIR=1 AVX=512y
@@ -16,12 +16,12 @@ make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg'
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg'
 CUDACPP_BUILDDIR='build.512z_m_inl0_hrd0'
+CUDACPP_BUILDDIR='build.512y_m_inl0_hrd0'
 CUDACPP_BUILDDIR='build.avx2_m_inl0_hrd0'
-CUDACPP_BUILDDIR='build.sse4_m_inl0_hrd0'
 CUDACPP_BUILDDIR='build.none_m_inl0_hrd0'
+CUDACPP_BUILDDIR='build.sse4_m_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg'
-CUDACPP_BUILDDIR='build.512y_m_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg'
 make[1]: Nothing to be done for 'all'.
@@ -33,7 +33,7 @@ make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/
 
 OMP_NUM_THREADS=
 
-DATE: 2023-11-24_15:38:32
+DATE: 2024-01-30_06:13:08
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 Working directory (run): /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg
@@ -59,9 +59,9 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggttg_x1_fortran > /tmp/ava
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.0972 [9.7196357922470764E-002] fbridge_mode=0
  [UNWEIGHT] Wrote 42 events (found 469 events)
- [COUNTERS] PROGRAM TOTAL          :    0.5629s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2337s
- [COUNTERS] Fortran MEs      ( 1 ) :    0.3292s for     8192 events => throughput is 2.49E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.5921s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2413s
+ [COUNTERS] Fortran MEs      ( 1 ) :    0.3507s for     8192 events => throughput is 2.34E+04 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x1 (create events.lhe) ***
 --------------------
@@ -84,9 +84,9 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggttg_x1_fortran > /tmp/ava
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.0972 [9.7196357922470764E-002] fbridge_mode=0
  [UNWEIGHT] Wrote 41 events (found 467 events)
- [COUNTERS] PROGRAM TOTAL          :    0.5518s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2262s
- [COUNTERS] Fortran MEs      ( 1 ) :    0.3256s for     8192 events => throughput is 2.52E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.5875s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2393s
+ [COUNTERS] Fortran MEs      ( 1 ) :    0.3482s for     8192 events => throughput is 2.35E+04 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x10 (create events.lhe) ***
 --------------------
@@ -107,11 +107,11 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggttg_x10_fortran > /tmp/av
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.08131 [8.1310872077655555E-002] fbridge_mode=0
+ [XSECTION] Cross section = 0.08131 [8.1310872077655569E-002] fbridge_mode=0
  [UNWEIGHT] Wrote 679 events (found 1787 events)
- [COUNTERS] PROGRAM TOTAL          :    5.0097s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.4326s
- [COUNTERS] Fortran MEs      ( 1 ) :    3.5771s for    90112 events => throughput is 2.52E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    5.3863s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.5579s
+ [COUNTERS] Fortran MEs      ( 1 ) :    3.8284s for    90112 events => throughput is 2.35E+04 events/s
 
 *** (2-none) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -132,15 +132,15 @@ Executing ' ./build.none_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttg_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.0972 [9.7196358763382007E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.0972 [9.7196358763382021E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 41 events (found 467 events)
- [COUNTERS] PROGRAM TOTAL          :    0.8974s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.5586s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.3388s for     8192 events => throughput is 2.42E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.9324s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.5834s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.3490s for     8192 events => throughput is 2.35E+04 events/s
 
 *** (2-none) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (9.7196357922470764E-002) and cpp (9.7196358763382007E-002) differ by less than 2E-4 (8.651674487936134e-09)
+OK! xsec from fortran (9.7196357922470764E-002) and cpp (9.7196358763382021E-002) differ by less than 2E-4 (8.651674487936134e-09)
 
 *** (2-none) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -167,27 +167,27 @@ Executing ' ./build.none_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttg_x1
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.08131 [8.1310872835011053E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 679 events (found 1787 events)
- [COUNTERS] PROGRAM TOTAL          :    5.5261s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.7672s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    3.7589s for    90112 events => throughput is 2.40E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    5.7762s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.9305s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    3.8457s for    90112 events => throughput is 2.34E+04 events/s
 
 *** (2-none) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (8.1310872077655555E-002) and cpp (8.1310872835011053E-002) differ by less than 2E-4 (9.31432020401246e-09)
+OK! xsec from fortran (8.1310872077655569E-002) and cpp (8.1310872835011053E-002) differ by less than 2E-4 (9.314319981967856e-09)
 
 *** (2-none) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+BRDHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.465437e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.405521e+04                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.473065e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.412039e+04                 )  sec^-1
 
 *** (2-sse4) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -208,15 +208,15 @@ Executing ' ./build.sse4_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttg_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.0972 [9.7196358804670396E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.0972 [9.7196358804670424E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 41 events (found 467 events)
- [COUNTERS] PROGRAM TOTAL          :    0.5634s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3926s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.1708s for     8192 events => throughput is 4.80E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.5895s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.4134s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.1762s for     8192 events => throughput is 4.65E+04 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (9.7196357922470764E-002) and cpp (9.7196358804670396E-002) differ by less than 2E-4 (9.076468021618211e-09)
+OK! xsec from fortran (9.7196357922470764E-002) and cpp (9.7196358804670424E-002) differ by less than 2E-4 (9.076468243662816e-09)
 
 *** (2-sse4) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -243,27 +243,27 @@ Executing ' ./build.sse4_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttg_x1
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.08131 [8.1310872836789727E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 679 events (found 1787 events)
- [COUNTERS] PROGRAM TOTAL          :    3.4683s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.5927s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    1.8757s for    90112 events => throughput is 4.80E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    3.7571s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.7661s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    1.9910s for    90112 events => throughput is 4.53E+04 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (8.1310872077655555E-002) and cpp (8.1310872836789727E-002) differ by less than 2E-4 (9.336195150311255e-09)
+OK! xsec from fortran (8.1310872077655569E-002) and cpp (8.1310872836789727E-002) differ by less than 2E-4 (9.33619492826665e-09)
 
 *** (2-sse4) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+BRDHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 4.905170e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.812264e+04                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 4.906590e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.790866e+04                 )  sec^-1
 
 *** (2-avx2) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -284,15 +284,15 @@ Executing ' ./build.avx2_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttg_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.0972 [9.7196358586501358E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.0972 [9.7196358586501386E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 41 events (found 467 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3954s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3088s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0866s for     8192 events => throughput is 9.45E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.4175s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3288s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0887s for     8192 events => throughput is 9.23E+04 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (9.7196357922470764E-002) and cpp (9.7196358586501358E-002) differ by less than 2E-4 (6.831846421917476e-09)
+OK! xsec from fortran (9.7196357922470764E-002) and cpp (9.7196358586501386E-002) differ by less than 2E-4 (6.831846643962081e-09)
 
 *** (2-avx2) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -317,29 +317,29 @@ Executing ' ./build.avx2_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttg_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.08131 [8.1310872708918333E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.08131 [8.1310872708918305E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 679 events (found 1787 events)
- [COUNTERS] PROGRAM TOTAL          :    2.4529s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.4970s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.9559s for    90112 events => throughput is 9.43E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.6655s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.6760s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.9895s for    90112 events => throughput is 9.11E+04 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (8.1310872077655555E-002) and cpp (8.1310872708918333E-002) differ by less than 2E-4 (7.763571563401683e-09)
+OK! xsec from fortran (8.1310872077655569E-002) and cpp (8.1310872708918305E-002) differ by less than 2E-4 (7.763571119312473e-09)
 
 *** (2-avx2) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+BRDHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 9.722917e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.369009e+04                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 9.639767e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.305282e+04                 )  sec^-1
 
 *** (2-512y) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -360,15 +360,15 @@ Executing ' ./build.512y_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttg_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.0972 [9.7196358586501358E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.0972 [9.7196358586501386E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 41 events (found 467 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3758s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3002s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0756s for     8192 events => throughput is 1.08E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3954s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3173s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0781s for     8192 events => throughput is 1.05E+05 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (9.7196357922470764E-002) and cpp (9.7196358586501358E-002) differ by less than 2E-4 (6.831846421917476e-09)
+OK! xsec from fortran (9.7196357922470764E-002) and cpp (9.7196358586501386E-002) differ by less than 2E-4 (6.831846643962081e-09)
 
 *** (2-512y) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -393,29 +393,29 @@ Executing ' ./build.512y_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttg_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.08131 [8.1310872708918333E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.08131 [8.1310872708918305E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 679 events (found 1787 events)
- [COUNTERS] PROGRAM TOTAL          :    2.3291s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.4914s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.8377s for    90112 events => throughput is 1.08E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.5272s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.6689s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.8582s for    90112 events => throughput is 1.05E+05 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (8.1310872077655555E-002) and cpp (8.1310872708918333E-002) differ by less than 2E-4 (7.763571563401683e-09)
+OK! xsec from fortran (8.1310872077655569E-002) and cpp (8.1310872708918305E-002) differ by less than 2E-4 (7.763571119312473e-09)
 
 *** (2-512y) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+BRDHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.100825e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.086568e+05                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.120448e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.099441e+05                 )  sec^-1
 
 *** (2-512z) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -436,15 +436,15 @@ Executing ' ./build.512z_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttg_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.0972 [9.7196358757578441E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.0972 [9.7196358586501386E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 41 events (found 467 events)
- [COUNTERS] PROGRAM TOTAL          :    0.4736s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3530s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.1206s for     8192 events => throughput is 6.79E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.4677s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3542s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.1134s for     8192 events => throughput is 7.22E+04 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (9.7196357922470764E-002) and cpp (9.7196358757578441E-002) differ by less than 2E-4 (8.591964695270349e-09)
+OK! xsec from fortran (9.7196357922470764E-002) and cpp (9.7196358586501386E-002) differ by less than 2E-4 (6.831846643962081e-09)
 
 *** (2-512z) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -469,29 +469,29 @@ Executing ' ./build.512z_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttg_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.08131 [8.1310872803699391E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.08131 [8.1310872708918305E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 679 events (found 1787 events)
- [COUNTERS] PROGRAM TOTAL          :    2.8047s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.5601s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    1.2446s for    90112 events => throughput is 7.24E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.9514s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.7052s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    1.2463s for    90112 events => throughput is 7.23E+04 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (8.1310872077655555E-002) and cpp (8.1310872803699391E-002) differ by less than 2E-4 (8.929234462939917e-09)
+OK! xsec from fortran (8.1310872077655569E-002) and cpp (8.1310872708918305E-002) differ by less than 2E-4 (7.763571119312473e-09)
 
 *** (2-512z) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+BRDHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 7.211134e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.364043e+04                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 7.316317e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.347069e+04                 )  sec^-1
 
 *** (3) EXECUTE MADEVENT_CUDA x1 (create events.lhe) ***
 --------------------
@@ -512,15 +512,15 @@ Executing ' ./build.none_m_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_ggttg_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.0972 [9.7196358102981245E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.0972 [9.7196358102981231E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 41 events (found 467 events)
- [COUNTERS] PROGRAM TOTAL          :    0.6693s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.6639s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0054s for     8192 events => throughput is 1.53E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.6813s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.6759s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0054s for     8192 events => throughput is 1.52E+06 events/s
 
 *** (3) Compare MADEVENT_CUDA x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (9.7196357922470764E-002) and cpp (9.7196358102981245E-002) differ by less than 2E-4 (1.8571733040317895e-09)
+OK! xsec from fortran (9.7196357922470764E-002) and cpp (9.7196358102981231E-002) differ by less than 2E-4 (1.8571730819871846e-09)
 
 *** (3) Compare MADEVENT_CUDA x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -545,58 +545,58 @@ Executing ' ./build.none_m_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_ggttg_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.08131 [8.1310872068634174E-002] fbridge_mode=1
+ [XSECTION] Cross section = 0.08131 [8.1310872068634160E-002] fbridge_mode=1
  [UNWEIGHT] Wrote 679 events (found 1787 events)
- [COUNTERS] PROGRAM TOTAL          :    1.8539s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.8310s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0229s for    90112 events => throughput is 3.94E+06 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.0449s
+ [COUNTERS] Fortran Overhead ( 0 ) :    2.0215s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0233s for    90112 events => throughput is 3.86E+06 events/s
 
 *** (3) Compare MADEVENT_CUDA x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (8.1310872077655555E-002) and cpp (8.1310872068634174E-002) differ by less than 2E-4 (1.1094924978749532e-10)
+OK! xsec from fortran (8.1310872077655569E-002) and cpp (8.1310872068634160E-002) differ by less than 2E-4 (1.109495828544027e-10)
 
 *** (3) Compare MADEVENT_CUDA x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cuda.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.622730e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.624283e+06                 )  sec^-1
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 8.249441e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.218126e+06                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 16384 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 9.646288e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.599538e+06                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 16384 32 1 ***
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.232584e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.232652e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 4096 128 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 9.632648e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.616286e+06                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 4096 128 1 ***
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.244267e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.243703e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 65536 8 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 9.616374e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.609732e+06                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 65536 8 1 ***
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.720039e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.728637e+06                 )  sec^-1
 
 TEST COMPLETED
diff --git a/epochX/cudacpp/tmad/logs_ggttgg_mad/log_ggttgg_mad_d_inl0_hrd0.txt b/epochX/cudacpp/tmad/logs_ggttgg_mad/log_ggttgg_mad_d_inl0_hrd0.txt
index 2c14264714..930476d789 100644
--- a/epochX/cudacpp/tmad/logs_ggttgg_mad/log_ggttgg_mad_d_inl0_hrd0.txt
+++ b/epochX/cudacpp/tmad/logs_ggttgg_mad/log_ggttgg_mad_d_inl0_hrd0.txt
@@ -2,10 +2,10 @@ Working directory (build): /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/g
 CUDACPP_BUILDDIR='.'
 
 
-make USEBUILDDIR=1 AVX=none
-make USEBUILDDIR=1 AVX=sse4
 
 
+make USEBUILDDIR=1 AVX=none
+make USEBUILDDIR=1 AVX=sse4
 make USEBUILDDIR=1 AVX=avx2
 make USEBUILDDIR=1 AVX=512y
 
@@ -15,17 +15,17 @@ make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
-CUDACPP_BUILDDIR='build.512z_d_inl0_hrd0'
-CUDACPP_BUILDDIR='build.512y_d_inl0_hrd0'
 CUDACPP_BUILDDIR='build.none_d_inl0_hrd0'
+CUDACPP_BUILDDIR='build.sse4_d_inl0_hrd0'
+CUDACPP_BUILDDIR='build.avx2_d_inl0_hrd0'
+CUDACPP_BUILDDIR='build.512y_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
+CUDACPP_BUILDDIR='build.512z_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
-CUDACPP_BUILDDIR='build.sse4_d_inl0_hrd0'
-CUDACPP_BUILDDIR='build.avx2_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
 make[1]: Nothing to be done for 'all'.
@@ -33,7 +33,7 @@ make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/
 
 OMP_NUM_THREADS=
 
-DATE: 2023-11-24_15:39:14
+DATE: 2024-01-30_06:13:52
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 Working directory (run): /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg
@@ -57,11 +57,11 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggttgg_x1_fortran > /tmp/av
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 2
- [XSECTION] Cross section = 0.0003628 [3.6277277311352988E-004] fbridge_mode=0
+ [XSECTION] Cross section = 0.0003628 [3.6277277311352982E-004] fbridge_mode=0
  [UNWEIGHT] Wrote 48 events (found 439 events)
- [COUNTERS] PROGRAM TOTAL          :    4.5067s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2862s
- [COUNTERS] Fortran MEs      ( 1 ) :    4.2205s for     8192 events => throughput is 1.94E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :    4.7127s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3035s
+ [COUNTERS] Fortran MEs      ( 1 ) :    4.4093s for     8192 events => throughput is 1.86E+03 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x1 (create events.lhe) ***
 --------------------
@@ -82,11 +82,11 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggttgg_x1_fortran > /tmp/av
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 2
- [XSECTION] Cross section = 0.0003628 [3.6277277311352988E-004] fbridge_mode=0
+ [XSECTION] Cross section = 0.0003628 [3.6277277311352982E-004] fbridge_mode=0
  [UNWEIGHT] Wrote 59 events (found 420 events)
- [COUNTERS] PROGRAM TOTAL          :    4.5163s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2798s
- [COUNTERS] Fortran MEs      ( 1 ) :    4.2365s for     8192 events => throughput is 1.93E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :    4.7196s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2934s
+ [COUNTERS] Fortran MEs      ( 1 ) :    4.4261s for     8192 events => throughput is 1.85E+03 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x10 (create events.lhe) ***
 --------------------
@@ -107,11 +107,11 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggttgg_x10_fortran > /tmp/a
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 2
- [XSECTION] Cross section = 0.000158 [1.5803725748421164E-004] fbridge_mode=0
+ [XSECTION] Cross section = 0.000158 [1.5803725748421158E-004] fbridge_mode=0
  [UNWEIGHT] Wrote 207 events (found 1235 events)
- [COUNTERS] PROGRAM TOTAL          :   48.4845s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.9328s
- [COUNTERS] Fortran MEs      ( 1 ) :   46.5516s for    90112 events => throughput is 1.94E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :   50.8146s
+ [COUNTERS] Fortran Overhead ( 0 ) :    2.0980s
+ [COUNTERS] Fortran MEs      ( 1 ) :   48.7166s for    90112 events => throughput is 1.85E+03 events/s
 
 *** (2-none) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -132,15 +132,15 @@ Executing ' ./build.none_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttgg_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 2
- [XSECTION] Cross section = 0.0003628 [3.6277277311352998E-004] fbridge_mode=1
+ [XSECTION] Cross section = 0.0003628 [3.6277277311352993E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 59 events (found 420 events)
- [COUNTERS] PROGRAM TOTAL          :    8.8031s
- [COUNTERS] Fortran Overhead ( 0 ) :    4.5022s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    4.3008s for     8192 events => throughput is 1.90E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :    9.5114s
+ [COUNTERS] Fortran Overhead ( 0 ) :    4.8392s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    4.6722s for     8192 events => throughput is 1.75E+03 events/s
 
 *** (2-none) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (3.6277277311352988E-004) and cpp (3.6277277311352998E-004) differ by less than 2E-14 (2.220446049250313e-16)
+OK! xsec from fortran (3.6277277311352982E-004) and cpp (3.6277277311352993E-004) differ by less than 2E-14 (2.220446049250313e-16)
 
 *** (2-none) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -165,29 +165,29 @@ Executing ' ./build.none_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttgg_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 2
- [XSECTION] Cross section = 0.000158 [1.5803725748421161E-004] fbridge_mode=1
+ [XSECTION] Cross section = 0.000158 [1.5803725748421150E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 207 events (found 1235 events)
- [COUNTERS] PROGRAM TOTAL          :   53.6260s
- [COUNTERS] Fortran Overhead ( 0 ) :    6.1417s
- [COUNTERS] CudaCpp MEs      ( 2 ) :   47.4843s for    90112 events => throughput is 1.90E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :   57.8440s
+ [COUNTERS] Fortran Overhead ( 0 ) :    6.6165s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :   51.2275s for    90112 events => throughput is 1.76E+03 events/s
 
 *** (2-none) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.5803725748421164E-004) and cpp (1.5803725748421161E-004) differ by less than 2E-14 (2.220446049250313e-16)
+OK! xsec from fortran (1.5803725748421158E-004) and cpp (1.5803725748421150E-004) differ by less than 2E-14 (5.551115123125783e-16)
 
 *** (2-none) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.955415e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.804400e+03                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.957400e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.805886e+03                 )  sec^-1
 
 *** (2-sse4) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -208,15 +208,15 @@ Executing ' ./build.sse4_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttgg_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 2
- [XSECTION] Cross section = 0.0003628 [3.6277277311352993E-004] fbridge_mode=1
+ [XSECTION] Cross section = 0.0003628 [3.6277277311352998E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 59 events (found 420 events)
- [COUNTERS] PROGRAM TOTAL          :    4.8602s
- [COUNTERS] Fortran Overhead ( 0 ) :    2.5384s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    2.3219s for     8192 events => throughput is 3.53E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :    5.0600s
+ [COUNTERS] Fortran Overhead ( 0 ) :    2.6542s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    2.4058s for     8192 events => throughput is 3.41E+03 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (3.6277277311352988E-004) and cpp (3.6277277311352993E-004) differ by less than 2E-14 (2.220446049250313e-16)
+OK! xsec from fortran (3.6277277311352982E-004) and cpp (3.6277277311352998E-004) differ by less than 2E-14 (4.440892098500626e-16)
 
 *** (2-sse4) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -241,29 +241,29 @@ Executing ' ./build.sse4_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttgg_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 2
- [XSECTION] Cross section = 0.000158 [1.5803725748421158E-004] fbridge_mode=1
+ [XSECTION] Cross section = 0.000158 [1.5803725748421156E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 207 events (found 1235 events)
- [COUNTERS] PROGRAM TOTAL          :   29.4526s
- [COUNTERS] Fortran Overhead ( 0 ) :    4.1609s
- [COUNTERS] CudaCpp MEs      ( 2 ) :   25.2916s for    90112 events => throughput is 3.56E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :   30.6354s
+ [COUNTERS] Fortran Overhead ( 0 ) :    4.3949s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :   26.2405s for    90112 events => throughput is 3.43E+03 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.5803725748421164E-004) and cpp (1.5803725748421158E-004) differ by less than 2E-14 (3.3306690738754696e-16)
+OK! xsec from fortran (1.5803725748421158E-004) and cpp (1.5803725748421156E-004) differ by less than 2E-14 (2.220446049250313e-16)
 
 *** (2-sse4) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.699774e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.603718e+03                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.692908e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.615187e+03                 )  sec^-1
 
 *** (2-avx2) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -286,13 +286,13 @@ Executing ' ./build.avx2_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttgg_x
  [XSECTION] ChannelId = 2
  [XSECTION] Cross section = 0.0003628 [3.6277277311353009E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 59 events (found 420 events)
- [COUNTERS] PROGRAM TOTAL          :    2.2393s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.2482s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.9911s for     8192 events => throughput is 8.27E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.3469s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.3045s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    1.0424s for     8192 events => throughput is 7.86E+03 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (3.6277277311352988E-004) and cpp (3.6277277311353009E-004) differ by less than 2E-14 (6.661338147750939e-16)
+OK! xsec from fortran (3.6277277311352982E-004) and cpp (3.6277277311353009E-004) differ by less than 2E-14 (6.661338147750939e-16)
 
 *** (2-avx2) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -319,27 +319,27 @@ Executing ' ./build.avx2_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttgg_x
  [XSECTION] ChannelId = 2
  [XSECTION] Cross section = 0.000158 [1.5803725748421158E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 207 events (found 1235 events)
- [COUNTERS] PROGRAM TOTAL          :   13.8246s
- [COUNTERS] Fortran Overhead ( 0 ) :    2.9004s
- [COUNTERS] CudaCpp MEs      ( 2 ) :   10.9241s for    90112 events => throughput is 8.25E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :   14.6028s
+ [COUNTERS] Fortran Overhead ( 0 ) :    3.0873s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :   11.5154s for    90112 events => throughput is 7.83E+03 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.5803725748421164E-004) and cpp (1.5803725748421158E-004) differ by less than 2E-14 (3.3306690738754696e-16)
+OK! xsec from fortran (1.5803725748421158E-004) and cpp (1.5803725748421158E-004) differ by less than 2E-14 (0.0)
 
 *** (2-avx2) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 8.470220e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.099934e+03                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 8.480723e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.056733e+03                 )  sec^-1
 
 *** (2-512y) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -362,13 +362,13 @@ Executing ' ./build.512y_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttgg_x
  [XSECTION] ChannelId = 2
  [XSECTION] Cross section = 0.0003628 [3.6277277311353009E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 59 events (found 420 events)
- [COUNTERS] PROGRAM TOTAL          :    2.0124s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.1378s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.8747s for     8192 events => throughput is 9.37E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.1181s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.1916s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.9265s for     8192 events => throughput is 8.84E+03 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (3.6277277311352988E-004) and cpp (3.6277277311353009E-004) differ by less than 2E-14 (6.661338147750939e-16)
+OK! xsec from fortran (3.6277277311352982E-004) and cpp (3.6277277311353009E-004) differ by less than 2E-14 (6.661338147750939e-16)
 
 *** (2-512y) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -395,27 +395,27 @@ Executing ' ./build.512y_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttgg_x
  [XSECTION] ChannelId = 2
  [XSECTION] Cross section = 0.000158 [1.5803725748421158E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 207 events (found 1235 events)
- [COUNTERS] PROGRAM TOTAL          :   12.5302s
- [COUNTERS] Fortran Overhead ( 0 ) :    2.7907s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    9.7395s for    90112 events => throughput is 9.25E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :   13.1007s
+ [COUNTERS] Fortran Overhead ( 0 ) :    2.9669s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :   10.1338s for    90112 events => throughput is 8.89E+03 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.5803725748421164E-004) and cpp (1.5803725748421158E-004) differ by less than 2E-14 (3.3306690738754696e-16)
+OK! xsec from fortran (1.5803725748421158E-004) and cpp (1.5803725748421158E-004) differ by less than 2E-14 (0.0)
 
 *** (2-512y) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 9.611254e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.168199e+03                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 9.611296e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.198606e+03                 )  sec^-1
 
 *** (2-512z) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -438,13 +438,13 @@ Executing ' ./build.512z_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttgg_x
  [XSECTION] ChannelId = 2
  [XSECTION] Cross section = 0.0003628 [3.6277277311353009E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 59 events (found 420 events)
- [COUNTERS] PROGRAM TOTAL          :    2.4817s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.3731s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    1.1087s for     8192 events => throughput is 7.39E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.6312s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.4569s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    1.1743s for     8192 events => throughput is 6.98E+03 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (3.6277277311352988E-004) and cpp (3.6277277311353009E-004) differ by less than 2E-14 (6.661338147750939e-16)
+OK! xsec from fortran (3.6277277311352982E-004) and cpp (3.6277277311353009E-004) differ by less than 2E-14 (6.661338147750939e-16)
 
 *** (2-512z) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -471,27 +471,27 @@ Executing ' ./build.512z_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttgg_x
  [XSECTION] ChannelId = 2
  [XSECTION] Cross section = 0.000158 [1.5803725748421158E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 207 events (found 1235 events)
- [COUNTERS] PROGRAM TOTAL          :   15.5304s
- [COUNTERS] Fortran Overhead ( 0 ) :    3.0362s
- [COUNTERS] CudaCpp MEs      ( 2 ) :   12.4941s for    90112 events => throughput is 7.21E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :   16.1783s
+ [COUNTERS] Fortran Overhead ( 0 ) :    3.2363s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :   12.9420s for    90112 events => throughput is 6.96E+03 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.5803725748421164E-004) and cpp (1.5803725748421158E-004) differ by less than 2E-14 (3.3306690738754696e-16)
+OK! xsec from fortran (1.5803725748421158E-004) and cpp (1.5803725748421158E-004) differ by less than 2E-14 (0.0)
 
 *** (2-512z) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 7.310034e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.039911e+03                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 7.219164e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.063323e+03                 )  sec^-1
 
 *** (3) EXECUTE MADEVENT_CUDA x1 (create events.lhe) ***
 --------------------
@@ -514,13 +514,13 @@ Executing ' ./build.none_d_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_ggttgg_
  [XSECTION] ChannelId = 2
  [XSECTION] Cross section = 0.0003628 [3.6277277311352998E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 59 events (found 420 events)
- [COUNTERS] PROGRAM TOTAL          :    0.8199s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.7868s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0331s for     8192 events => throughput is 2.48E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.8345s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.8014s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0331s for     8192 events => throughput is 2.47E+05 events/s
 
 *** (3) Compare MADEVENT_CUDA x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (3.6277277311352988E-004) and cpp (3.6277277311352998E-004) differ by less than 2E-14 (2.220446049250313e-16)
+OK! xsec from fortran (3.6277277311352982E-004) and cpp (3.6277277311352998E-004) differ by less than 2E-14 (4.440892098500626e-16)
 
 *** (3) Compare MADEVENT_CUDA x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -545,58 +545,58 @@ Executing ' ./build.none_d_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_ggttgg_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 2
- [XSECTION] Cross section = 0.000158 [1.5803725748421161E-004] fbridge_mode=1
+ [XSECTION] Cross section = 0.000158 [1.5803725748421166E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 207 events (found 1235 events)
- [COUNTERS] PROGRAM TOTAL          :    2.8197s
- [COUNTERS] Fortran Overhead ( 0 ) :    2.4552s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.3644s for    90112 events => throughput is 2.47E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.9375s
+ [COUNTERS] Fortran Overhead ( 0 ) :    2.5725s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.3650s for    90112 events => throughput is 2.47E+05 events/s
 
 *** (3) Compare MADEVENT_CUDA x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.5803725748421164E-004) and cpp (1.5803725748421161E-004) differ by less than 2E-14 (2.220446049250313e-16)
+OK! xsec from fortran (1.5803725748421158E-004) and cpp (1.5803725748421166E-004) differ by less than 2E-14 (4.440892098500626e-16)
 
 *** (3) Compare MADEVENT_CUDA x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cuda.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.292727e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.275540e+05                 )  sec^-1
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.508942e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.510631e+05                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 16384 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 4.110003e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.114953e+05                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 16384 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 4.156098e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.167963e+05                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 4096 128 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 4.100637e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.105704e+05                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 4096 128 1 ***
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 4.153684e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.168745e+05                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 65536 8 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 4.109892e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.099883e+05                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 65536 8 1 ***
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.425957e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.425014e+05                 )  sec^-1
 
 TEST COMPLETED
diff --git a/epochX/cudacpp/tmad/logs_ggttgg_mad/log_ggttgg_mad_f_inl0_hrd0.txt b/epochX/cudacpp/tmad/logs_ggttgg_mad/log_ggttgg_mad_f_inl0_hrd0.txt
index dbf6975e6c..5e8ad575df 100644
--- a/epochX/cudacpp/tmad/logs_ggttgg_mad/log_ggttgg_mad_f_inl0_hrd0.txt
+++ b/epochX/cudacpp/tmad/logs_ggttgg_mad/log_ggttgg_mad_f_inl0_hrd0.txt
@@ -16,16 +16,16 @@ make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
 CUDACPP_BUILDDIR='build.sse4_f_inl0_hrd0'
-CUDACPP_BUILDDIR='build.512y_f_inl0_hrd0'
+CUDACPP_BUILDDIR='build.512z_f_inl0_hrd0'
 CUDACPP_BUILDDIR='build.none_f_inl0_hrd0'
+CUDACPP_BUILDDIR='build.512y_f_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
+CUDACPP_BUILDDIR='build.avx2_f_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
-CUDACPP_BUILDDIR='build.512z_f_inl0_hrd0'
-CUDACPP_BUILDDIR='build.avx2_f_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
 make[1]: Nothing to be done for 'all'.
@@ -33,7 +33,7 @@ make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/
 
 OMP_NUM_THREADS=
 
-DATE: 2023-11-24_15:43:32
+DATE: 2024-01-30_06:18:23
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 Working directory (run): /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg
@@ -57,11 +57,11 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggttgg_x1_fortran > /tmp/av
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 2
- [XSECTION] Cross section = 0.0003628 [3.6277277311352988E-004] fbridge_mode=0
+ [XSECTION] Cross section = 0.0003628 [3.6277277311352982E-004] fbridge_mode=0
  [UNWEIGHT] Wrote 48 events (found 439 events)
- [COUNTERS] PROGRAM TOTAL          :    4.5240s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2823s
- [COUNTERS] Fortran MEs      ( 1 ) :    4.2417s for     8192 events => throughput is 1.93E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :    4.7226s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2990s
+ [COUNTERS] Fortran MEs      ( 1 ) :    4.4236s for     8192 events => throughput is 1.85E+03 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x1 (create events.lhe) ***
 --------------------
@@ -82,11 +82,11 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggttgg_x1_fortran > /tmp/av
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 2
- [XSECTION] Cross section = 0.0003628 [3.6277277311352988E-004] fbridge_mode=0
+ [XSECTION] Cross section = 0.0003628 [3.6277277311352982E-004] fbridge_mode=0
  [UNWEIGHT] Wrote 59 events (found 420 events)
- [COUNTERS] PROGRAM TOTAL          :    4.5067s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2792s
- [COUNTERS] Fortran MEs      ( 1 ) :    4.2274s for     8192 events => throughput is 1.94E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :    4.7231s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2979s
+ [COUNTERS] Fortran MEs      ( 1 ) :    4.4253s for     8192 events => throughput is 1.85E+03 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x10 (create events.lhe) ***
 --------------------
@@ -107,11 +107,11 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggttgg_x10_fortran > /tmp/a
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 2
- [XSECTION] Cross section = 0.000158 [1.5803725748421164E-004] fbridge_mode=0
+ [XSECTION] Cross section = 0.000158 [1.5803725748421158E-004] fbridge_mode=0
  [UNWEIGHT] Wrote 207 events (found 1235 events)
- [COUNTERS] PROGRAM TOTAL          :   48.8698s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.9436s
- [COUNTERS] Fortran MEs      ( 1 ) :   46.9262s for    90112 events => throughput is 1.92E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :   50.8315s
+ [COUNTERS] Fortran Overhead ( 0 ) :    2.1047s
+ [COUNTERS] Fortran MEs      ( 1 ) :   48.7267s for    90112 events => throughput is 1.85E+03 events/s
 
 *** (2-none) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -132,15 +132,15 @@ Executing ' ./build.none_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttgg_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 2
- [XSECTION] Cross section = 0.0003628 [3.6277396490802749E-004] fbridge_mode=1
+ [XSECTION] Cross section = 0.0003628 [3.6277396352122325E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 59 events (found 420 events)
- [COUNTERS] PROGRAM TOTAL          :    8.6169s
- [COUNTERS] Fortran Overhead ( 0 ) :    4.3960s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    4.2209s for     8192 events => throughput is 1.94E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :    8.7124s
+ [COUNTERS] Fortran Overhead ( 0 ) :    4.4408s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    4.2715s for     8192 events => throughput is 1.92E+03 events/s
 
 *** (2-none) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (3.6277277311352988E-004) and cpp (3.6277396490802749E-004) differ by less than 4E-4 (3.2852368918590003e-06)
+OK! xsec from fortran (3.6277277311352982E-004) and cpp (3.6277396352122325E-004) differ by less than 4E-4 (3.2814141017745158e-06)
 
 *** (2-none) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -165,29 +165,29 @@ Executing ' ./build.none_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttgg_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 2
- [XSECTION] Cross section = 0.000158 [1.5803774602344628E-004] fbridge_mode=1
+ [XSECTION] Cross section = 0.000158 [1.5803774048965294E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 207 events (found 1235 events)
- [COUNTERS] PROGRAM TOTAL          :   52.7280s
- [COUNTERS] Fortran Overhead ( 0 ) :    6.0827s
- [COUNTERS] CudaCpp MEs      ( 2 ) :   46.6453s for    90112 events => throughput is 1.93E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :   53.4678s
+ [COUNTERS] Fortran Overhead ( 0 ) :    6.2349s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :   47.2329s for    90112 events => throughput is 1.91E+03 events/s
 
 *** (2-none) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.5803725748421164E-004) and cpp (1.5803774602344628E-004) differ by less than 4E-4 (3.0912915247593986e-06)
+OK! xsec from fortran (1.5803725748421158E-004) and cpp (1.5803774048965294E-004) differ by less than 4E-4 (3.056275773571926e-06)
 
 *** (2-none) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.998671e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.973797e+03                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.001275e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.974372e+03                 )  sec^-1
 
 *** (2-sse4) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -208,15 +208,15 @@ Executing ' ./build.sse4_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttgg_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 2
- [XSECTION] Cross section = 0.0003628 [3.6277389126121586E-004] fbridge_mode=1
+ [XSECTION] Cross section = 0.0003628 [3.6277387698033752E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 59 events (found 420 events)
- [COUNTERS] PROGRAM TOTAL          :    2.5615s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.4108s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    1.1507s for     8192 events => throughput is 7.12E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.6573s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.4642s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    1.1930s for     8192 events => throughput is 6.87E+03 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (3.6277277311352988E-004) and cpp (3.6277389126121586E-004) differ by less than 4E-4 (3.0822260348450925e-06)
+OK! xsec from fortran (3.6277277311352982E-004) and cpp (3.6277387698033752E-004) differ by less than 4E-4 (3.0428601303089664e-06)
 
 *** (2-sse4) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -241,29 +241,29 @@ Executing ' ./build.sse4_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttgg_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 2
- [XSECTION] Cross section = 0.000158 [1.5803771887543366E-004] fbridge_mode=1
+ [XSECTION] Cross section = 0.000158 [1.5803770691658365E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 207 events (found 1235 events)
- [COUNTERS] PROGRAM TOTAL          :   15.7555s
- [COUNTERS] Fortran Overhead ( 0 ) :    3.0766s
- [COUNTERS] CudaCpp MEs      ( 2 ) :   12.6789s for    90112 events => throughput is 7.11E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :   16.3220s
+ [COUNTERS] Fortran Overhead ( 0 ) :    3.2267s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :   13.0952s for    90112 events => throughput is 6.88E+03 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.5803725748421164E-004) and cpp (1.5803771887543366E-004) differ by less than 4E-4 (2.9195091675315865e-06)
+OK! xsec from fortran (1.5803725748421158E-004) and cpp (1.5803770691658365E-004) differ by less than 4E-4 (2.8438380874629132e-06)
 
 *** (2-sse4) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 7.214660e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.126754e+03                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 7.260665e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.170734e+03                 )  sec^-1
 
 *** (2-avx2) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -284,15 +284,15 @@ Executing ' ./build.avx2_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttgg_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 2
- [XSECTION] Cross section = 0.0003628 [3.6277390198115864E-004] fbridge_mode=1
+ [XSECTION] Cross section = 0.0003628 [3.6277388844638422E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 59 events (found 420 events)
- [COUNTERS] PROGRAM TOTAL          :    1.2961s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.7839s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.5123s for     8192 events => throughput is 1.60E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    1.3306s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.8070s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.5236s for     8192 events => throughput is 1.56E+04 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (3.6277277311352988E-004) and cpp (3.6277390198115864E-004) differ by less than 4E-4 (3.111776055053639e-06)
+OK! xsec from fortran (3.6277277311352982E-004) and cpp (3.6277388844638422E-004) differ by less than 4E-4 (3.074466820685018e-06)
 
 *** (2-avx2) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -317,29 +317,29 @@ Executing ' ./build.avx2_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttgg_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 2
- [XSECTION] Cross section = 0.000158 [1.5803774416711566E-004] fbridge_mode=1
+ [XSECTION] Cross section = 0.000158 [1.5803773310773457E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 207 events (found 1235 events)
- [COUNTERS] PROGRAM TOTAL          :    8.1063s
- [COUNTERS] Fortran Overhead ( 0 ) :    2.4519s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    5.6543s for    90112 events => throughput is 1.59E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    8.3793s
+ [COUNTERS] Fortran Overhead ( 0 ) :    2.5829s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    5.7964s for    90112 events => throughput is 1.55E+04 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.5803725748421164E-004) and cpp (1.5803774416711566E-004) differ by less than 4E-4 (3.079545366491132e-06)
+OK! xsec from fortran (1.5803725748421158E-004) and cpp (1.5803773310773457E-004) differ by less than 4E-4 (3.0095657856943347e-06)
 
 *** (2-avx2) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.649408e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.587083e+04                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.639467e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.594558e+04                 )  sec^-1
 
 *** (2-512y) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -360,15 +360,15 @@ Executing ' ./build.512y_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttgg_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 2
- [XSECTION] Cross section = 0.0003628 [3.6277390198115864E-004] fbridge_mode=1
+ [XSECTION] Cross section = 0.0003628 [3.6277388844638422E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 59 events (found 420 events)
- [COUNTERS] PROGRAM TOTAL          :    1.1678s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.7178s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.4500s for     8192 events => throughput is 1.82E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    1.2103s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.7457s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.4646s for     8192 events => throughput is 1.76E+04 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (3.6277277311352988E-004) and cpp (3.6277390198115864E-004) differ by less than 4E-4 (3.111776055053639e-06)
+OK! xsec from fortran (3.6277277311352982E-004) and cpp (3.6277388844638422E-004) differ by less than 4E-4 (3.074466820685018e-06)
 
 *** (2-512y) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -393,29 +393,29 @@ Executing ' ./build.512y_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttgg_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 2
- [XSECTION] Cross section = 0.000158 [1.5803774416711566E-004] fbridge_mode=1
+ [XSECTION] Cross section = 0.000158 [1.5803773310773457E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 207 events (found 1235 events)
- [COUNTERS] PROGRAM TOTAL          :    7.3020s
- [COUNTERS] Fortran Overhead ( 0 ) :    2.3817s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    4.9203s for    90112 events => throughput is 1.83E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    7.6352s
+ [COUNTERS] Fortran Overhead ( 0 ) :    2.5218s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    5.1134s for    90112 events => throughput is 1.76E+04 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.5803725748421164E-004) and cpp (1.5803774416711566E-004) differ by less than 4E-4 (3.079545366491132e-06)
+OK! xsec from fortran (1.5803725748421158E-004) and cpp (1.5803773310773457E-004) differ by less than 4E-4 (3.0095657856943347e-06)
 
 *** (2-512y) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.885655e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.751909e+04                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.876738e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.760230e+04                 )  sec^-1
 
 *** (2-512z) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -436,15 +436,15 @@ Executing ' ./build.512z_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttgg_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 2
- [XSECTION] Cross section = 0.0003628 [3.6277396394633404E-004] fbridge_mode=1
+ [XSECTION] Cross section = 0.0003628 [3.6277396133530942E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 59 events (found 420 events)
- [COUNTERS] PROGRAM TOTAL          :    1.3886s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.8337s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.5549s for     8192 events => throughput is 1.48E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    1.4750s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.8777s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.5973s for     8192 events => throughput is 1.37E+04 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (3.6277277311352988E-004) and cpp (3.6277396394633404E-004) differ by less than 4E-4 (3.2825859392904277e-06)
+OK! xsec from fortran (3.6277277311352982E-004) and cpp (3.6277396133530942E-004) differ by less than 4E-4 (3.2753885288450135e-06)
 
 *** (2-512z) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -469,29 +469,29 @@ Executing ' ./build.512z_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttgg_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 2
- [XSECTION] Cross section = 0.000158 [1.5803777741065333E-004] fbridge_mode=1
+ [XSECTION] Cross section = 0.000158 [1.5803777739454609E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 207 events (found 1235 events)
- [COUNTERS] PROGRAM TOTAL          :    8.7321s
- [COUNTERS] Fortran Overhead ( 0 ) :    2.5116s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    6.2205s for    90112 events => throughput is 1.45E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    9.1796s
+ [COUNTERS] Fortran Overhead ( 0 ) :    2.7011s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    6.4785s for    90112 events => throughput is 1.39E+04 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.5803725748421164E-004) and cpp (1.5803777741065333E-004) differ by less than 4E-4 (3.2898979009932106e-06)
+OK! xsec from fortran (1.5803725748421158E-004) and cpp (1.5803777739454609E-004) differ by less than 4E-4 (3.2897959809652377e-06)
 
 *** (2-512z) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.496235e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.411903e+04                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.469765e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.410632e+04                 )  sec^-1
 
 *** (3) EXECUTE MADEVENT_CUDA x1 (create events.lhe) ***
 --------------------
@@ -512,15 +512,15 @@ Executing ' ./build.none_f_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_ggttgg_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 2
- [XSECTION] Cross section = 0.0003628 [3.6277400478491260E-004] fbridge_mode=1
+ [XSECTION] Cross section = 0.0003628 [3.6277400478491265E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 59 events (found 420 events)
- [COUNTERS] PROGRAM TOTAL          :    0.7802s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.7588s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0214s for     8192 events => throughput is 3.82E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.7967s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.7754s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0214s for     8192 events => throughput is 3.83E+05 events/s
 
 *** (3) Compare MADEVENT_CUDA x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (3.6277277311352988E-004) and cpp (3.6277400478491260E-004) differ by less than 4E-4 (3.3951593780834344e-06)
+OK! xsec from fortran (3.6277277311352982E-004) and cpp (3.6277400478491265E-004) differ by less than 4E-4 (3.395159378305479e-06)
 
 *** (3) Compare MADEVENT_CUDA x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -547,56 +547,56 @@ Executing ' ./build.none_f_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_ggttgg_
  [XSECTION] ChannelId = 2
  [XSECTION] Cross section = 0.000158 [1.5803779990154892E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 207 events (found 1235 events)
- [COUNTERS] PROGRAM TOTAL          :    2.6640s
- [COUNTERS] Fortran Overhead ( 0 ) :    2.4295s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.2346s for    90112 events => throughput is 3.84E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.7835s
+ [COUNTERS] Fortran Overhead ( 0 ) :    2.5473s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.2361s for    90112 events => throughput is 3.82E+05 events/s
 
 *** (3) Compare MADEVENT_CUDA x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.5803725748421164E-004) and cpp (1.5803779990154892E-004) differ by less than 4E-4 (3.4322117830054566e-06)
+OK! xsec from fortran (1.5803725748421158E-004) and cpp (1.5803779990154892E-004) differ by less than 4E-4 (3.432211783227501e-06)
 
 *** (3) Compare MADEVENT_CUDA x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cuda.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.580570e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.582485e+05                 )  sec^-1
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.934731e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.942798e+05                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 16384 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 8.492770e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.492976e+05                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 16384 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 8.733354e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.638150e+05                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 4096 128 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 8.509140e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.493239e+05                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 4096 128 1 ***
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 8.727290e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.638925e+05                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 65536 8 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 8.483888e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.453709e+05                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 65536 8 1 ***
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.531454e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.527726e+05                 )  sec^-1
 
 TEST COMPLETED
diff --git a/epochX/cudacpp/tmad/logs_ggttgg_mad/log_ggttgg_mad_m_inl0_hrd0.txt b/epochX/cudacpp/tmad/logs_ggttgg_mad/log_ggttgg_mad_m_inl0_hrd0.txt
index cfbc1973b9..a372850ebe 100644
--- a/epochX/cudacpp/tmad/logs_ggttgg_mad/log_ggttgg_mad_m_inl0_hrd0.txt
+++ b/epochX/cudacpp/tmad/logs_ggttgg_mad/log_ggttgg_mad_m_inl0_hrd0.txt
@@ -1,11 +1,11 @@
 Working directory (build): /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg
 CUDACPP_BUILDDIR='.'
 
-make USEBUILDDIR=1 AVX=none
 
 
-make USEBUILDDIR=1 AVX=sse4
 
+make USEBUILDDIR=1 AVX=none
+make USEBUILDDIR=1 AVX=sse4
 make USEBUILDDIR=1 AVX=avx2
 make USEBUILDDIR=1 AVX=512y
 
@@ -16,14 +16,14 @@ make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
 CUDACPP_BUILDDIR='build.none_m_inl0_hrd0'
-CUDACPP_BUILDDIR='build.sse4_m_inl0_hrd0'
-CUDACPP_BUILDDIR='build.512y_m_inl0_hrd0'
 CUDACPP_BUILDDIR='build.avx2_m_inl0_hrd0'
-CUDACPP_BUILDDIR='build.512z_m_inl0_hrd0'
+CUDACPP_BUILDDIR='build.512y_m_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
+CUDACPP_BUILDDIR='build.512z_m_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
+CUDACPP_BUILDDIR='build.sse4_m_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
 make[1]: Nothing to be done for 'all'.
@@ -33,7 +33,7 @@ make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/
 
 OMP_NUM_THREADS=
 
-DATE: 2023-11-24_15:46:56
+DATE: 2024-01-30_06:21:53
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 Working directory (run): /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg
@@ -57,11 +57,11 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggttgg_x1_fortran > /tmp/av
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 2
- [XSECTION] Cross section = 0.0003628 [3.6277277311352988E-004] fbridge_mode=0
+ [XSECTION] Cross section = 0.0003628 [3.6277277311352982E-004] fbridge_mode=0
  [UNWEIGHT] Wrote 48 events (found 439 events)
- [COUNTERS] PROGRAM TOTAL          :    4.5513s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2808s
- [COUNTERS] Fortran MEs      ( 1 ) :    4.2704s for     8192 events => throughput is 1.92E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :    4.7321s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3006s
+ [COUNTERS] Fortran MEs      ( 1 ) :    4.4315s for     8192 events => throughput is 1.85E+03 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x1 (create events.lhe) ***
 --------------------
@@ -82,11 +82,11 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggttgg_x1_fortran > /tmp/av
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 2
- [XSECTION] Cross section = 0.0003628 [3.6277277311352988E-004] fbridge_mode=0
+ [XSECTION] Cross section = 0.0003628 [3.6277277311352982E-004] fbridge_mode=0
  [UNWEIGHT] Wrote 59 events (found 420 events)
- [COUNTERS] PROGRAM TOTAL          :    4.5272s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2803s
- [COUNTERS] Fortran MEs      ( 1 ) :    4.2469s for     8192 events => throughput is 1.93E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :    4.7365s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2954s
+ [COUNTERS] Fortran MEs      ( 1 ) :    4.4410s for     8192 events => throughput is 1.84E+03 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x10 (create events.lhe) ***
 --------------------
@@ -107,11 +107,11 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggttgg_x10_fortran > /tmp/a
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 2
- [XSECTION] Cross section = 0.000158 [1.5803725748421164E-004] fbridge_mode=0
+ [XSECTION] Cross section = 0.000158 [1.5803725748421158E-004] fbridge_mode=0
  [UNWEIGHT] Wrote 207 events (found 1235 events)
- [COUNTERS] PROGRAM TOTAL          :   48.8098s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.9495s
- [COUNTERS] Fortran MEs      ( 1 ) :   46.8603s for    90112 events => throughput is 1.92E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :   50.8810s
+ [COUNTERS] Fortran Overhead ( 0 ) :    2.1085s
+ [COUNTERS] Fortran MEs      ( 1 ) :   48.7725s for    90112 events => throughput is 1.85E+03 events/s
 
 *** (2-none) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -134,13 +134,13 @@ Executing ' ./build.none_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttgg_x
  [XSECTION] ChannelId = 2
  [XSECTION] Cross section = 0.0003628 [3.6277277432965013E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 59 events (found 420 events)
- [COUNTERS] PROGRAM TOTAL          :    9.0359s
- [COUNTERS] Fortran Overhead ( 0 ) :    4.6142s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    4.4217s for     8192 events => throughput is 1.85E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :    9.6340s
+ [COUNTERS] Fortran Overhead ( 0 ) :    4.9027s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    4.7313s for     8192 events => throughput is 1.73E+03 events/s
 
 *** (2-none) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (3.6277277311352988E-004) and cpp (3.6277277432965013E-004) differ by less than 2E-4 (3.352291999547674e-09)
+OK! xsec from fortran (3.6277277311352982E-004) and cpp (3.6277277432965013E-004) differ by less than 2E-4 (3.352291999547674e-09)
 
 *** (2-none) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -165,29 +165,29 @@ Executing ' ./build.none_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttgg_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 2
- [XSECTION] Cross section = 0.000158 [1.5803725813026109E-004] fbridge_mode=1
+ [XSECTION] Cross section = 0.000158 [1.5803725813026107E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 207 events (found 1235 events)
- [COUNTERS] PROGRAM TOTAL          :   54.8666s
- [COUNTERS] Fortran Overhead ( 0 ) :    6.2546s
- [COUNTERS] CudaCpp MEs      ( 2 ) :   48.6120s for    90112 events => throughput is 1.85E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :   58.7075s
+ [COUNTERS] Fortran Overhead ( 0 ) :    6.6716s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :   52.0359s for    90112 events => throughput is 1.73E+03 events/s
 
 *** (2-none) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.5803725748421164E-004) and cpp (1.5803725813026109E-004) differ by less than 2E-4 (4.087956639864387e-09)
+OK! xsec from fortran (1.5803725748421158E-004) and cpp (1.5803725813026107E-004) differ by less than 2E-4 (4.087956861908992e-09)
 
 *** (2-none) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+BRDHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.917331e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.784786e+03                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.930542e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.785841e+03                 )  sec^-1
 
 *** (2-sse4) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -208,15 +208,15 @@ Executing ' ./build.sse4_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttgg_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 2
- [XSECTION] Cross section = 0.0003628 [3.6277277430934464E-004] fbridge_mode=1
+ [XSECTION] Cross section = 0.0003628 [3.6277277430934459E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 59 events (found 420 events)
- [COUNTERS] PROGRAM TOTAL          :    4.7788s
- [COUNTERS] Fortran Overhead ( 0 ) :    2.4963s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    2.2825s for     8192 events => throughput is 3.59E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :    5.0010s
+ [COUNTERS] Fortran Overhead ( 0 ) :    2.6173s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    2.3837s for     8192 events => throughput is 3.44E+03 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (3.6277277311352988E-004) and cpp (3.6277277430934464E-004) differ by less than 2E-4 (3.296318995538172e-09)
+OK! xsec from fortran (3.6277277311352982E-004) and cpp (3.6277277430934459E-004) differ by less than 2E-4 (3.296318995538172e-09)
 
 *** (2-sse4) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -241,29 +241,29 @@ Executing ' ./build.sse4_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttgg_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 2
- [XSECTION] Cross section = 0.000158 [1.5803725816246317E-004] fbridge_mode=1
+ [XSECTION] Cross section = 0.000158 [1.5803725816246315E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 207 events (found 1235 events)
- [COUNTERS] PROGRAM TOTAL          :   29.3040s
- [COUNTERS] Fortran Overhead ( 0 ) :    4.1597s
- [COUNTERS] CudaCpp MEs      ( 2 ) :   25.1443s for    90112 events => throughput is 3.58E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :   30.6721s
+ [COUNTERS] Fortran Overhead ( 0 ) :    4.4448s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :   26.2273s for    90112 events => throughput is 3.44E+03 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.5803725748421164E-004) and cpp (1.5803725816246317E-004) differ by less than 2E-4 (4.291719202242916e-09)
+OK! xsec from fortran (1.5803725748421158E-004) and cpp (1.5803725816246315E-004) differ by less than 2E-4 (4.291719424287521e-09)
 
 *** (2-sse4) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+BRDHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.675549e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.519557e+03                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.694435e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.539213e+03                 )  sec^-1
 
 *** (2-avx2) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -286,13 +286,13 @@ Executing ' ./build.avx2_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttgg_x
  [XSECTION] ChannelId = 2
  [XSECTION] Cross section = 0.0003628 [3.6277277419683297E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 59 events (found 420 events)
- [COUNTERS] PROGRAM TOTAL          :    2.2311s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.2435s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.9875s for     8192 events => throughput is 8.30E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.3362s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.3065s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    1.0298s for     8192 events => throughput is 7.96E+03 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (3.6277277311352988E-004) and cpp (3.6277277419683297E-004) differ by less than 2E-4 (2.9861753070292707e-09)
+OK! xsec from fortran (3.6277277311352982E-004) and cpp (3.6277277419683297E-004) differ by less than 2E-4 (2.9861755290738756e-09)
 
 *** (2-avx2) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -319,27 +319,27 @@ Executing ' ./build.avx2_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttgg_x
  [XSECTION] ChannelId = 2
  [XSECTION] Cross section = 0.000158 [1.5803725810769321E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 207 events (found 1235 events)
- [COUNTERS] PROGRAM TOTAL          :   13.8690s
- [COUNTERS] Fortran Overhead ( 0 ) :    2.9050s
- [COUNTERS] CudaCpp MEs      ( 2 ) :   10.9640s for    90112 events => throughput is 8.22E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :   14.5206s
+ [COUNTERS] Fortran Overhead ( 0 ) :    3.0822s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :   11.4384s for    90112 events => throughput is 7.88E+03 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.5803725748421164E-004) and cpp (1.5803725810769321E-004) differ by less than 2E-4 (3.945155535589606e-09)
+OK! xsec from fortran (1.5803725748421158E-004) and cpp (1.5803725810769321E-004) differ by less than 2E-4 (3.945155979678816e-09)
 
 *** (2-avx2) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+BRDHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 8.536816e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.081577e+03                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 8.511555e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.110336e+03                 )  sec^-1
 
 *** (2-512y) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -362,13 +362,13 @@ Executing ' ./build.512y_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttgg_x
  [XSECTION] ChannelId = 2
  [XSECTION] Cross section = 0.0003628 [3.6277277419683297E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 59 events (found 420 events)
- [COUNTERS] PROGRAM TOTAL          :    2.0078s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.1337s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.8741s for     8192 events => throughput is 9.37E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.0953s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.1807s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.9146s for     8192 events => throughput is 8.96E+03 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (3.6277277311352988E-004) and cpp (3.6277277419683297E-004) differ by less than 2E-4 (2.9861753070292707e-09)
+OK! xsec from fortran (3.6277277311352982E-004) and cpp (3.6277277419683297E-004) differ by less than 2E-4 (2.9861755290738756e-09)
 
 *** (2-512y) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -395,27 +395,27 @@ Executing ' ./build.512y_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttgg_x
  [XSECTION] ChannelId = 2
  [XSECTION] Cross section = 0.000158 [1.5803725810769321E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 207 events (found 1235 events)
- [COUNTERS] PROGRAM TOTAL          :   12.4367s
- [COUNTERS] Fortran Overhead ( 0 ) :    2.7931s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    9.6436s for    90112 events => throughput is 9.34E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :   13.0092s
+ [COUNTERS] Fortran Overhead ( 0 ) :    2.9623s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :   10.0469s for    90112 events => throughput is 8.97E+03 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.5803725748421164E-004) and cpp (1.5803725810769321E-004) differ by less than 2E-4 (3.945155535589606e-09)
+OK! xsec from fortran (1.5803725748421158E-004) and cpp (1.5803725810769321E-004) differ by less than 2E-4 (3.945155979678816e-09)
 
 *** (2-512y) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+BRDHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 9.658673e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.305280e+03                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 9.653676e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.278381e+03                 )  sec^-1
 
 *** (2-512z) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -438,13 +438,13 @@ Executing ' ./build.512z_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttgg_x
  [XSECTION] ChannelId = 2
  [XSECTION] Cross section = 0.0003628 [3.6277277419683297E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 59 events (found 420 events)
- [COUNTERS] PROGRAM TOTAL          :    2.5071s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.3899s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    1.1172s for     8192 events => throughput is 7.33E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.6707s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.4755s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    1.1952s for     8192 events => throughput is 6.85E+03 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (3.6277277311352988E-004) and cpp (3.6277277419683297E-004) differ by less than 2E-4 (2.9861753070292707e-09)
+OK! xsec from fortran (3.6277277311352982E-004) and cpp (3.6277277419683297E-004) differ by less than 2E-4 (2.9861755290738756e-09)
 
 *** (2-512z) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -471,27 +471,27 @@ Executing ' ./build.512z_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttgg_x
  [XSECTION] ChannelId = 2
  [XSECTION] Cross section = 0.000158 [1.5803725810769321E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 207 events (found 1235 events)
- [COUNTERS] PROGRAM TOTAL          :   15.4633s
- [COUNTERS] Fortran Overhead ( 0 ) :    3.0792s
- [COUNTERS] CudaCpp MEs      ( 2 ) :   12.3840s for    90112 events => throughput is 7.28E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :   16.3107s
+ [COUNTERS] Fortran Overhead ( 0 ) :    3.2346s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :   13.0761s for    90112 events => throughput is 6.89E+03 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.5803725748421164E-004) and cpp (1.5803725810769321E-004) differ by less than 2E-4 (3.945155535589606e-09)
+OK! xsec from fortran (1.5803725748421158E-004) and cpp (1.5803725810769321E-004) differ by less than 2E-4 (3.945155979678816e-09)
 
 *** (2-512z) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+BRDHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 7.401661e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.979300e+03                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 7.428706e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.981216e+03                 )  sec^-1
 
 *** (3) EXECUTE MADEVENT_CUDA x1 (create events.lhe) ***
 --------------------
@@ -512,15 +512,15 @@ Executing ' ./build.none_m_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_ggttgg_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 2
- [XSECTION] Cross section = 0.0003628 [3.6277277293084707E-004] fbridge_mode=1
+ [XSECTION] Cross section = 0.0003628 [3.6277277293084701E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 59 events (found 420 events)
- [COUNTERS] PROGRAM TOTAL          :    0.8222s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.7891s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0330s for     8192 events => throughput is 2.48E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.8344s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.8014s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0331s for     8192 events => throughput is 2.48E+05 events/s
 
 *** (3) Compare MADEVENT_CUDA x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (3.6277277311352988E-004) and cpp (3.6277277293084707E-004) differ by less than 2E-4 (5.03573627241849e-10)
+OK! xsec from fortran (3.6277277311352982E-004) and cpp (3.6277277293084701E-004) differ by less than 2E-4 (5.03573627241849e-10)
 
 *** (3) Compare MADEVENT_CUDA x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -547,56 +547,56 @@ Executing ' ./build.none_m_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_ggttgg_
  [XSECTION] ChannelId = 2
  [XSECTION] Cross section = 0.000158 [1.5803725738731039E-004] fbridge_mode=1
  [UNWEIGHT] Wrote 207 events (found 1235 events)
- [COUNTERS] PROGRAM TOTAL          :    2.8170s
- [COUNTERS] Fortran Overhead ( 0 ) :    2.4535s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.3635s for    90112 events => throughput is 2.48E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.9283s
+ [COUNTERS] Fortran Overhead ( 0 ) :    2.5651s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.3632s for    90112 events => throughput is 2.48E+05 events/s
 
 *** (3) Compare MADEVENT_CUDA x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.5803725748421164E-004) and cpp (1.5803725738731039E-004) differ by less than 2E-4 (6.131544161291913e-10)
+OK! xsec from fortran (1.5803725748421158E-004) and cpp (1.5803725738731039E-004) differ by less than 2E-4 (6.131540830622839e-10)
 
 *** (3) Compare MADEVENT_CUDA x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cuda.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.293320e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.286633e+05                 )  sec^-1
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.523576e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.522365e+05                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 16384 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 4.111282e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.122442e+05                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 16384 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 4.159178e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.148040e+05                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 4096 128 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 4.106507e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.112221e+05                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 4096 128 1 ***
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 4.157638e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.164785e+05                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 65536 8 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 4.099330e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.108117e+05                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 65536 8 1 ***
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.430536e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.430780e+05                 )  sec^-1
 
 TEST COMPLETED
diff --git a/epochX/cudacpp/tmad/logs_ggttggg_mad/log_ggttggg_mad_d_inl0_hrd0.txt b/epochX/cudacpp/tmad/logs_ggttggg_mad/log_ggttggg_mad_d_inl0_hrd0.txt
index 4f16911127..bc47a109df 100644
--- a/epochX/cudacpp/tmad/logs_ggttggg_mad/log_ggttggg_mad_d_inl0_hrd0.txt
+++ b/epochX/cudacpp/tmad/logs_ggttggg_mad/log_ggttggg_mad_d_inl0_hrd0.txt
@@ -1,12 +1,12 @@
 Working directory (build): /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg
 CUDACPP_BUILDDIR='.'
 
-make USEBUILDDIR=1 AVX=none
 
+make USEBUILDDIR=1 AVX=none
 
-make USEBUILDDIR=1 AVX=avx2
 
 make USEBUILDDIR=1 AVX=sse4
+make USEBUILDDIR=1 AVX=avx2
 make USEBUILDDIR=1 AVX=512y
 
 make USEBUILDDIR=1 AVX=512z
@@ -15,10 +15,10 @@ make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg'
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg'
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg'
-CUDACPP_BUILDDIR='build.none_d_inl0_hrd0'
 CUDACPP_BUILDDIR='build.512z_d_inl0_hrd0'
 CUDACPP_BUILDDIR='build.512y_d_inl0_hrd0'
-CUDACPP_BUILDDIR='build.avx2_d_inl0_hrd0'
+CUDACPP_BUILDDIR='build.sse4_d_inl0_hrd0'
+CUDACPP_BUILDDIR='build.none_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg'
 make[1]: Nothing to be done for 'all'.
@@ -27,13 +27,13 @@ make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg'
-CUDACPP_BUILDDIR='build.sse4_d_inl0_hrd0'
+CUDACPP_BUILDDIR='build.avx2_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg'
 
 OMP_NUM_THREADS=
 
-DATE: 2023-11-24_15:52:44
+DATE: 2024-01-30_06:27:59
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 Working directory (run): /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg
@@ -57,11 +57,11 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggttggg_x1_fortran > /tmp/a
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 1.169e-06 [1.1693100945435802E-006] fbridge_mode=0
+ [XSECTION] Cross section = 1.169e-06 [1.1693100945435806E-006] fbridge_mode=0
  [UNWEIGHT] Wrote 1 events (found 166 events)
- [COUNTERS] PROGRAM TOTAL          :   98.5141s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.4738s
- [COUNTERS] Fortran MEs      ( 1 ) :   98.0403s for     8192 events => throughput is 8.36E+01 events/s
+ [COUNTERS] PROGRAM TOTAL          :  101.9143s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.4823s
+ [COUNTERS] Fortran MEs      ( 1 ) :  101.4320s for     8192 events => throughput is 8.08E+01 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x1 (create events.lhe) ***
 --------------------
@@ -82,11 +82,11 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggttggg_x1_fortran > /tmp/a
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 1.169e-06 [1.1693100945435802E-006] fbridge_mode=0
+ [XSECTION] Cross section = 1.169e-06 [1.1693100945435806E-006] fbridge_mode=0
  [UNWEIGHT] Wrote 15 events (found 163 events)
- [COUNTERS] PROGRAM TOTAL          :   98.4757s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.4682s
- [COUNTERS] Fortran MEs      ( 1 ) :   98.0075s for     8192 events => throughput is 8.36E+01 events/s
+ [COUNTERS] PROGRAM TOTAL          :  101.8572s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.4740s
+ [COUNTERS] Fortran MEs      ( 1 ) :  101.3832s for     8192 events => throughput is 8.08E+01 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x10 (create events.lhe) ***
 --------------------
@@ -107,11 +107,11 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggttggg_x10_fortran > /tmp/
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 2.136e-07 [2.1358436158813976E-007] fbridge_mode=0
+ [XSECTION] Cross section = 2.136e-07 [2.1358436158813979E-007] fbridge_mode=0
  [UNWEIGHT] Wrote 84 events (found 808 events)
- [COUNTERS] PROGRAM TOTAL          : 1079.6019s
- [COUNTERS] Fortran Overhead ( 0 ) :    4.3004s
- [COUNTERS] Fortran MEs      ( 1 ) : 1075.3015s for    90112 events => throughput is 8.38E+01 events/s
+ [COUNTERS] PROGRAM TOTAL          : 1118.0575s
+ [COUNTERS] Fortran Overhead ( 0 ) :    4.3730s
+ [COUNTERS] Fortran MEs      ( 1 ) : 1113.6844s for    90112 events => throughput is 8.09E+01 events/s
 
 *** (2-none) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -134,13 +134,13 @@ Executing ' ./build.none_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttggg_
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 1.169e-06 [1.1693100945435831E-006] fbridge_mode=1
  [UNWEIGHT] Wrote 15 events (found 163 events)
- [COUNTERS] PROGRAM TOTAL          :  225.7194s
- [COUNTERS] Fortran Overhead ( 0 ) :  104.5284s
- [COUNTERS] CudaCpp MEs      ( 2 ) :  121.1910s for     8192 events => throughput is 6.76E+01 events/s
+ [COUNTERS] PROGRAM TOTAL          :  222.3358s
+ [COUNTERS] Fortran Overhead ( 0 ) :  102.7450s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :  119.5908s for     8192 events => throughput is 6.85E+01 events/s
 
 *** (2-none) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.1693100945435802E-006) and cpp (1.1693100945435831E-006) differ by less than 2E-14 (2.4424906541753444e-15)
+OK! xsec from fortran (1.1693100945435806E-006) and cpp (1.1693100945435831E-006) differ by less than 2E-14 (2.220446049250313e-15)
 
 *** (2-none) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -165,29 +165,29 @@ Executing ' ./build.none_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttggg_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 2.136e-07 [2.1358436158813953E-007] fbridge_mode=1
+ [XSECTION] Cross section = 2.136e-07 [2.1358436158813950E-007] fbridge_mode=1
  [UNWEIGHT] Wrote 84 events (found 808 events)
- [COUNTERS] PROGRAM TOTAL          : 1418.6305s
- [COUNTERS] Fortran Overhead ( 0 ) :  107.9097s
- [COUNTERS] CudaCpp MEs      ( 2 ) : 1310.7208s for    90112 events => throughput is 6.87E+01 events/s
+ [COUNTERS] PROGRAM TOTAL          : 1439.6055s
+ [COUNTERS] Fortran Overhead ( 0 ) :  107.7197s
+ [COUNTERS] CudaCpp MEs      ( 2 ) : 1331.8857s for    90112 events => throughput is 6.77E+01 events/s
 
 *** (2-none) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (2.1358436158813976E-007) and cpp (2.1358436158813953E-007) differ by less than 2E-14 (1.1102230246251565e-15)
+OK! xsec from fortran (2.1358436158813979E-007) and cpp (2.1358436158813950E-007) differ by less than 2E-14 (1.3322676295501878e-15)
 
 *** (2-none) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 7.929728e+01                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.948640e+01                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 6.986885e+01                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.570768e+01                 )  sec^-1
 
 *** (2-sse4) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -208,15 +208,15 @@ Executing ' ./build.sse4_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttggg_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 1.169e-06 [1.1693100945435827E-006] fbridge_mode=1
+ [XSECTION] Cross section = 1.169e-06 [1.1693100945435831E-006] fbridge_mode=1
  [UNWEIGHT] Wrote 15 events (found 163 events)
- [COUNTERS] PROGRAM TOTAL          :  110.1960s
- [COUNTERS] Fortran Overhead ( 0 ) :   50.7639s
- [COUNTERS] CudaCpp MEs      ( 2 ) :   59.4321s for     8192 events => throughput is 1.38E+02 events/s
+ [COUNTERS] PROGRAM TOTAL          :  115.4155s
+ [COUNTERS] Fortran Overhead ( 0 ) :   52.8722s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :   62.5433s for     8192 events => throughput is 1.31E+02 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.1693100945435802E-006) and cpp (1.1693100945435827E-006) differ by less than 2E-14 (2.220446049250313e-15)
+OK! xsec from fortran (1.1693100945435806E-006) and cpp (1.1693100945435831E-006) differ by less than 2E-14 (2.220446049250313e-15)
 
 *** (2-sse4) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -243,27 +243,27 @@ Executing ' ./build.sse4_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttggg_
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 2.136e-07 [2.1358436158813958E-007] fbridge_mode=1
  [UNWEIGHT] Wrote 84 events (found 808 events)
- [COUNTERS] PROGRAM TOTAL          :  706.5283s
- [COUNTERS] Fortran Overhead ( 0 ) :   54.7734s
- [COUNTERS] CudaCpp MEs      ( 2 ) :  651.7549s for    90112 events => throughput is 1.38E+02 events/s
+ [COUNTERS] PROGRAM TOTAL          :  742.6112s
+ [COUNTERS] Fortran Overhead ( 0 ) :   56.7177s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :  685.8936s for    90112 events => throughput is 1.31E+02 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (2.1358436158813976E-007) and cpp (2.1358436158813958E-007) differ by less than 2E-14 (8.881784197001252e-16)
+OK! xsec from fortran (2.1358436158813979E-007) and cpp (2.1358436158813958E-007) differ by less than 2E-14 (9.992007221626409e-16)
 
 *** (2-sse4) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.649511e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.569503e+02                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.642121e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.568080e+02                 )  sec^-1
 
 *** (2-avx2) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -284,15 +284,15 @@ Executing ' ./build.avx2_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttggg_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 1.169e-06 [1.1693100945435829E-006] fbridge_mode=1
+ [XSECTION] Cross section = 1.169e-06 [1.1693100945435827E-006] fbridge_mode=1
  [UNWEIGHT] Wrote 15 events (found 163 events)
- [COUNTERS] PROGRAM TOTAL          :   51.5856s
- [COUNTERS] Fortran Overhead ( 0 ) :   23.9016s
- [COUNTERS] CudaCpp MEs      ( 2 ) :   27.6840s for     8192 events => throughput is 2.96E+02 events/s
+ [COUNTERS] PROGRAM TOTAL          :   53.0772s
+ [COUNTERS] Fortran Overhead ( 0 ) :   24.7915s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :   28.2857s for     8192 events => throughput is 2.90E+02 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.1693100945435802E-006) and cpp (1.1693100945435829E-006) differ by less than 2E-14 (2.4424906541753444e-15)
+OK! xsec from fortran (1.1693100945435806E-006) and cpp (1.1693100945435827E-006) differ by less than 2E-14 (1.7763568394002505e-15)
 
 *** (2-avx2) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -319,27 +319,27 @@ Executing ' ./build.avx2_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttggg_
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 2.136e-07 [2.1358436158813958E-007] fbridge_mode=1
  [UNWEIGHT] Wrote 84 events (found 808 events)
- [COUNTERS] PROGRAM TOTAL          :  331.4778s
- [COUNTERS] Fortran Overhead ( 0 ) :   27.5520s
- [COUNTERS] CudaCpp MEs      ( 2 ) :  303.9259s for    90112 events => throughput is 2.96E+02 events/s
+ [COUNTERS] PROGRAM TOTAL          :  340.2848s
+ [COUNTERS] Fortran Overhead ( 0 ) :   28.6537s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :  311.6311s for    90112 events => throughput is 2.89E+02 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (2.1358436158813976E-007) and cpp (2.1358436158813958E-007) differ by less than 2E-14 (8.881784197001252e-16)
+OK! xsec from fortran (2.1358436158813979E-007) and cpp (2.1358436158813958E-007) differ by less than 2E-14 (9.992007221626409e-16)
 
 *** (2-avx2) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.538462e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.398486e+02                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.559381e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.399363e+02                 )  sec^-1
 
 *** (2-512y) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -360,15 +360,15 @@ Executing ' ./build.512y_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttggg_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 1.169e-06 [1.1693100945435829E-006] fbridge_mode=1
+ [XSECTION] Cross section = 1.169e-06 [1.1693100945435827E-006] fbridge_mode=1
  [UNWEIGHT] Wrote 15 events (found 163 events)
- [COUNTERS] PROGRAM TOTAL          :   45.7105s
- [COUNTERS] Fortran Overhead ( 0 ) :   20.8116s
- [COUNTERS] CudaCpp MEs      ( 2 ) :   24.8989s for     8192 events => throughput is 3.29E+02 events/s
+ [COUNTERS] PROGRAM TOTAL          :   46.8647s
+ [COUNTERS] Fortran Overhead ( 0 ) :   21.6683s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :   25.1965s for     8192 events => throughput is 3.25E+02 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.1693100945435802E-006) and cpp (1.1693100945435829E-006) differ by less than 2E-14 (2.4424906541753444e-15)
+OK! xsec from fortran (1.1693100945435806E-006) and cpp (1.1693100945435827E-006) differ by less than 2E-14 (1.7763568394002505e-15)
 
 *** (2-512y) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -395,27 +395,27 @@ Executing ' ./build.512y_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttggg_
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 2.136e-07 [2.1358436158813958E-007] fbridge_mode=1
  [UNWEIGHT] Wrote 84 events (found 808 events)
- [COUNTERS] PROGRAM TOTAL          :  299.0798s
- [COUNTERS] Fortran Overhead ( 0 ) :   24.6512s
- [COUNTERS] CudaCpp MEs      ( 2 ) :  274.4286s for    90112 events => throughput is 3.28E+02 events/s
+ [COUNTERS] PROGRAM TOTAL          :  302.3706s
+ [COUNTERS] Fortran Overhead ( 0 ) :   25.4811s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :  276.8895s for    90112 events => throughput is 3.25E+02 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (2.1358436158813976E-007) and cpp (2.1358436158813958E-007) differ by less than 2E-14 (8.881784197001252e-16)
+OK! xsec from fortran (2.1358436158813979E-007) and cpp (2.1358436158813958E-007) differ by less than 2E-14 (9.992007221626409e-16)
 
 *** (2-512y) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.987523e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.882449e+02                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.957972e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.889860e+02                 )  sec^-1
 
 *** (2-512z) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -436,15 +436,15 @@ Executing ' ./build.512z_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttggg_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 1.169e-06 [1.1693100945435829E-006] fbridge_mode=1
+ [XSECTION] Cross section = 1.169e-06 [1.1693100945435827E-006] fbridge_mode=1
  [UNWEIGHT] Wrote 15 events (found 163 events)
- [COUNTERS] PROGRAM TOTAL          :   47.6904s
- [COUNTERS] Fortran Overhead ( 0 ) :   23.3628s
- [COUNTERS] CudaCpp MEs      ( 2 ) :   24.3276s for     8192 events => throughput is 3.37E+02 events/s
+ [COUNTERS] PROGRAM TOTAL          :   50.0231s
+ [COUNTERS] Fortran Overhead ( 0 ) :   24.6335s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :   25.3895s for     8192 events => throughput is 3.23E+02 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.1693100945435802E-006) and cpp (1.1693100945435829E-006) differ by less than 2E-14 (2.4424906541753444e-15)
+OK! xsec from fortran (1.1693100945435806E-006) and cpp (1.1693100945435827E-006) differ by less than 2E-14 (1.7763568394002505e-15)
 
 *** (2-512z) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -471,27 +471,27 @@ Executing ' ./build.512z_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttggg_
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 2.136e-07 [2.1358436158813958E-007] fbridge_mode=1
  [UNWEIGHT] Wrote 84 events (found 808 events)
- [COUNTERS] PROGRAM TOTAL          :  293.8850s
- [COUNTERS] Fortran Overhead ( 0 ) :   27.2969s
- [COUNTERS] CudaCpp MEs      ( 2 ) :  266.5881s for    90112 events => throughput is 3.38E+02 events/s
+ [COUNTERS] PROGRAM TOTAL          :  308.5181s
+ [COUNTERS] Fortran Overhead ( 0 ) :   28.4720s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :  280.0461s for    90112 events => throughput is 3.22E+02 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (2.1358436158813976E-007) and cpp (2.1358436158813958E-007) differ by less than 2E-14 (8.881784197001252e-16)
+OK! xsec from fortran (2.1358436158813979E-007) and cpp (2.1358436158813958E-007) differ by less than 2E-14 (9.992007221626409e-16)
 
 *** (2-512z) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.560488e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.386684e+02                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.577427e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.385729e+02                 )  sec^-1
 
 *** (3) EXECUTE MADEVENT_CUDA x1 (create events.lhe) ***
 --------------------
@@ -512,15 +512,15 @@ Executing ' ./build.none_d_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_ggttggg
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 1.169e-06 [1.1693100945435838E-006] fbridge_mode=1
+ [XSECTION] Cross section = 1.169e-06 [1.1693100945435829E-006] fbridge_mode=1
  [UNWEIGHT] Wrote 15 events (found 163 events)
- [COUNTERS] PROGRAM TOTAL          :    4.2450s
- [COUNTERS] Fortran Overhead ( 0 ) :    3.1602s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    1.0848s for     8192 events => throughput is 7.55E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :    4.2467s
+ [COUNTERS] Fortran Overhead ( 0 ) :    3.1625s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    1.0842s for     8192 events => throughput is 7.56E+03 events/s
 
 *** (3) Compare MADEVENT_CUDA x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.1693100945435802E-006) and cpp (1.1693100945435838E-006) differ by less than 2E-14 (3.1086244689504383e-15)
+OK! xsec from fortran (1.1693100945435806E-006) and cpp (1.1693100945435829E-006) differ by less than 2E-14 (1.9984014443252818e-15)
 
 *** (3) Compare MADEVENT_CUDA x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -545,58 +545,58 @@ Executing ' ./build.none_d_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_ggttggg
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 2.136e-07 [2.1358436158813958E-007] fbridge_mode=1
+ [XSECTION] Cross section = 2.136e-07 [2.1358436158813960E-007] fbridge_mode=1
  [UNWEIGHT] Wrote 84 events (found 808 events)
- [COUNTERS] PROGRAM TOTAL          :   18.7885s
- [COUNTERS] Fortran Overhead ( 0 ) :    6.9034s
- [COUNTERS] CudaCpp MEs      ( 2 ) :   11.8851s for    90112 events => throughput is 7.58E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :   18.9518s
+ [COUNTERS] Fortran Overhead ( 0 ) :    7.0338s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :   11.9179s for    90112 events => throughput is 7.56E+03 events/s
 
 *** (3) Compare MADEVENT_CUDA x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (2.1358436158813976E-007) and cpp (2.1358436158813958E-007) differ by less than 2E-14 (8.881784197001252e-16)
+OK! xsec from fortran (2.1358436158813979E-007) and cpp (2.1358436158813960E-007) differ by less than 2E-14 (8.881784197001252e-16)
 
 *** (3) Compare MADEVENT_CUDA x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cuda.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 7.552834e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.528868e+03                 )  sec^-1
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 9.253212e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.249701e+03                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 512 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 9.219027e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.231891e+03                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 512 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 9.567176e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.557033e+03                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 128 128 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 9.273376e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.244700e+03                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 128 128 1 ***
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 9.442640e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.446855e+03                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 2048 8 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 9.208706e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.214530e+03                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 2048 8 1 ***
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.238633e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.244468e+03                 )  sec^-1
 
 TEST COMPLETED
diff --git a/epochX/cudacpp/tmad/logs_ggttggg_mad/log_ggttggg_mad_f_inl0_hrd0.txt b/epochX/cudacpp/tmad/logs_ggttggg_mad/log_ggttggg_mad_f_inl0_hrd0.txt
index 8e8ecf354b..c35aa0a017 100644
--- a/epochX/cudacpp/tmad/logs_ggttggg_mad/log_ggttggg_mad_f_inl0_hrd0.txt
+++ b/epochX/cudacpp/tmad/logs_ggttggg_mad/log_ggttggg_mad_f_inl0_hrd0.txt
@@ -3,8 +3,8 @@ CUDACPP_BUILDDIR='.'
 
 
 
-
 make USEBUILDDIR=1 AVX=none
+
 make USEBUILDDIR=1 AVX=sse4
 make USEBUILDDIR=1 AVX=avx2
 make USEBUILDDIR=1 AVX=512y
@@ -16,12 +16,12 @@ make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg'
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg'
 CUDACPP_BUILDDIR='build.none_f_inl0_hrd0'
-CUDACPP_BUILDDIR='build.avx2_f_inl0_hrd0'
-CUDACPP_BUILDDIR='build.512z_f_inl0_hrd0'
-CUDACPP_BUILDDIR='build.sse4_f_inl0_hrd0'
-CUDACPP_BUILDDIR='build.512y_f_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg'
+CUDACPP_BUILDDIR='build.sse4_f_inl0_hrd0'
+CUDACPP_BUILDDIR='build.avx2_f_inl0_hrd0'
+CUDACPP_BUILDDIR='build.512y_f_inl0_hrd0'
+CUDACPP_BUILDDIR='build.512z_f_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg'
 make[1]: Nothing to be done for 'all'.
@@ -33,7 +33,7 @@ make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/
 
 OMP_NUM_THREADS=
 
-DATE: 2023-11-24_17:21:33
+DATE: 2024-01-30_07:59:04
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 Working directory (run): /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg
@@ -57,11 +57,11 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggttggg_x1_fortran > /tmp/a
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 1.169e-06 [1.1693100945435802E-006] fbridge_mode=0
+ [XSECTION] Cross section = 1.169e-06 [1.1693100945435806E-006] fbridge_mode=0
  [UNWEIGHT] Wrote 1 events (found 166 events)
- [COUNTERS] PROGRAM TOTAL          :   98.7413s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.4669s
- [COUNTERS] Fortran MEs      ( 1 ) :   98.2745s for     8192 events => throughput is 8.34E+01 events/s
+ [COUNTERS] PROGRAM TOTAL          :  101.8466s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.4779s
+ [COUNTERS] Fortran MEs      ( 1 ) :  101.3687s for     8192 events => throughput is 8.08E+01 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x1 (create events.lhe) ***
 --------------------
@@ -82,11 +82,11 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggttggg_x1_fortran > /tmp/a
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 1.169e-06 [1.1693100945435802E-006] fbridge_mode=0
+ [XSECTION] Cross section = 1.169e-06 [1.1693100945435806E-006] fbridge_mode=0
  [UNWEIGHT] Wrote 15 events (found 163 events)
- [COUNTERS] PROGRAM TOTAL          :   98.9213s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.4778s
- [COUNTERS] Fortran MEs      ( 1 ) :   98.4435s for     8192 events => throughput is 8.32E+01 events/s
+ [COUNTERS] PROGRAM TOTAL          :  101.7818s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.4752s
+ [COUNTERS] Fortran MEs      ( 1 ) :  101.3066s for     8192 events => throughput is 8.09E+01 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x10 (create events.lhe) ***
 --------------------
@@ -107,11 +107,11 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggttggg_x10_fortran > /tmp/
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 2.136e-07 [2.1358436158813976E-007] fbridge_mode=0
+ [XSECTION] Cross section = 2.136e-07 [2.1358436158813979E-007] fbridge_mode=0
  [UNWEIGHT] Wrote 84 events (found 808 events)
- [COUNTERS] PROGRAM TOTAL          : 1080.6656s
- [COUNTERS] Fortran Overhead ( 0 ) :    4.3073s
- [COUNTERS] Fortran MEs      ( 1 ) : 1076.3584s for    90112 events => throughput is 8.37E+01 events/s
+ [COUNTERS] PROGRAM TOTAL          : 1118.6550s
+ [COUNTERS] Fortran Overhead ( 0 ) :    4.3767s
+ [COUNTERS] Fortran MEs      ( 1 ) : 1114.2783s for    90112 events => throughput is 8.09E+01 events/s
 
 *** (2-none) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -132,15 +132,15 @@ Executing ' ./build.none_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttggg_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 1.169e-06 [1.1694768344939596E-006] fbridge_mode=1
+ [XSECTION] Cross section = 1.169e-06 [1.1694768374083672E-006] fbridge_mode=1
  [UNWEIGHT] Wrote 15 events (found 163 events)
- [COUNTERS] PROGRAM TOTAL          :  202.3759s
- [COUNTERS] Fortran Overhead ( 0 ) :   92.2956s
- [COUNTERS] CudaCpp MEs      ( 2 ) :  110.0803s for     8192 events => throughput is 7.44E+01 events/s
+ [COUNTERS] PROGRAM TOTAL          :  205.5797s
+ [COUNTERS] Fortran Overhead ( 0 ) :   95.6059s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :  109.9738s for     8192 events => throughput is 7.45E+01 events/s
 
 *** (2-none) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.1693100945435802E-006) and cpp (1.1694768344939596E-006) differ by less than 4E-4 (0.00014259686216466783)
+OK! xsec from fortran (1.1693100945435806E-006) and cpp (1.1694768374083672E-006) differ by less than 4E-4 (0.00014259935458071915)
 
 *** (2-none) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -165,29 +165,29 @@ Executing ' ./build.none_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttggg_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 2.136e-07 [2.1361436150871156E-007] fbridge_mode=1
+ [XSECTION] Cross section = 2.136e-07 [2.1361435710758843E-007] fbridge_mode=1
  [UNWEIGHT] Wrote 84 events (found 808 events)
- [COUNTERS] PROGRAM TOTAL          : 1304.1851s
- [COUNTERS] Fortran Overhead ( 0 ) :   97.1400s
- [COUNTERS] CudaCpp MEs      ( 2 ) : 1207.0450s for    90112 events => throughput is 7.47E+01 events/s
+ [COUNTERS] PROGRAM TOTAL          : 1305.3326s
+ [COUNTERS] Fortran Overhead ( 0 ) :   98.5377s
+ [COUNTERS] CudaCpp MEs      ( 2 ) : 1206.7949s for    90112 events => throughput is 7.47E+01 events/s
 
 *** (2-none) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (2.1358436158813976E-007) and cpp (2.1361436150871156E-007) differ by less than 4E-4 (0.00014045934987350073)
+OK! xsec from fortran (2.1358436158813979E-007) and cpp (2.1361435710758843E-007) differ by less than 4E-4 (0.0001404387438554977)
 
 *** (2-none) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 8.937347e+01                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.692219e+01                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 8.901357e+01                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.699275e+01                 )  sec^-1
 
 *** (2-sse4) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -208,15 +208,15 @@ Executing ' ./build.sse4_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttggg_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 1.169e-06 [1.1694765850750953E-006] fbridge_mode=1
+ [XSECTION] Cross section = 1.169e-06 [1.1694765360831655E-006] fbridge_mode=1
  [UNWEIGHT] Wrote 15 events (found 163 events)
- [COUNTERS] PROGRAM TOTAL          :   50.8509s
- [COUNTERS] Fortran Overhead ( 0 ) :   24.0533s
- [COUNTERS] CudaCpp MEs      ( 2 ) :   26.7976s for     8192 events => throughput is 3.06E+02 events/s
+ [COUNTERS] PROGRAM TOTAL          :   52.0497s
+ [COUNTERS] Fortran Overhead ( 0 ) :   24.6638s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :   27.3859s for     8192 events => throughput is 2.99E+02 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.1693100945435802E-006) and cpp (1.1694765850750953E-006) differ by less than 4E-4 (0.00014238355787066226)
+OK! xsec from fortran (1.1693100945435806E-006) and cpp (1.1694765360831655E-006) differ by less than 4E-4 (0.00014234165972015766)
 
 *** (2-sse4) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -241,29 +241,29 @@ Executing ' ./build.sse4_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttggg_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 2.136e-07 [2.1361430669586527E-007] fbridge_mode=1
+ [XSECTION] Cross section = 2.136e-07 [2.1361429212586563E-007] fbridge_mode=1
  [UNWEIGHT] Wrote 84 events (found 808 events)
- [COUNTERS] PROGRAM TOTAL          :  321.9393s
- [COUNTERS] Fortran Overhead ( 0 ) :   27.9125s
- [COUNTERS] CudaCpp MEs      ( 2 ) :  294.0268s for    90112 events => throughput is 3.06E+02 events/s
+ [COUNTERS] PROGRAM TOTAL          :  336.4854s
+ [COUNTERS] Fortran Overhead ( 0 ) :   29.3396s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :  307.1459s for    90112 events => throughput is 2.93E+02 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (2.1358436158813976E-007) and cpp (2.1361430669586527E-007) differ by less than 4E-4 (0.00014020271663550687)
+OK! xsec from fortran (2.1358436158813979E-007) and cpp (2.1361429212586563E-007) differ by less than 4E-4 (0.00014013450003202976)
 
 *** (2-sse4) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.520695e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.371429e+02                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.523057e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.391230e+02                 )  sec^-1
 
 *** (2-avx2) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -284,15 +284,15 @@ Executing ' ./build.avx2_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttggg_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 1.169e-06 [1.1694764951124567E-006] fbridge_mode=1
+ [XSECTION] Cross section = 1.169e-06 [1.1694764906356561E-006] fbridge_mode=1
  [UNWEIGHT] Wrote 15 events (found 163 events)
- [COUNTERS] PROGRAM TOTAL          :   26.0191s
- [COUNTERS] Fortran Overhead ( 0 ) :   12.1224s
- [COUNTERS] CudaCpp MEs      ( 2 ) :   13.8967s for     8192 events => throughput is 5.89E+02 events/s
+ [COUNTERS] PROGRAM TOTAL          :   27.0721s
+ [COUNTERS] Fortran Overhead ( 0 ) :   12.6637s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :   14.4085s for     8192 events => throughput is 5.69E+02 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.1693100945435802E-006) and cpp (1.1694764951124567E-006) differ by less than 4E-4 (0.00014230662135994443)
+OK! xsec from fortran (1.1693100945435806E-006) and cpp (1.1694764906356561E-006) differ by less than 4E-4 (0.0001423027927767162)
 
 *** (2-avx2) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -317,29 +317,29 @@ Executing ' ./build.avx2_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttggg_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 2.136e-07 [2.1361430425531218E-007] fbridge_mode=1
+ [XSECTION] Cross section = 2.136e-07 [2.1361429111797059E-007] fbridge_mode=1
  [UNWEIGHT] Wrote 84 events (found 808 events)
- [COUNTERS] PROGRAM TOTAL          :  171.5242s
- [COUNTERS] Fortran Overhead ( 0 ) :   15.8508s
- [COUNTERS] CudaCpp MEs      ( 2 ) :  155.6733s for    90112 events => throughput is 5.79E+02 events/s
+ [COUNTERS] PROGRAM TOTAL          :  174.3265s
+ [COUNTERS] Fortran Overhead ( 0 ) :   16.5080s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :  157.8185s for    90112 events => throughput is 5.71E+02 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (2.1358436158813976E-007) and cpp (2.1361430425531218E-007) differ by less than 4E-4 (0.0001401912899885449)
+OK! xsec from fortran (2.1358436158813979E-007) and cpp (2.1361429111797059E-007) differ by less than 4E-4 (0.00014012978107680318)
 
 *** (2-avx2) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 7.026908e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.735262e+02                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 7.019796e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.739370e+02                 )  sec^-1
 
 *** (2-512y) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -360,15 +360,15 @@ Executing ' ./build.512y_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttggg_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 1.169e-06 [1.1694764951124567E-006] fbridge_mode=1
+ [XSECTION] Cross section = 1.169e-06 [1.1694764906356561E-006] fbridge_mode=1
  [UNWEIGHT] Wrote 15 events (found 163 events)
- [COUNTERS] PROGRAM TOTAL          :   23.1132s
- [COUNTERS] Fortran Overhead ( 0 ) :   10.7368s
- [COUNTERS] CudaCpp MEs      ( 2 ) :   12.3764s for     8192 events => throughput is 6.62E+02 events/s
+ [COUNTERS] PROGRAM TOTAL          :   23.8611s
+ [COUNTERS] Fortran Overhead ( 0 ) :   11.1738s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :   12.6873s for     8192 events => throughput is 6.46E+02 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.1693100945435802E-006) and cpp (1.1694764951124567E-006) differ by less than 4E-4 (0.00014230662135994443)
+OK! xsec from fortran (1.1693100945435806E-006) and cpp (1.1694764906356561E-006) differ by less than 4E-4 (0.0001423027927767162)
 
 *** (2-512y) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -393,29 +393,29 @@ Executing ' ./build.512y_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttggg_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 2.136e-07 [2.1361430425531218E-007] fbridge_mode=1
+ [XSECTION] Cross section = 2.136e-07 [2.1361429111797059E-007] fbridge_mode=1
  [UNWEIGHT] Wrote 84 events (found 808 events)
- [COUNTERS] PROGRAM TOTAL          :  151.0249s
- [COUNTERS] Fortran Overhead ( 0 ) :   14.5293s
- [COUNTERS] CudaCpp MEs      ( 2 ) :  136.4957s for    90112 events => throughput is 6.60E+02 events/s
+ [COUNTERS] PROGRAM TOTAL          :  154.4819s
+ [COUNTERS] Fortran Overhead ( 0 ) :   15.1638s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :  139.3180s for    90112 events => throughput is 6.47E+02 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (2.1358436158813976E-007) and cpp (2.1361430425531218E-007) differ by less than 4E-4 (0.0001401912899885449)
+OK! xsec from fortran (2.1358436158813979E-007) and cpp (2.1361429111797059E-007) differ by less than 4E-4 (0.00014012978107680318)
 
 *** (2-512y) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 7.945001e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.672552e+02                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 7.958337e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.678478e+02                 )  sec^-1
 
 *** (2-512z) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -436,15 +436,15 @@ Executing ' ./build.512z_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttggg_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 1.169e-06 [1.1694767957195604E-006] fbridge_mode=1
+ [XSECTION] Cross section = 1.169e-06 [1.1694768276769753E-006] fbridge_mode=1
  [UNWEIGHT] Wrote 15 events (found 163 events)
- [COUNTERS] PROGRAM TOTAL          :   23.9825s
- [COUNTERS] Fortran Overhead ( 0 ) :   11.9109s
- [COUNTERS] CudaCpp MEs      ( 2 ) :   12.0717s for     8192 events => throughput is 6.79E+02 events/s
+ [COUNTERS] PROGRAM TOTAL          :   25.4092s
+ [COUNTERS] Fortran Overhead ( 0 ) :   12.6834s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :   12.7257s for     8192 events => throughput is 6.44E+02 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.1693100945435802E-006) and cpp (1.1694767957195604E-006) differ by less than 4E-4 (0.00014256370209930758)
+OK! xsec from fortran (1.1693100945435806E-006) and cpp (1.1694768276769753E-006) differ by less than 4E-4 (0.00014259103224434355)
 
 *** (2-512z) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -469,29 +469,29 @@ Executing ' ./build.512z_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttggg_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 2.136e-07 [2.1361435956349820E-007] fbridge_mode=1
+ [XSECTION] Cross section = 2.136e-07 [2.1361435948756818E-007] fbridge_mode=1
  [UNWEIGHT] Wrote 84 events (found 808 events)
- [COUNTERS] PROGRAM TOTAL          :  148.9306s
- [COUNTERS] Fortran Overhead ( 0 ) :   15.5819s
- [COUNTERS] CudaCpp MEs      ( 2 ) :  133.3487s for    90112 events => throughput is 6.76E+02 events/s
+ [COUNTERS] PROGRAM TOTAL          :  156.3015s
+ [COUNTERS] Fortran Overhead ( 0 ) :   16.4754s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :  139.8261s for    90112 events => throughput is 6.44E+02 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (2.1358436158813976E-007) and cpp (2.1361435956349820E-007) differ by less than 4E-4 (0.00014045024240250115)
+OK! xsec from fortran (2.1358436158813979E-007) and cpp (2.1361435948756818E-007) differ by less than 4E-4 (0.00014044988689865257)
 
 *** (2-512z) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 7.105766e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.776081e+02                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 7.113491e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.750726e+02                 )  sec^-1
 
 *** (3) EXECUTE MADEVENT_CUDA x1 (create events.lhe) ***
 --------------------
@@ -512,15 +512,15 @@ Executing ' ./build.none_f_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_ggttggg
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 1.169e-06 [1.1694770708195000E-006] fbridge_mode=1
+ [XSECTION] Cross section = 1.169e-06 [1.1694770708194997E-006] fbridge_mode=1
  [UNWEIGHT] Wrote 15 events (found 163 events)
- [COUNTERS] PROGRAM TOTAL          :    2.4884s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.9881s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.5003s for     8192 events => throughput is 1.64E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.5003s
+ [COUNTERS] Fortran Overhead ( 0 ) :    2.0019s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.4984s for     8192 events => throughput is 1.64E+04 events/s
 
 *** (3) Compare MADEVENT_CUDA x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.1693100945435802E-006) and cpp (1.1694770708195000E-006) differ by less than 4E-4 (0.00014279896898083955)
+OK! xsec from fortran (1.1693100945435806E-006) and cpp (1.1694770708194997E-006) differ by less than 4E-4 (0.00014279896898039546)
 
 *** (3) Compare MADEVENT_CUDA x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -545,58 +545,58 @@ Executing ' ./build.none_f_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_ggttggg
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 2.136e-07 [2.1361443477565659E-007] fbridge_mode=1
+ [XSECTION] Cross section = 2.136e-07 [2.1361443477565656E-007] fbridge_mode=1
  [UNWEIGHT] Wrote 84 events (found 808 events)
- [COUNTERS] PROGRAM TOTAL          :   11.1878s
- [COUNTERS] Fortran Overhead ( 0 ) :    5.7450s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    5.4428s for    90112 events => throughput is 1.66E+04 events/s
+ [COUNTERS] PROGRAM TOTAL          :   11.2881s
+ [COUNTERS] Fortran Overhead ( 0 ) :    5.8695s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    5.4186s for    90112 events => throughput is 1.66E+04 events/s
 
 *** (3) Compare MADEVENT_CUDA x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (2.1358436158813976E-007) and cpp (2.1361443477565659E-007) differ by less than 4E-4 (0.0001408023850304474)
+OK! xsec from fortran (2.1358436158813979E-007) and cpp (2.1361443477565656E-007) differ by less than 4E-4 (0.00014080238503022535)
 
 *** (3) Compare MADEVENT_CUDA x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cuda.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.635622e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.635547e+04                 )  sec^-1
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.625946e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.633264e+04                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 512 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.339621e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.309560e+04                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 512 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.381945e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.405493e+04                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 128 128 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.320446e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.341562e+04                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 128 128 1 ***
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.325431e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.341458e+04                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 2048 8 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.300828e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.336833e+04                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 2048 8 1 ***
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 6.393528e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.413620e+03                 )  sec^-1
 
 TEST COMPLETED
diff --git a/epochX/cudacpp/tmad/logs_ggttggg_mad/log_ggttggg_mad_m_inl0_hrd0.txt b/epochX/cudacpp/tmad/logs_ggttggg_mad/log_ggttggg_mad_m_inl0_hrd0.txt
index 963e0ec416..b9faa14c51 100644
--- a/epochX/cudacpp/tmad/logs_ggttggg_mad/log_ggttggg_mad_m_inl0_hrd0.txt
+++ b/epochX/cudacpp/tmad/logs_ggttggg_mad/log_ggttggg_mad_m_inl0_hrd0.txt
@@ -4,8 +4,8 @@ CUDACPP_BUILDDIR='.'
 
 
 make USEBUILDDIR=1 AVX=none
-
 make USEBUILDDIR=1 AVX=sse4
+
 make USEBUILDDIR=1 AVX=avx2
 make USEBUILDDIR=1 AVX=512y
 
@@ -15,10 +15,11 @@ make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg'
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg'
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg'
-CUDACPP_BUILDDIR='build.avx2_m_inl0_hrd0'
-CUDACPP_BUILDDIR='build.512y_m_inl0_hrd0'
-CUDACPP_BUILDDIR='build.none_m_inl0_hrd0'
 CUDACPP_BUILDDIR='build.512z_m_inl0_hrd0'
+CUDACPP_BUILDDIR='build.sse4_m_inl0_hrd0'
+CUDACPP_BUILDDIR='build.none_m_inl0_hrd0'
+CUDACPP_BUILDDIR='build.512y_m_inl0_hrd0'
+CUDACPP_BUILDDIR='build.avx2_m_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg'
 make[1]: Nothing to be done for 'all'.
@@ -27,13 +28,12 @@ make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg'
-CUDACPP_BUILDDIR='build.sse4_m_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg'
 
 OMP_NUM_THREADS=
 
-DATE: 2023-11-24_18:28:54
+DATE: 2024-01-30_09:07:55
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 Working directory (run): /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg
@@ -57,11 +57,11 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggttggg_x1_fortran > /tmp/a
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 1.169e-06 [1.1693100945435802E-006] fbridge_mode=0
+ [XSECTION] Cross section = 1.169e-06 [1.1693100945435806E-006] fbridge_mode=0
  [UNWEIGHT] Wrote 1 events (found 166 events)
- [COUNTERS] PROGRAM TOTAL          :   98.6622s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.4669s
- [COUNTERS] Fortran MEs      ( 1 ) :   98.1953s for     8192 events => throughput is 8.34E+01 events/s
+ [COUNTERS] PROGRAM TOTAL          :  101.9697s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.4771s
+ [COUNTERS] Fortran MEs      ( 1 ) :  101.4926s for     8192 events => throughput is 8.07E+01 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x1 (create events.lhe) ***
 --------------------
@@ -82,11 +82,11 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggttggg_x1_fortran > /tmp/a
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 1.169e-06 [1.1693100945435802E-006] fbridge_mode=0
+ [XSECTION] Cross section = 1.169e-06 [1.1693100945435806E-006] fbridge_mode=0
  [UNWEIGHT] Wrote 15 events (found 163 events)
- [COUNTERS] PROGRAM TOTAL          :   98.6106s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.4686s
- [COUNTERS] Fortran MEs      ( 1 ) :   98.1420s for     8192 events => throughput is 8.35E+01 events/s
+ [COUNTERS] PROGRAM TOTAL          :  101.6914s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.4764s
+ [COUNTERS] Fortran MEs      ( 1 ) :  101.2150s for     8192 events => throughput is 8.09E+01 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x10 (create events.lhe) ***
 --------------------
@@ -107,11 +107,11 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_ggttggg_x10_fortran > /tmp/
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 2.136e-07 [2.1358436158813976E-007] fbridge_mode=0
+ [XSECTION] Cross section = 2.136e-07 [2.1358436158813979E-007] fbridge_mode=0
  [UNWEIGHT] Wrote 84 events (found 808 events)
- [COUNTERS] PROGRAM TOTAL          : 1080.6388s
- [COUNTERS] Fortran Overhead ( 0 ) :    4.2788s
- [COUNTERS] Fortran MEs      ( 1 ) : 1076.3600s for    90112 events => throughput is 8.37E+01 events/s
+ [COUNTERS] PROGRAM TOTAL          : 1118.2550s
+ [COUNTERS] Fortran Overhead ( 0 ) :    4.3831s
+ [COUNTERS] Fortran MEs      ( 1 ) : 1113.8719s for    90112 events => throughput is 8.09E+01 events/s
 
 *** (2-none) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -132,15 +132,15 @@ Executing ' ./build.none_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttggg_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 1.169e-06 [1.1693101016896846E-006] fbridge_mode=1
+ [XSECTION] Cross section = 1.169e-06 [1.1693101016896844E-006] fbridge_mode=1
  [UNWEIGHT] Wrote 15 events (found 163 events)
- [COUNTERS] PROGRAM TOTAL          :  227.6225s
- [COUNTERS] Fortran Overhead ( 0 ) :  105.5451s
- [COUNTERS] CudaCpp MEs      ( 2 ) :  122.0773s for     8192 events => throughput is 6.71E+01 events/s
+ [COUNTERS] PROGRAM TOTAL          :  224.2502s
+ [COUNTERS] Fortran Overhead ( 0 ) :  103.5045s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :  120.7457s for     8192 events => throughput is 6.78E+01 events/s
 
 *** (2-none) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.1693100945435802E-006) and cpp (1.1693101016896846E-006) differ by less than 2E-4 (6.111385175699979e-09)
+OK! xsec from fortran (1.1693100945435806E-006) and cpp (1.1693101016896844E-006) differ by less than 2E-4 (6.1113847316107694e-09)
 
 *** (2-none) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -167,27 +167,27 @@ Executing ' ./build.none_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttggg_
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 2.136e-07 [2.1358436275882778E-007] fbridge_mode=1
  [UNWEIGHT] Wrote 84 events (found 808 events)
- [COUNTERS] PROGRAM TOTAL          : 1457.2560s
- [COUNTERS] Fortran Overhead ( 0 ) :  110.3750s
- [COUNTERS] CudaCpp MEs      ( 2 ) : 1346.8810s for    90112 events => throughput is 6.69E+01 events/s
+ [COUNTERS] PROGRAM TOTAL          : 1439.9019s
+ [COUNTERS] Fortran Overhead ( 0 ) :  107.2148s
+ [COUNTERS] CudaCpp MEs      ( 2 ) : 1332.6870s for    90112 events => throughput is 6.76E+01 events/s
 
 *** (2-none) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (2.1358436158813976E-007) and cpp (2.1358436275882778E-007) differ by less than 2E-4 (5.48115042242614e-09)
+OK! xsec from fortran (2.1358436158813979E-007) and cpp (2.1358436275882778E-007) differ by less than 2E-4 (5.48115042242614e-09)
 
 *** (2-none) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+BRDHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 7.788273e+01                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.977492e+01                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 7.741929e+01                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.962894e+01                 )  sec^-1
 
 *** (2-sse4) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -210,13 +210,13 @@ Executing ' ./build.sse4_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttggg_
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 1.169e-06 [1.1693101020910778E-006] fbridge_mode=1
  [UNWEIGHT] Wrote 15 events (found 163 events)
- [COUNTERS] PROGRAM TOTAL          :  115.6749s
- [COUNTERS] Fortran Overhead ( 0 ) :   52.7434s
- [COUNTERS] CudaCpp MEs      ( 2 ) :   62.9314s for     8192 events => throughput is 1.30E+02 events/s
+ [COUNTERS] PROGRAM TOTAL          :  114.2875s
+ [COUNTERS] Fortran Overhead ( 0 ) :   54.1504s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :   60.1370s for     8192 events => throughput is 1.36E+02 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.1693100945435802E-006) and cpp (1.1693101020910778E-006) differ by less than 2E-4 (6.454658807442115e-09)
+OK! xsec from fortran (1.1693100945435806E-006) and cpp (1.1693101020910778E-006) differ by less than 2E-4 (6.454658363352905e-09)
 
 *** (2-sse4) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -241,29 +241,29 @@ Executing ' ./build.sse4_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttggg_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 2.136e-07 [2.1358436284111598E-007] fbridge_mode=1
+ [XSECTION] Cross section = 2.136e-07 [2.1358436284111587E-007] fbridge_mode=1
  [UNWEIGHT] Wrote 84 events (found 808 events)
- [COUNTERS] PROGRAM TOTAL          :  738.7127s
- [COUNTERS] Fortran Overhead ( 0 ) :   56.7745s
- [COUNTERS] CudaCpp MEs      ( 2 ) :  681.9382s for    90112 events => throughput is 1.32E+02 events/s
+ [COUNTERS] PROGRAM TOTAL          :  713.5498s
+ [COUNTERS] Fortran Overhead ( 0 ) :   58.1042s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :  655.4456s for    90112 events => throughput is 1.37E+02 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (2.1358436158813976E-007) and cpp (2.1358436284111598E-007) differ by less than 2E-4 (5.866422903011426e-09)
+OK! xsec from fortran (2.1358436158813979E-007) and cpp (2.1358436284111587E-007) differ by less than 2E-4 (5.866422458922216e-09)
 
 *** (2-sse4) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+BRDHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.576187e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.524488e+02                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.573070e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.529646e+02                 )  sec^-1
 
 *** (2-avx2) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -284,15 +284,15 @@ Executing ' ./build.avx2_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttggg_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 1.169e-06 [1.1693101021831071E-006] fbridge_mode=1
+ [XSECTION] Cross section = 1.169e-06 [1.1693101021831069E-006] fbridge_mode=1
  [UNWEIGHT] Wrote 15 events (found 163 events)
- [COUNTERS] PROGRAM TOTAL          :   49.9825s
- [COUNTERS] Fortran Overhead ( 0 ) :   22.7532s
- [COUNTERS] CudaCpp MEs      ( 2 ) :   27.2294s for     8192 events => throughput is 3.01E+02 events/s
+ [COUNTERS] PROGRAM TOTAL          :   50.7719s
+ [COUNTERS] Fortran Overhead ( 0 ) :   23.4236s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :   27.3483s for     8192 events => throughput is 3.00E+02 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.1693100945435802E-006) and cpp (1.1693101021831071E-006) differ by less than 2E-4 (6.5333627397023974e-09)
+OK! xsec from fortran (1.1693100945435806E-006) and cpp (1.1693101021831069E-006) differ by less than 2E-4 (6.533362073568583e-09)
 
 *** (2-avx2) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -317,29 +317,29 @@ Executing ' ./build.avx2_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttggg_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 2.136e-07 [2.1358436281462142E-007] fbridge_mode=1
+ [XSECTION] Cross section = 2.136e-07 [2.1358436281462147E-007] fbridge_mode=1
  [UNWEIGHT] Wrote 84 events (found 808 events)
- [COUNTERS] PROGRAM TOTAL          :  330.1981s
- [COUNTERS] Fortran Overhead ( 0 ) :   26.8024s
- [COUNTERS] CudaCpp MEs      ( 2 ) :  303.3957s for    90112 events => throughput is 2.97E+02 events/s
+ [COUNTERS] PROGRAM TOTAL          :  326.3386s
+ [COUNTERS] Fortran Overhead ( 0 ) :   27.1544s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :  299.1842s for    90112 events => throughput is 3.01E+02 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (2.1358436158813976E-007) and cpp (2.1358436281462142E-007) differ by less than 2E-4 (5.742375686068613e-09)
+OK! xsec from fortran (2.1358436158813979E-007) and cpp (2.1358436281462147E-007) differ by less than 2E-4 (5.7423759081132175e-09)
 
 *** (2-avx2) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+BRDHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.700163e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.567666e+02                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.716172e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.597479e+02                 )  sec^-1
 
 *** (2-512y) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -360,15 +360,15 @@ Executing ' ./build.512y_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttggg_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 1.169e-06 [1.1693101021831071E-006] fbridge_mode=1
+ [XSECTION] Cross section = 1.169e-06 [1.1693101021831069E-006] fbridge_mode=1
  [UNWEIGHT] Wrote 15 events (found 163 events)
- [COUNTERS] PROGRAM TOTAL          :   43.9707s
- [COUNTERS] Fortran Overhead ( 0 ) :   19.9051s
- [COUNTERS] CudaCpp MEs      ( 2 ) :   24.0656s for     8192 events => throughput is 3.40E+02 events/s
+ [COUNTERS] PROGRAM TOTAL          :   45.3921s
+ [COUNTERS] Fortran Overhead ( 0 ) :   20.5216s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :   24.8705s for     8192 events => throughput is 3.29E+02 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.1693100945435802E-006) and cpp (1.1693101021831071E-006) differ by less than 2E-4 (6.5333627397023974e-09)
+OK! xsec from fortran (1.1693100945435806E-006) and cpp (1.1693101021831069E-006) differ by less than 2E-4 (6.533362073568583e-09)
 
 *** (2-512y) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -393,29 +393,29 @@ Executing ' ./build.512y_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttggg_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 2.136e-07 [2.1358436281462142E-007] fbridge_mode=1
+ [XSECTION] Cross section = 2.136e-07 [2.1358436281462147E-007] fbridge_mode=1
  [UNWEIGHT] Wrote 84 events (found 808 events)
- [COUNTERS] PROGRAM TOTAL          :  289.2968s
- [COUNTERS] Fortran Overhead ( 0 ) :   23.7412s
- [COUNTERS] CudaCpp MEs      ( 2 ) :  265.5556s for    90112 events => throughput is 3.39E+02 events/s
+ [COUNTERS] PROGRAM TOTAL          :  301.9509s
+ [COUNTERS] Fortran Overhead ( 0 ) :   24.4062s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :  277.5446s for    90112 events => throughput is 3.25E+02 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (2.1358436158813976E-007) and cpp (2.1358436281462142E-007) differ by less than 2E-4 (5.742375686068613e-09)
+OK! xsec from fortran (2.1358436158813979E-007) and cpp (2.1358436281462147E-007) differ by less than 2E-4 (5.7423759081132175e-09)
 
 *** (2-512y) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+BRDHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 4.209510e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.101718e+02                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 4.226810e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.081457e+02                 )  sec^-1
 
 *** (2-512z) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -436,15 +436,15 @@ Executing ' ./build.512z_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttggg_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 1.169e-06 [1.1693101021831071E-006] fbridge_mode=1
+ [XSECTION] Cross section = 1.169e-06 [1.1693101021831069E-006] fbridge_mode=1
  [UNWEIGHT] Wrote 15 events (found 163 events)
- [COUNTERS] PROGRAM TOTAL          :   46.6653s
- [COUNTERS] Fortran Overhead ( 0 ) :   22.7630s
- [COUNTERS] CudaCpp MEs      ( 2 ) :   23.9023s for     8192 events => throughput is 3.43E+02 events/s
+ [COUNTERS] PROGRAM TOTAL          :   48.9503s
+ [COUNTERS] Fortran Overhead ( 0 ) :   23.9070s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :   25.0433s for     8192 events => throughput is 3.27E+02 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.1693100945435802E-006) and cpp (1.1693101021831071E-006) differ by less than 2E-4 (6.5333627397023974e-09)
+OK! xsec from fortran (1.1693100945435806E-006) and cpp (1.1693101021831069E-006) differ by less than 2E-4 (6.533362073568583e-09)
 
 *** (2-512z) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -469,29 +469,29 @@ Executing ' ./build.512z_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_ggttggg_
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 2.136e-07 [2.1358436281462142E-007] fbridge_mode=1
+ [XSECTION] Cross section = 2.136e-07 [2.1358436281462147E-007] fbridge_mode=1
  [UNWEIGHT] Wrote 84 events (found 808 events)
- [COUNTERS] PROGRAM TOTAL          :  296.3640s
- [COUNTERS] Fortran Overhead ( 0 ) :   26.8877s
- [COUNTERS] CudaCpp MEs      ( 2 ) :  269.4763s for    90112 events => throughput is 3.34E+02 events/s
+ [COUNTERS] PROGRAM TOTAL          :  301.8849s
+ [COUNTERS] Fortran Overhead ( 0 ) :   27.8392s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :  274.0457s for    90112 events => throughput is 3.29E+02 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (2.1358436158813976E-007) and cpp (2.1358436281462142E-007) differ by less than 2E-4 (5.742375686068613e-09)
+OK! xsec from fortran (2.1358436158813979E-007) and cpp (2.1358436281462147E-007) differ by less than 2E-4 (5.7423759081132175e-09)
 
 *** (2-512z) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+BRDHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.629406e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.501339e+02                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.640208e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.509355e+02                 )  sec^-1
 
 *** (3) EXECUTE MADEVENT_CUDA x1 (create events.lhe) ***
 --------------------
@@ -514,13 +514,13 @@ Executing ' ./build.none_m_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_ggttggg
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 1.169e-06 [1.1693100942770687E-006] fbridge_mode=1
  [UNWEIGHT] Wrote 15 events (found 163 events)
- [COUNTERS] PROGRAM TOTAL          :    3.5613s
- [COUNTERS] Fortran Overhead ( 0 ) :    2.6950s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.8663s for     8192 events => throughput is 9.46E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :    3.5767s
+ [COUNTERS] Fortran Overhead ( 0 ) :    2.7138s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.8630s for     8192 events => throughput is 9.49E+03 events/s
 
 *** (3) Compare MADEVENT_CUDA x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (1.1693100945435802E-006) and cpp (1.1693100942770687E-006) differ by less than 2E-4 (2.2792201459509442e-10)
+OK! xsec from fortran (1.1693100945435806E-006) and cpp (1.1693100942770687E-006) differ by less than 2E-4 (2.279223476620018e-10)
 
 *** (3) Compare MADEVENT_CUDA x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -547,56 +547,56 @@ Executing ' ./build.none_m_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_ggttggg
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 2.136e-07 [2.1358436157495368E-007] fbridge_mode=1
  [UNWEIGHT] Wrote 84 events (found 808 events)
- [COUNTERS] PROGRAM TOTAL          :   15.9559s
- [COUNTERS] Fortran Overhead ( 0 ) :    6.4625s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    9.4934s for    90112 events => throughput is 9.49E+03 events/s
+ [COUNTERS] PROGRAM TOTAL          :   16.0811s
+ [COUNTERS] Fortran Overhead ( 0 ) :    6.5876s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    9.4935s for    90112 events => throughput is 9.49E+03 events/s
 
 *** (3) Compare MADEVENT_CUDA x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (2.1358436158813976E-007) and cpp (2.1358436157495368E-007) differ by less than 2E-4 (6.173705990875078e-11)
+OK! xsec from fortran (2.1358436158813979E-007) and cpp (2.1358436157495368E-007) differ by less than 2E-4 (6.173717093105324e-11)
 
 *** (3) Compare MADEVENT_CUDA x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cuda.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 9.436689e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.427839e+03                 )  sec^-1
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.080384e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.087147e+04                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 512 32 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.108886e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.109838e+04                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 512 32 1 ***
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.161919e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.157967e+04                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 128 128 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.113576e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.106789e+04                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 128 128 1 ***
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.109641e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.114403e+04                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 2048 8 1 --bridge ***
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.111246e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.111816e+04                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 2048 8 1 ***
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.639262e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.650481e+03                 )  sec^-1
 
 TEST COMPLETED
diff --git a/epochX/cudacpp/tmad/logs_gqttq_mad/log_gqttq_mad_d_inl0_hrd0.txt b/epochX/cudacpp/tmad/logs_gqttq_mad/log_gqttq_mad_d_inl0_hrd0.txt
index f27ee93a59..1fb13570ed 100644
--- a/epochX/cudacpp/tmad/logs_gqttq_mad/log_gqttq_mad_d_inl0_hrd0.txt
+++ b/epochX/cudacpp/tmad/logs_gqttq_mad/log_gqttq_mad_d_inl0_hrd0.txt
@@ -3,9 +3,9 @@ CUDACPP_BUILDDIR='.'
 
 
 make USEBUILDDIR=1 AVX=none
-
 make USEBUILDDIR=1 AVX=sse4
 
+
 make USEBUILDDIR=1 AVX=avx2
 make USEBUILDDIR=1 AVX=512y
 
@@ -15,13 +15,13 @@ make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu'
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu'
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu'
-CUDACPP_BUILDDIR='build.512y_d_inl0_hrd0'
 CUDACPP_BUILDDIR='build.sse4_d_inl0_hrd0'
+CUDACPP_BUILDDIR='build.none_d_inl0_hrd0'
 CUDACPP_BUILDDIR='build.512z_d_inl0_hrd0'
-CUDACPP_BUILDDIR='build.avx2_d_inl0_hrd0'
+CUDACPP_BUILDDIR='build.512y_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu'
-CUDACPP_BUILDDIR='build.none_d_inl0_hrd0'
+CUDACPP_BUILDDIR='build.avx2_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu'
 make[1]: Nothing to be done for 'all'.
@@ -33,7 +33,7 @@ make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/
 
 OMP_NUM_THREADS=
 
-DATE: 2023-11-24_15:51:16
+DATE: 2024-01-30_06:26:25
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 Working directory (run): /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu
@@ -59,9 +59,9 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_gqttq_x1_fortran > /tmp/ava
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.2605 [0.26050333309703716] fbridge_mode=0
  [UNWEIGHT] Wrote 78 events (found 561 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3170s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2461s
- [COUNTERS] Fortran MEs      ( 1 ) :    0.0708s for     8192 events => throughput is 1.16E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3322s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2581s
+ [COUNTERS] Fortran MEs      ( 1 ) :    0.0741s for     8192 events => throughput is 1.11E+05 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x1 (create events.lhe) ***
 --------------------
@@ -84,9 +84,9 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_gqttq_x1_fortran > /tmp/ava
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.2605 [0.26050333309703716] fbridge_mode=0
  [UNWEIGHT] Wrote 81 events (found 540 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3076s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2367s
- [COUNTERS] Fortran MEs      ( 1 ) :    0.0710s for     8192 events => throughput is 1.15E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3239s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2500s
+ [COUNTERS] Fortran MEs      ( 1 ) :    0.0739s for     8192 events => throughput is 1.11E+05 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x10 (create events.lhe) ***
 --------------------
@@ -107,11 +107,11 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_gqttq_x10_fortran > /tmp/av
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.218 [0.21801182648615874] fbridge_mode=0
+ [XSECTION] Cross section = 0.218 [0.21801182648615872] fbridge_mode=0
  [UNWEIGHT] Wrote 853 events (found 1849 events)
- [COUNTERS] PROGRAM TOTAL          :    2.2394s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.4635s
- [COUNTERS] Fortran MEs      ( 1 ) :    0.7759s for    90112 events => throughput is 1.16E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.4169s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.6004s
+ [COUNTERS] Fortran MEs      ( 1 ) :    0.8166s for    90112 events => throughput is 1.10E+05 events/s
 
 *** (2-none) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -132,15 +132,15 @@ Executing ' ./build.none_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_gqttq_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.2605 [0.26050333309703716] fbridge_mode=1
+ [XSECTION] Cross section = 0.2605 [0.26050333309703710] fbridge_mode=1
  [UNWEIGHT] Wrote 81 events (found 540 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3933s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3167s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0766s for     8192 events => throughput is 1.07E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.4205s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3380s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0825s for     8192 events => throughput is 9.92E+04 events/s
 
 *** (2-none) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.26050333309703716) and cpp (0.26050333309703716) differ by less than 2E-14 (0.0)
+OK! xsec from fortran (0.26050333309703716) and cpp (0.26050333309703710) differ by less than 2E-14 (2.220446049250313e-16)
 
 *** (2-none) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -165,29 +165,29 @@ Executing ' ./build.none_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_gqttq_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.218 [0.21801182648615863] fbridge_mode=1
+ [XSECTION] Cross section = 0.218 [0.21801182648615872] fbridge_mode=1
  [UNWEIGHT] Wrote 853 events (found 1849 events)
- [COUNTERS] PROGRAM TOTAL          :    2.4082s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.5618s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.8464s for    90112 events => throughput is 1.06E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.6157s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.7094s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.9063s for    90112 events => throughput is 9.94E+04 events/s
 
 *** (2-none) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.21801182648615874) and cpp (0.21801182648615863) differ by less than 2E-14 (5.551115123125783e-16)
+OK! xsec from fortran (0.21801182648615872) and cpp (0.21801182648615872) differ by less than 2E-14 (0.0)
 
 *** (2-none) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.081037e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.011343e+05                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.086206e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.009603e+05                 )  sec^-1
 
 *** (2-sse4) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -208,15 +208,15 @@ Executing ' ./build.sse4_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_gqttq_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.2605 [0.26050333309703733] fbridge_mode=1
+ [XSECTION] Cross section = 0.2605 [0.26050333309703727] fbridge_mode=1
  [UNWEIGHT] Wrote 81 events (found 540 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3295s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2869s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0426s for     8192 events => throughput is 1.92E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3388s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2965s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0423s for     8192 events => throughput is 1.94E+05 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.26050333309703716) and cpp (0.26050333309703733) differ by less than 2E-14 (6.661338147750939e-16)
+OK! xsec from fortran (0.26050333309703716) and cpp (0.26050333309703727) differ by less than 2E-14 (4.440892098500626e-16)
 
 *** (2-sse4) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -243,27 +243,27 @@ Executing ' ./build.sse4_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_gqttq_x1
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.218 [0.21801182648615872] fbridge_mode=1
  [UNWEIGHT] Wrote 853 events (found 1849 events)
- [COUNTERS] PROGRAM TOTAL          :    2.0022s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.5391s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.4630s for    90112 events => throughput is 1.95E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.1314s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.6642s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.4672s for    90112 events => throughput is 1.93E+05 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.21801182648615874) and cpp (0.21801182648615872) differ by less than 2E-14 (1.1102230246251565e-16)
+OK! xsec from fortran (0.21801182648615872) and cpp (0.21801182648615872) differ by less than 2E-14 (0.0)
 
 *** (2-sse4) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.967710e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.951695e+05                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.951729e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.960178e+05                 )  sec^-1
 
 *** (2-avx2) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -284,15 +284,15 @@ Executing ' ./build.avx2_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_gqttq_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.2605 [0.26050333309703733] fbridge_mode=1
+ [XSECTION] Cross section = 0.2605 [0.26050333309703727] fbridge_mode=1
  [UNWEIGHT] Wrote 81 events (found 540 events)
- [COUNTERS] PROGRAM TOTAL          :    0.2892s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2650s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0242s for     8192 events => throughput is 3.38E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3038s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2788s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0250s for     8192 events => throughput is 3.28E+05 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.26050333309703716) and cpp (0.26050333309703733) differ by less than 2E-14 (6.661338147750939e-16)
+OK! xsec from fortran (0.26050333309703716) and cpp (0.26050333309703727) differ by less than 2E-14 (4.440892098500626e-16)
 
 *** (2-avx2) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -317,29 +317,29 @@ Executing ' ./build.avx2_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_gqttq_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.218 [0.21801182648615863] fbridge_mode=1
+ [XSECTION] Cross section = 0.218 [0.21801182648615869] fbridge_mode=1
  [UNWEIGHT] Wrote 853 events (found 1849 events)
- [COUNTERS] PROGRAM TOTAL          :    1.7692s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.5041s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.2652s for    90112 events => throughput is 3.40E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    1.9222s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.6468s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.2754s for    90112 events => throughput is 3.27E+05 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.21801182648615874) and cpp (0.21801182648615863) differ by less than 2E-14 (5.551115123125783e-16)
+OK! xsec from fortran (0.21801182648615872) and cpp (0.21801182648615869) differ by less than 2E-14 (1.1102230246251565e-16)
 
 *** (2-avx2) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.443138e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.305728e+05                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.417632e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.308522e+05                 )  sec^-1
 
 *** (2-512y) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -360,15 +360,15 @@ Executing ' ./build.512y_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_gqttq_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.2605 [0.26050333309703733] fbridge_mode=1
+ [XSECTION] Cross section = 0.2605 [0.26050333309703727] fbridge_mode=1
  [UNWEIGHT] Wrote 81 events (found 540 events)
- [COUNTERS] PROGRAM TOTAL          :    0.2906s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2677s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0228s for     8192 events => throughput is 3.59E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.2978s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2759s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0219s for     8192 events => throughput is 3.74E+05 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.26050333309703716) and cpp (0.26050333309703733) differ by less than 2E-14 (6.661338147750939e-16)
+OK! xsec from fortran (0.26050333309703716) and cpp (0.26050333309703727) differ by less than 2E-14 (4.440892098500626e-16)
 
 *** (2-512y) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -393,29 +393,29 @@ Executing ' ./build.512y_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_gqttq_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.218 [0.21801182648615863] fbridge_mode=1
+ [XSECTION] Cross section = 0.218 [0.21801182648615869] fbridge_mode=1
  [UNWEIGHT] Wrote 853 events (found 1849 events)
- [COUNTERS] PROGRAM TOTAL          :    1.7686s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.5286s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.2400s for    90112 events => throughput is 3.75E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    1.8887s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.6471s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.2415s for    90112 events => throughput is 3.73E+05 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.21801182648615874) and cpp (0.21801182648615863) differ by less than 2E-14 (5.551115123125783e-16)
+OK! xsec from fortran (0.21801182648615872) and cpp (0.21801182648615869) differ by less than 2E-14 (1.1102230246251565e-16)
 
 *** (2-512y) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.729118e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.792718e+05                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.747676e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.861917e+05                 )  sec^-1
 
 *** (2-512z) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -436,15 +436,15 @@ Executing ' ./build.512z_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_gqttq_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.2605 [0.26050333309703733] fbridge_mode=1
+ [XSECTION] Cross section = 0.2605 [0.26050333309703727] fbridge_mode=1
  [UNWEIGHT] Wrote 81 events (found 540 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3058s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2735s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0323s for     8192 events => throughput is 2.54E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3207s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2877s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0330s for     8192 events => throughput is 2.48E+05 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.26050333309703716) and cpp (0.26050333309703733) differ by less than 2E-14 (6.661338147750939e-16)
+OK! xsec from fortran (0.26050333309703716) and cpp (0.26050333309703727) differ by less than 2E-14 (4.440892098500626e-16)
 
 *** (2-512z) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -469,29 +469,29 @@ Executing ' ./build.512z_d_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_gqttq_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.218 [0.21801182648615863] fbridge_mode=1
+ [XSECTION] Cross section = 0.218 [0.21801182648615869] fbridge_mode=1
  [UNWEIGHT] Wrote 853 events (found 1849 events)
- [COUNTERS] PROGRAM TOTAL          :    1.8682s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.5123s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.3559s for    90112 events => throughput is 2.53E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.0189s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.6567s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.3622s for    90112 events => throughput is 2.49E+05 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.21801182648615874) and cpp (0.21801182648615863) differ by less than 2E-14 (5.551115123125783e-16)
+OK! xsec from fortran (0.21801182648615872) and cpp (0.21801182648615869) differ by less than 2E-14 (1.1102230246251565e-16)
 
 *** (2-512z) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.352575e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.535117e+05                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.365058e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.517520e+05                 )  sec^-1
 
 *** (3) EXECUTE MADEVENT_CUDA x1 (create events.lhe) ***
 --------------------
@@ -514,8 +514,8 @@ Executing ' ./build.none_d_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_gqttq_x
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.2605 [0.26050333309703733] fbridge_mode=1
  [UNWEIGHT] Wrote 81 events (found 540 events)
- [COUNTERS] PROGRAM TOTAL          :    0.6681s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.6674s
+ [COUNTERS] PROGRAM TOTAL          :    0.6879s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.6872s
  [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0007s for     8192 events => throughput is 1.21E+07 events/s
 
 *** (3) Compare MADEVENT_CUDA x1 xsec to MADEVENT_FORTRAN xsec ***
@@ -547,56 +547,56 @@ Executing ' ./build.none_d_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_gqttq_x
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.218 [0.21801182648615869] fbridge_mode=1
  [UNWEIGHT] Wrote 853 events (found 1849 events)
- [COUNTERS] PROGRAM TOTAL          :    1.9647s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.9564s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0083s for    90112 events => throughput is 1.09E+07 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.0638s
+ [COUNTERS] Fortran Overhead ( 0 ) :    2.0555s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0083s for    90112 events => throughput is 1.08E+07 events/s
 
 *** (3) Compare MADEVENT_CUDA x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.21801182648615874) and cpp (0.21801182648615869) differ by less than 2E-14 (2.220446049250313e-16)
+OK! xsec from fortran (0.21801182648615872) and cpp (0.21801182648615869) differ by less than 2E-14 (1.1102230246251565e-16)
 
 *** (3) Compare MADEVENT_CUDA x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cuda.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.568354e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.535376e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 4.005124e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.131781e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 16384 32 1 --bridge ***
-Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.387855e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.380880e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 16384 32 1 ***
-Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 6.497397e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.511409e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 4096 128 1 --bridge ***
-Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.373048e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.374806e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 4096 128 1 ***
-Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 6.771015e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.787335e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 65536 8 1 --bridge ***
-Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.382200e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.380768e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 65536 8 1 ***
-Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.783427e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.782273e+07                 )  sec^-1
 
 TEST COMPLETED
diff --git a/epochX/cudacpp/tmad/logs_gqttq_mad/log_gqttq_mad_f_inl0_hrd0.txt b/epochX/cudacpp/tmad/logs_gqttq_mad/log_gqttq_mad_f_inl0_hrd0.txt
index f629b5c150..4985f151b2 100644
--- a/epochX/cudacpp/tmad/logs_gqttq_mad/log_gqttq_mad_f_inl0_hrd0.txt
+++ b/epochX/cudacpp/tmad/logs_gqttq_mad/log_gqttq_mad_f_inl0_hrd0.txt
@@ -2,10 +2,10 @@ Working directory (build): /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/g
 CUDACPP_BUILDDIR='.'
 
 
-
 make USEBUILDDIR=1 AVX=none
 make USEBUILDDIR=1 AVX=sse4
 
+
 make USEBUILDDIR=1 AVX=avx2
 make USEBUILDDIR=1 AVX=512y
 
@@ -16,12 +16,12 @@ make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu'
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu'
 CUDACPP_BUILDDIR='build.none_f_inl0_hrd0'
-CUDACPP_BUILDDIR='build.sse4_f_inl0_hrd0'
-CUDACPP_BUILDDIR='build.512y_f_inl0_hrd0'
 CUDACPP_BUILDDIR='build.avx2_f_inl0_hrd0'
-CUDACPP_BUILDDIR='build.512z_f_inl0_hrd0'
+CUDACPP_BUILDDIR='build.512y_f_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu'
+CUDACPP_BUILDDIR='build.sse4_f_inl0_hrd0'
+CUDACPP_BUILDDIR='build.512z_f_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu'
 make[1]: Nothing to be done for 'all'.
@@ -33,7 +33,7 @@ make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/
 
 OMP_NUM_THREADS=
 
-DATE: 2023-11-24_15:51:46
+DATE: 2024-01-30_06:26:56
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 Working directory (run): /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu
@@ -59,9 +59,9 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_gqttq_x1_fortran > /tmp/ava
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.2605 [0.26050333309703716] fbridge_mode=0
  [UNWEIGHT] Wrote 78 events (found 561 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3117s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2399s
- [COUNTERS] Fortran MEs      ( 1 ) :    0.0718s for     8192 events => throughput is 1.14E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3277s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2529s
+ [COUNTERS] Fortran MEs      ( 1 ) :    0.0748s for     8192 events => throughput is 1.10E+05 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x1 (create events.lhe) ***
 --------------------
@@ -84,9 +84,9 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_gqttq_x1_fortran > /tmp/ava
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.2605 [0.26050333309703716] fbridge_mode=0
  [UNWEIGHT] Wrote 81 events (found 540 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3099s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2384s
- [COUNTERS] Fortran MEs      ( 1 ) :    0.0715s for     8192 events => throughput is 1.15E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3225s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2485s
+ [COUNTERS] Fortran MEs      ( 1 ) :    0.0740s for     8192 events => throughput is 1.11E+05 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x10 (create events.lhe) ***
 --------------------
@@ -107,11 +107,11 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_gqttq_x10_fortran > /tmp/av
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.218 [0.21801182648615874] fbridge_mode=0
+ [XSECTION] Cross section = 0.218 [0.21801182648615872] fbridge_mode=0
  [UNWEIGHT] Wrote 853 events (found 1849 events)
- [COUNTERS] PROGRAM TOTAL          :    2.2455s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.4617s
- [COUNTERS] Fortran MEs      ( 1 ) :    0.7838s for    90112 events => throughput is 1.15E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.4071s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.5926s
+ [COUNTERS] Fortran MEs      ( 1 ) :    0.8145s for    90112 events => throughput is 1.11E+05 events/s
 
 *** (2-none) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -132,15 +132,15 @@ Executing ' ./build.none_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_gqttq_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.2605 [0.26050316058770007] fbridge_mode=1
+ [XSECTION] Cross section = 0.2605 [0.26050314903825744] fbridge_mode=1
  [UNWEIGHT] Wrote 81 events (found 540 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3857s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3127s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0731s for     8192 events => throughput is 1.12E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.4015s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3277s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0738s for     8192 events => throughput is 1.11E+05 events/s
 
 *** (2-none) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.26050333309703716) and cpp (0.26050316058770007) differ by less than 4E-4 (6.622154696822591e-07)
+OK! xsec from fortran (0.26050333309703716) and cpp (0.26050314903825744) differ by less than 4E-4 (7.065505747139156e-07)
 
 *** (2-none) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -165,29 +165,29 @@ Executing ' ./build.none_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_gqttq_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.218 [0.21801182797520666] fbridge_mode=1
+ [XSECTION] Cross section = 0.218 [0.21801181770186087] fbridge_mode=1
  [UNWEIGHT] Wrote 853 events (found 1849 events)
- [COUNTERS] PROGRAM TOTAL          :    2.3684s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.5619s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.8065s for    90112 events => throughput is 1.12E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.5139s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.7000s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.8139s for    90112 events => throughput is 1.11E+05 events/s
 
 *** (2-none) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.21801182648615874) and cpp (0.21801182797520666) differ by less than 4E-4 (6.830124466006282e-09)
+OK! xsec from fortran (0.21801182648615872) and cpp (0.21801181770186087) differ by less than 4E-4 (4.0292758352045155e-08)
 
 *** (2-none) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.132102e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.131056e+05                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.140766e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.131141e+05                 )  sec^-1
 
 *** (2-sse4) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -208,15 +208,15 @@ Executing ' ./build.sse4_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_gqttq_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.2605 [0.26050313133963987] fbridge_mode=1
+ [XSECTION] Cross section = 0.2605 [0.26050310835231938] fbridge_mode=1
  [UNWEIGHT] Wrote 81 events (found 540 events)
- [COUNTERS] PROGRAM TOTAL          :    0.2925s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2662s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0263s for     8192 events => throughput is 3.12E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3069s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2800s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0269s for     8192 events => throughput is 3.04E+05 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.26050333309703716) and cpp (0.26050313133963987) differ by less than 4E-4 (7.744906558304621e-07)
+OK! xsec from fortran (0.26050333309703716) and cpp (0.26050310835231938) differ by less than 4E-4 (8.627325996934943e-07)
 
 *** (2-sse4) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -241,29 +241,29 @@ Executing ' ./build.sse4_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_gqttq_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.218 [0.21801179276862181] fbridge_mode=1
+ [XSECTION] Cross section = 0.218 [0.21801177817838580] fbridge_mode=1
  [UNWEIGHT] Wrote 853 events (found 1849 events)
- [COUNTERS] PROGRAM TOTAL          :    1.7892s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.5021s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.2870s for    90112 events => throughput is 3.14E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    1.9246s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.6468s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.2778s for    90112 events => throughput is 3.24E+05 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.21801182648615874) and cpp (0.21801179276862181) differ by less than 4E-4 (1.5465921032742358e-07)
+OK! xsec from fortran (0.21801182648615872) and cpp (0.21801177817838580) differ by less than 4E-4 (2.2158326773435988e-07)
 
 *** (2-sse4) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.189560e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.299610e+05                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.133717e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.290596e+05                 )  sec^-1
 
 *** (2-avx2) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -284,15 +284,15 @@ Executing ' ./build.avx2_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_gqttq_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.2605 [0.26050313344346482] fbridge_mode=1
+ [XSECTION] Cross section = 0.2605 [0.26050310803492405] fbridge_mode=1
  [UNWEIGHT] Wrote 81 events (found 540 events)
- [COUNTERS] PROGRAM TOTAL          :    0.2638s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2508s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0130s for     8192 events => throughput is 6.28E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.2803s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2672s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0131s for     8192 events => throughput is 6.25E+05 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.26050333309703716) and cpp (0.26050313344346482) differ by less than 4E-4 (7.664146557395668e-07)
+OK! xsec from fortran (0.26050333309703716) and cpp (0.26050310803492405) differ by less than 4E-4 (8.639509921914978e-07)
 
 *** (2-avx2) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -317,29 +317,29 @@ Executing ' ./build.avx2_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_gqttq_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.218 [0.21801179137376883] fbridge_mode=1
+ [XSECTION] Cross section = 0.218 [0.21801177493542723] fbridge_mode=1
  [UNWEIGHT] Wrote 853 events (found 1849 events)
- [COUNTERS] PROGRAM TOTAL          :    1.6261s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.4856s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.1405s for    90112 events => throughput is 6.42E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    1.7784s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.6325s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.1459s for    90112 events => throughput is 6.18E+05 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.21801182648615874) and cpp (0.21801179137376883) differ by less than 4E-4 (1.6105727140836024e-07)
+OK! xsec from fortran (0.21801182648615872) and cpp (0.21801177493542723) differ by less than 4E-4 (2.364584175129636e-07)
 
 *** (2-avx2) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 6.270432e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.309270e+05                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 6.321323e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.331612e+05                 )  sec^-1
 
 *** (2-512y) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -360,15 +360,15 @@ Executing ' ./build.512y_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_gqttq_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.2605 [0.26050313344346482] fbridge_mode=1
+ [XSECTION] Cross section = 0.2605 [0.26050310803492405] fbridge_mode=1
  [UNWEIGHT] Wrote 81 events (found 540 events)
- [COUNTERS] PROGRAM TOTAL          :    0.2747s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2611s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0135s for     8192 events => throughput is 6.06E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.2783s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2663s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0120s for     8192 events => throughput is 6.84E+05 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.26050333309703716) and cpp (0.26050313344346482) differ by less than 4E-4 (7.664146557395668e-07)
+OK! xsec from fortran (0.26050333309703716) and cpp (0.26050310803492405) differ by less than 4E-4 (8.639509921914978e-07)
 
 *** (2-512y) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -393,29 +393,29 @@ Executing ' ./build.512y_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_gqttq_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.218 [0.21801179137376883] fbridge_mode=1
+ [XSECTION] Cross section = 0.218 [0.21801177493542723] fbridge_mode=1
  [UNWEIGHT] Wrote 853 events (found 1849 events)
- [COUNTERS] PROGRAM TOTAL          :    1.6570s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.5241s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.1330s for    90112 events => throughput is 6.78E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    1.7659s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.6337s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.1322s for    90112 events => throughput is 6.82E+05 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.21801182648615874) and cpp (0.21801179137376883) differ by less than 4E-4 (1.6105727140836024e-07)
+OK! xsec from fortran (0.21801182648615872) and cpp (0.21801177493542723) differ by less than 4E-4 (2.364584175129636e-07)
 
 *** (2-512y) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 6.825162e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.987405e+05                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 6.838486e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.079276e+05                 )  sec^-1
 
 *** (2-512z) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -438,9 +438,9 @@ Executing ' ./build.512z_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_gqttq_x1
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.2605 [0.26050317064561834] fbridge_mode=1
  [UNWEIGHT] Wrote 81 events (found 540 events)
- [COUNTERS] PROGRAM TOTAL          :    0.2742s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2582s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0161s for     8192 events => throughput is 5.10E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.2869s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2697s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0173s for     8192 events => throughput is 4.74E+05 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
@@ -471,27 +471,27 @@ Executing ' ./build.512z_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_gqttq_x1
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.218 [0.21801182143140752] fbridge_mode=1
  [UNWEIGHT] Wrote 853 events (found 1849 events)
- [COUNTERS] PROGRAM TOTAL          :    1.7047s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.5193s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.1853s for    90112 events => throughput is 4.86E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    1.8266s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.6389s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.1877s for    90112 events => throughput is 4.80E+05 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.21801182648615874) and cpp (0.21801182143140752) differ by less than 4E-4 (2.3185674380421517e-08)
+OK! xsec from fortran (0.21801182648615872) and cpp (0.21801182143140752) differ by less than 4E-4 (2.3185674269399215e-08)
 
 *** (2-512z) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 4.717607e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.948471e+05                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 4.789235e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.943070e+05                 )  sec^-1
 
 *** (3) EXECUTE MADEVENT_CUDA x1 (create events.lhe) ***
 --------------------
@@ -514,9 +514,9 @@ Executing ' ./build.none_f_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_gqttq_x
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.2605 [0.26050319131407651] fbridge_mode=1
  [UNWEIGHT] Wrote 81 events (found 540 events)
- [COUNTERS] PROGRAM TOTAL          :    0.6713s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.6707s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0005s for     8192 events => throughput is 1.54E+07 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.6865s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.6860s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0005s for     8192 events => throughput is 1.55E+07 events/s
 
 *** (3) Compare MADEVENT_CUDA x1 xsec to MADEVENT_FORTRAN xsec ***
 
@@ -547,56 +547,56 @@ Executing ' ./build.none_f_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_gqttq_x
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.218 [0.21801186038252196] fbridge_mode=1
  [UNWEIGHT] Wrote 853 events (found 1849 events)
- [COUNTERS] PROGRAM TOTAL          :    1.9216s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.9152s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0064s for    90112 events => throughput is 1.41E+07 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.0580s
+ [COUNTERS] Fortran Overhead ( 0 ) :    2.0515s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0065s for    90112 events => throughput is 1.38E+07 events/s
 
 *** (3) Compare MADEVENT_CUDA x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.21801182648615874) and cpp (0.21801186038252196) differ by less than 4E-4 (1.5547946996541384e-07)
+OK! xsec from fortran (0.21801182648615872) and cpp (0.21801186038252196) differ by less than 4E-4 (1.5547946996541384e-07)
 
 *** (3) Compare MADEVENT_CUDA x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cuda.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.688652e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.646730e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 6.419660e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.486298e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 16384 32 1 --bridge ***
-Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 4.806306e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.789487e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 16384 32 1 ***
-Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.722148e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.699254e+08                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 4096 128 1 --bridge ***
-Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 4.808789e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.776807e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 4096 128 1 ***
-Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.809471e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.780761e+08                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 65536 8 1 --bridge ***
-Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 4.332763e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.361160e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 65536 8 1 ***
-Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 6.066283e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.995376e+07                 )  sec^-1
 
 TEST COMPLETED
diff --git a/epochX/cudacpp/tmad/logs_gqttq_mad/log_gqttq_mad_m_inl0_hrd0.txt b/epochX/cudacpp/tmad/logs_gqttq_mad/log_gqttq_mad_m_inl0_hrd0.txt
index 8fb8683f4e..44df8a9e3d 100644
--- a/epochX/cudacpp/tmad/logs_gqttq_mad/log_gqttq_mad_m_inl0_hrd0.txt
+++ b/epochX/cudacpp/tmad/logs_gqttq_mad/log_gqttq_mad_m_inl0_hrd0.txt
@@ -2,10 +2,10 @@ Working directory (build): /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/g
 CUDACPP_BUILDDIR='.'
 
 
-
 make USEBUILDDIR=1 AVX=none
 
 make USEBUILDDIR=1 AVX=sse4
+
 make USEBUILDDIR=1 AVX=avx2
 make USEBUILDDIR=1 AVX=512y
 
@@ -15,13 +15,13 @@ make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu'
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu'
 make[1]: Entering directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu'
+CUDACPP_BUILDDIR='build.512y_m_inl0_hrd0'
+CUDACPP_BUILDDIR='build.avx2_m_inl0_hrd0'
 CUDACPP_BUILDDIR='build.512z_m_inl0_hrd0'
-CUDACPP_BUILDDIR='build.none_m_inl0_hrd0'
 CUDACPP_BUILDDIR='build.sse4_m_inl0_hrd0'
-CUDACPP_BUILDDIR='build.avx2_m_inl0_hrd0'
+CUDACPP_BUILDDIR='build.none_m_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu'
-CUDACPP_BUILDDIR='build.512y_m_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu'
 make[1]: Nothing to be done for 'all'.
@@ -33,7 +33,7 @@ make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/
 
 OMP_NUM_THREADS=
 
-DATE: 2023-11-24_15:52:14
+DATE: 2024-01-30_06:27:27
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 Working directory (run): /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu
@@ -59,9 +59,9 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_gqttq_x1_fortran > /tmp/ava
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.2605 [0.26050333309703716] fbridge_mode=0
  [UNWEIGHT] Wrote 78 events (found 561 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3110s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2397s
- [COUNTERS] Fortran MEs      ( 1 ) :    0.0713s for     8192 events => throughput is 1.15E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3260s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2519s
+ [COUNTERS] Fortran MEs      ( 1 ) :    0.0740s for     8192 events => throughput is 1.11E+05 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x1 (create events.lhe) ***
 --------------------
@@ -84,9 +84,9 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_gqttq_x1_fortran > /tmp/ava
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.2605 [0.26050333309703716] fbridge_mode=0
  [UNWEIGHT] Wrote 81 events (found 540 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3123s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2407s
- [COUNTERS] Fortran MEs      ( 1 ) :    0.0716s for     8192 events => throughput is 1.14E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3236s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2490s
+ [COUNTERS] Fortran MEs      ( 1 ) :    0.0746s for     8192 events => throughput is 1.10E+05 events/s
 
 *** (1) EXECUTE MADEVENT_FORTRAN x10 (create events.lhe) ***
 --------------------
@@ -107,11 +107,11 @@ Executing ' ./madevent_fortran < /tmp/avalassi/input_gqttq_x10_fortran > /tmp/av
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.218 [0.21801182648615874] fbridge_mode=0
+ [XSECTION] Cross section = 0.218 [0.21801182648615872] fbridge_mode=0
  [UNWEIGHT] Wrote 853 events (found 1849 events)
- [COUNTERS] PROGRAM TOTAL          :    2.2374s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.4535s
- [COUNTERS] Fortran MEs      ( 1 ) :    0.7839s for    90112 events => throughput is 1.15E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.4069s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.5940s
+ [COUNTERS] Fortran MEs      ( 1 ) :    0.8130s for    90112 events => throughput is 1.11E+05 events/s
 
 *** (2-none) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -134,9 +134,9 @@ Executing ' ./build.none_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_gqttq_x1
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.2605 [0.26050333282657206] fbridge_mode=1
  [UNWEIGHT] Wrote 81 events (found 540 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3957s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.3178s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0780s for     8192 events => throughput is 1.05E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.4196s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.3367s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0829s for     8192 events => throughput is 9.88E+04 events/s
 
 *** (2-none) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
@@ -165,29 +165,29 @@ Executing ' ./build.none_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_gqttq_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.218 [0.21801182636608796] fbridge_mode=1
+ [XSECTION] Cross section = 0.218 [0.21801182636608801] fbridge_mode=1
  [UNWEIGHT] Wrote 853 events (found 1849 events)
- [COUNTERS] PROGRAM TOTAL          :    2.4134s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.5610s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.8524s for    90112 events => throughput is 1.06E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.6225s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.7110s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.9116s for    90112 events => throughput is 9.89E+04 events/s
 
 *** (2-none) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.21801182648615874) and cpp (0.21801182636608796) differ by less than 2E-4 (5.507535538740171e-10)
+OK! xsec from fortran (0.21801182648615872) and cpp (0.21801182636608801) differ by less than 2E-4 (5.507531097848073e-10)
 
 *** (2-none) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+BRDHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.065790e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.000214e+05                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.078999e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.992747e+04                 )  sec^-1
 
 *** (2-sse4) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -208,15 +208,15 @@ Executing ' ./build.sse4_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_gqttq_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.2605 [0.26050333282657201] fbridge_mode=1
+ [XSECTION] Cross section = 0.2605 [0.26050333282657212] fbridge_mode=1
  [UNWEIGHT] Wrote 81 events (found 540 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3246s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2839s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0407s for     8192 events => throughput is 2.01E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3393s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2968s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0425s for     8192 events => throughput is 1.93E+05 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.26050333309703716) and cpp (0.26050333282657201) differ by less than 2E-4 (1.0382406046005599e-09)
+OK! xsec from fortran (0.26050333309703716) and cpp (0.26050333282657212) differ by less than 2E-4 (1.0382402715336525e-09)
 
 *** (2-sse4) Compare MADEVENT_CPP x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -241,29 +241,29 @@ Executing ' ./build.sse4_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_gqttq_x1
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.218 [0.21801182636608810] fbridge_mode=1
+ [XSECTION] Cross section = 0.218 [0.21801182636608804] fbridge_mode=1
  [UNWEIGHT] Wrote 853 events (found 1849 events)
- [COUNTERS] PROGRAM TOTAL          :    1.9711s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.5247s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.4464s for    90112 events => throughput is 2.02E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.1463s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.6727s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.4735s for    90112 events => throughput is 1.90E+05 events/s
 
 *** (2-sse4) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.21801182648615874) and cpp (0.21801182636608810) differ by less than 2E-4 (5.507528877402024e-10)
+OK! xsec from fortran (0.21801182648615872) and cpp (0.21801182636608804) differ by less than 2E-4 (5.507529987625048e-10)
 
 *** (2-sse4) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+BRDHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.994361e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.936317e+05                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.977417e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.928004e+05                 )  sec^-1
 
 *** (2-avx2) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -286,9 +286,9 @@ Executing ' ./build.avx2_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_gqttq_x1
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.2605 [0.26050333291481387] fbridge_mode=1
  [UNWEIGHT] Wrote 81 events (found 540 events)
- [COUNTERS] PROGRAM TOTAL          :    0.2878s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2640s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0238s for     8192 events => throughput is 3.45E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3040s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2793s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0246s for     8192 events => throughput is 3.33E+05 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
@@ -319,27 +319,27 @@ Executing ' ./build.avx2_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_gqttq_x1
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.218 [0.21801182638680733] fbridge_mode=1
  [UNWEIGHT] Wrote 853 events (found 1849 events)
- [COUNTERS] PROGRAM TOTAL          :    1.7622s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.4988s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.2634s for    90112 events => throughput is 3.42E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    1.9238s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.6513s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.2724s for    90112 events => throughput is 3.31E+05 events/s
 
 *** (2-avx2) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.21801182648615874) and cpp (0.21801182638680733) differ by less than 2E-4 (4.557156874085422e-10)
+OK! xsec from fortran (0.21801182648615872) and cpp (0.21801182638680733) differ by less than 2E-4 (4.557155763862397e-10)
 
 *** (2-avx2) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+BRDHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.380399e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.358469e+05                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.457644e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.409112e+05                 )  sec^-1
 
 *** (2-512y) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -362,9 +362,9 @@ Executing ' ./build.512y_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_gqttq_x1
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.2605 [0.26050333291481387] fbridge_mode=1
  [UNWEIGHT] Wrote 81 events (found 540 events)
- [COUNTERS] PROGRAM TOTAL          :    0.2826s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2612s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0213s for     8192 events => throughput is 3.84E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.2986s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2774s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0212s for     8192 events => throughput is 3.86E+05 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
@@ -395,27 +395,27 @@ Executing ' ./build.512y_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_gqttq_x1
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.218 [0.21801182638680733] fbridge_mode=1
  [UNWEIGHT] Wrote 853 events (found 1849 events)
- [COUNTERS] PROGRAM TOTAL          :    1.7310s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.4992s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.2318s for    90112 events => throughput is 3.89E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    1.8897s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.6536s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.2361s for    90112 events => throughput is 3.82E+05 events/s
 
 *** (2-512y) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.21801182648615874) and cpp (0.21801182638680733) differ by less than 2E-4 (4.557156874085422e-10)
+OK! xsec from fortran (0.21801182648615872) and cpp (0.21801182638680733) differ by less than 2E-4 (4.557155763862397e-10)
 
 *** (2-512y) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+BRDHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.844575e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.861136e+05                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.903023e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.873456e+05                 )  sec^-1
 
 *** (2-512z) EXECUTE MADEVENT_CPP x1 (create events.lhe) ***
 --------------------
@@ -438,9 +438,9 @@ Executing ' ./build.512z_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_gqttq_x1
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.2605 [0.26050333291481387] fbridge_mode=1
  [UNWEIGHT] Wrote 81 events (found 540 events)
- [COUNTERS] PROGRAM TOTAL          :    0.3081s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.2751s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0330s for     8192 events => throughput is 2.48E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.3231s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.2893s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0338s for     8192 events => throughput is 2.42E+05 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x1 xsec to MADEVENT_FORTRAN xsec ***
 
@@ -471,27 +471,27 @@ Executing ' ./build.512z_m_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_gqttq_x1
  [XSECTION] ChannelId = 1
  [XSECTION] Cross section = 0.218 [0.21801182638680733] fbridge_mode=1
  [UNWEIGHT] Wrote 853 events (found 1849 events)
- [COUNTERS] PROGRAM TOTAL          :    1.8866s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.5159s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.3706s for    90112 events => throughput is 2.43E+05 events/s
+ [COUNTERS] PROGRAM TOTAL          :    2.0365s
+ [COUNTERS] Fortran Overhead ( 0 ) :    1.6643s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.3722s for    90112 events => throughput is 2.42E+05 events/s
 
 *** (2-512z) Compare MADEVENT_CPP x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.21801182648615874) and cpp (0.21801182638680733) differ by less than 2E-4 (4.557156874085422e-10)
+OK! xsec from fortran (0.21801182648615872) and cpp (0.21801182638680733) differ by less than 2E-4 (4.557155763862397e-10)
 
 *** (2-512z) Compare MADEVENT_CPP x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cpp.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE CHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+BRDHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.345187e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.448669e+05                 )  sec^-1
 
 *** EXECUTE CHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.413954e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.422091e+05                 )  sec^-1
 
 *** (3) EXECUTE MADEVENT_CUDA x1 (create events.lhe) ***
 --------------------
@@ -512,15 +512,15 @@ Executing ' ./build.none_m_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_gqttq_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.2605 [0.26050333301029693] fbridge_mode=1
+ [XSECTION] Cross section = 0.2605 [0.26050333301029699] fbridge_mode=1
  [UNWEIGHT] Wrote 81 events (found 540 events)
- [COUNTERS] PROGRAM TOTAL          :    0.6718s
- [COUNTERS] Fortran Overhead ( 0 ) :    0.6711s
- [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0007s for     8192 events => throughput is 1.18E+07 events/s
+ [COUNTERS] PROGRAM TOTAL          :    0.6889s
+ [COUNTERS] Fortran Overhead ( 0 ) :    0.6882s
+ [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0007s for     8192 events => throughput is 1.21E+07 events/s
 
 *** (3) Compare MADEVENT_CUDA x1 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.26050333309703716) and cpp (0.26050333301029693) differ by less than 2E-4 (3.329716502520341e-10)
+OK! xsec from fortran (0.26050333309703716) and cpp (0.26050333301029699) differ by less than 2E-4 (3.329714282074292e-10)
 
 *** (3) Compare MADEVENT_CUDA x1 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
@@ -545,58 +545,58 @@ Executing ' ./build.none_m_inl0_hrd0/madevent_cuda < /tmp/avalassi/input_gqttq_x
  [XSECTION] MultiChannel = TRUE
  [XSECTION] Configuration = 1
  [XSECTION] ChannelId = 1
- [XSECTION] Cross section = 0.218 [0.21801182637219935] fbridge_mode=1
+ [XSECTION] Cross section = 0.218 [0.21801182637219937] fbridge_mode=1
  [UNWEIGHT] Wrote 853 events (found 1849 events)
- [COUNTERS] PROGRAM TOTAL          :    1.9585s
- [COUNTERS] Fortran Overhead ( 0 ) :    1.9503s
+ [COUNTERS] PROGRAM TOTAL          :    2.0663s
+ [COUNTERS] Fortran Overhead ( 0 ) :    2.0581s
  [COUNTERS] CudaCpp MEs      ( 2 ) :    0.0082s for    90112 events => throughput is 1.10E+07 events/s
 
 *** (3) Compare MADEVENT_CUDA x10 xsec to MADEVENT_FORTRAN xsec ***
 
-OK! xsec from fortran (0.21801182648615874) and cpp (0.21801182637219935) differ by less than 2E-4 (5.227211996583492e-10)
+OK! xsec from fortran (0.21801182648615872) and cpp (0.21801182637219937) differ by less than 2E-4 (5.227208665914418e-10)
 
 *** (3) Compare MADEVENT_CUDA x10 events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
 
 OK! events.lhe.cuda.10 and events.lhe.ref.10 are identical
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 --bridge ***
-Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.574709e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.534715e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(8192) -p 256 32 1 ***
-Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 3.987655e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.123919e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 16384 32 1 --bridge ***
-Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.382555e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.382422e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX) -p 16384 32 1 ***
-Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 6.503146e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.503129e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 4096 128 1 --bridge ***
-Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.367209e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.385930e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX128THR) -p 4096 128 1 ***
-Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 6.807426e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.826918e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 65536 8 1 --bridge ***
-Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 2.377892e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.379565e+07                 )  sec^-1
 
 *** EXECUTE GCHECK(MAX8THR) -p 65536 8 1 ***
-Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GU_TTXU_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
-EvtsPerSec[MECalcOnly] (3a) = ( 1.781332e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.789199e+07                 )  sec^-1
 
 TEST COMPLETED
diff --git a/epochX/cudacpp/tmad/madX.sh b/epochX/cudacpp/tmad/madX.sh
index d03ca9f65f..991ab62ea2 100755
--- a/epochX/cudacpp/tmad/madX.sh
+++ b/epochX/cudacpp/tmad/madX.sh
@@ -420,7 +420,13 @@ printf "\nOMP_NUM_THREADS=$OMP_NUM_THREADS\n"
 
 printf "\nDATE: $(date '+%Y-%m-%d_%H:%M:%S')\n\n"
 
-if nvidia-smi -L > /dev/null 2>&1; then gpuTxt="$(nvidia-smi -L | wc -l)x $(nvidia-smi -L | awk '{print $3,$4}' | sort -u)"; else gpuTxt=none; fi
+if nvidia-smi -L > /dev/null 2>&1; then
+  gpuTxt="$(nvidia-smi -L | wc -l)x $(nvidia-smi -L | awk '{print $3,$4}' | sort -u)"
+elif rocm-smi -i > /dev/null 2>&1; then
+  gpuTxt="$(rocm-smi --showproductname | grep 'Card series' | awk '{print $5,$6,$7}')"
+else
+  gpuTxt=none
+fi
 if [ "${unames}" == "Darwin" ]; then 
   cpuTxt=$(sysctl -h machdep.cpu.brand_string)
   cpuTxt=${cpuTxt/machdep.cpu.brand_string: }
diff --git a/epochX/cudacpp/tput/allTees.sh b/epochX/cudacpp/tput/allTees.sh
index f67fa5eccf..9be4f5d4fc 100755
--- a/epochX/cudacpp/tput/allTees.sh
+++ b/epochX/cudacpp/tput/allTees.sh
@@ -5,12 +5,14 @@
 
 scrdir=$(cd $(dirname $0); pwd)
 
-# By default, use the madevent+cudacpp version of code and tee scripts
-sa=
+# By default, use the madevent+cudacpp version of code and tee scripts (use -sa to use the standalone version instead)
+# By default, build and run all tests (use -makeonly to only build all tests)
+opts=
 suff=".mad"
 
 # Parse command line arguments
 ggttggg=-ggttggg
+rndhst=-curhst
 while [ "$1" != "" ]; do
   if [ "$1" == "-short" ]; then
     # Short (no ggttggg) or long version?
@@ -22,11 +24,19 @@ while [ "$1" != "" ]; do
     shift
   elif [ "$1" == "-sa" ]; then
     # Use standalone_cudacpp builds instead of madevent+cudacpp?
-    sa=-sa
+    opts+=" -sa"
     suff=".sa"
     shift
+  elif [ "$1" == "-makeonly" ]; then
+    # Only build all tests instead of building and running them?
+    opts+=" -makeonly"
+    shift
+  elif [ "$1" == "-hip" ]; then
+    # Random numbers use rocrand instead of curand?
+    rndhst=-rorhst
+    shift
   else
-    echo "Usage: $0 [-short] [-e] [-sa]"
+    echo "Usage: $0 [-short] [-e] [-sa] [-makeonly] [-hip]"
     exit 1
   fi
 done
@@ -40,7 +50,7 @@ started="STARTED  AT $(date)"
 
 # (36/78) Six logs (double/float/mixed x hrd0/hrd1 x inl0) in each of the six processes
 \rm -rf gg_ttggg${suff}/lib/build.none_*
-cmd="./tput/teeThroughputX.sh -mix -hrd -makej -eemumu -ggtt -ggttg -ggttgg -gqttq $ggttggg -makeclean ${sa}"
+cmd="./tput/teeThroughputX.sh -mix -hrd -makej -eemumu -ggtt -ggttg -ggttgg -gqttq $ggttggg -makeclean ${opts}"
 $cmd; status=$?
 ended1="$cmd\nENDED(1) AT $(date) [Status=$status]"
 tmp1=$(mktemp)
@@ -49,29 +59,29 @@ ls -ltr ee_mumu${suff}/lib/build.none_*_inl0_hrd* gg_tt${suff}/lib/build.none_*_
 # (48/78) Four extra logs (double/float x hrd0/hrd1 x inl1) only in three of the six processes
 \rm -rf gg_ttg${suff}/lib/build.none_*
 \rm -rf gg_ttggg${suff}/lib/build.none_*
-cmd="./tput/teeThroughputX.sh -flt -hrd -makej -eemumu -ggtt -ggttgg -inlonly -makeclean ${sa}"
+cmd="./tput/teeThroughputX.sh -flt -hrd -makej -eemumu -ggtt -ggttgg -inlonly -makeclean ${opts}"
 $cmd; status=$?
 ended2="$cmd\nENDED(2) AT $(date) [Status=$status]"
 tmp2=$(mktemp)
 ls -ltr ee_mumu${suff}/lib/build.none_*_inl1_hrd* gg_tt${suff}/lib/build.none_*_inl1_hrd* gg_tt*g${suff}/lib/build.none_*_inl1_hrd* | egrep -v '(total|\./|\.build|_common|^$)' > $tmp2
 
 # (60/78) Two extra logs (double/float x hrd0 x inl0 + bridge) in all six processes (rebuild from cache)
-cmd="./tput/teeThroughputX.sh -makej -eemumu -ggtt -ggttg -gqttq -ggttgg $ggttggg -flt -bridge -makeclean ${sa}"
+cmd="./tput/teeThroughputX.sh -makej -eemumu -ggtt -ggttg -gqttq -ggttgg $ggttggg -flt -bridge -makeclean ${opts}"
 $cmd; status=$?
 ended3="$cmd\nENDED(3) AT $(date) [Status=$status]"
 
 # (66/78) Two extra logs (double/float x hrd0 x inl0 + rmbhst) only in three of the six processes (no rebuild needed)
-cmd="./tput/teeThroughputX.sh -eemumu -ggtt -ggttgg -flt -rmbhst ${sa}"
+cmd="./tput/teeThroughputX.sh -eemumu -ggtt -ggttgg -flt -rmbhst ${opts}"
 $cmd; status=$?
 ended4="$cmd\nENDED(4) AT $(date) [Status=$status]"
 
-# (72/78) Two extra logs (double/float x hrd0 x inl0 + curhst) only in three of the six processes (no rebuild needed)
-cmd="./tput/teeThroughputX.sh -eemumu -ggtt -ggttgg -flt -curhst ${sa}"
+# (72/78) Two extra logs (double/float x hrd0 x inl0 + rndhst) only in three of the six processes (no rebuild needed)
+cmd="./tput/teeThroughputX.sh -eemumu -ggtt -ggttgg -flt ${rndhst} ${opts}"
 $cmd; status=$?
 ended5="$cmd\nENDED(5) AT $(date) [Status=$status]"
 
 # (78/78) Two extra logs (double/float x hrd0 x inl0 + common) only in three of the six processes (no rebuild needed)
-cmd="./tput/teeThroughputX.sh -eemumu -ggtt -ggttgg -flt -common ${sa}"
+cmd="./tput/teeThroughputX.sh -eemumu -ggtt -ggttgg -flt -common ${opts}"
 $cmd; status=$?
 ended6="$cmd\nENDED(6) AT $(date) [Status=$status]"
 
@@ -92,8 +102,8 @@ echo -e "$ended5"
 if [ "$ggttggg" == "" ]; then
   echo
   echo "To complete the test for ggttggg type:"
-  echo "  ./tput/teeThroughputX.sh -flt -hrd -makej -ggttggg -makeclean ${sa}"
-  echo "  ./tput/teeThroughputX.sh -makej -ggttggg -flt -bridge -makeclean ${sa}"
+  echo "  ./tput/teeThroughputX.sh -flt -hrd -makej -ggttggg -makeclean ${opts}"
+  echo "  ./tput/teeThroughputX.sh -makej -ggttggg -flt -bridge -makeclean ${opts}"
 fi
 
 # Print out any errors in the logs
diff --git a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_d_inl0_hrd0.txt b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_d_inl0_hrd0.txt
index 352d1c6fba..15dbd5f8d1 100644
--- a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_d_inl0_hrd0.txt
+++ b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_d_inl0_hrd0.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum'
 
-DATE: 2023-11-24_14:27:52
+DATE: 2024-01-30_04:51:46
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_d_inl0_hrd0/gcheck.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 5.424219e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.270852e+08                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.136825e+08                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.572573e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.281942e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.116391e+08                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     0.828494 sec
-     2,799,509,879      cycles                           #    2.917 GHz                    
-     4,362,735,667      instructions                     #    1.56  insn per cycle         
-       1.170786520 seconds time elapsed
+TOTAL       :     0.839714 sec
+     2,719,217,340      cycles                           #    2.832 GHz                    
+     4,277,615,433      instructions                     #    1.57  insn per cycle         
+       1.175143775 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_d_inl0_hrd0/gcheck.exe -p 2048 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 166
@@ -72,20 +72,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_d_inl0_hrd0/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.137542e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.334591e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.334591e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 9.879157e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.147243e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.147243e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     5.926738 sec
-    18,326,467,169      cycles                           #    3.090 GHz                    
-    43,971,697,990      instructions                     #    2.40  insn per cycle         
-       5.940094102 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  433) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     6.790847 sec
+    19,539,640,504      cycles                           #    2.876 GHz                    
+    46,935,351,432      instructions                     #    2.40  insn per cycle         
+       6.804517518 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  472) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -99,20 +99,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_d_inl0_hrd0/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.696452e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.222211e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.222211e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.545376e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.021398e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.021398e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     4.102636 sec
-    12,781,292,854      cycles                           #    3.112 GHz                    
-    30,998,546,187      instructions                     #    2.43  insn per cycle         
-       4.125957125 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 1644) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     4.488904 sec
+    12,869,370,410      cycles                           #    2.864 GHz                    
+    31,186,180,279      instructions                     #    2.42  insn per cycle         
+       4.505888529 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 1626) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -126,20 +126,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_d_inl0_hrd0/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.105968e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.947022e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.947022e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.955981e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.735873e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.735873e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     3.375754 sec
-    10,081,613,505      cycles                           #    2.982 GHz                    
-    19,365,068,208      instructions                     #    1.92  insn per cycle         
-       3.393379455 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1966) (512y:    0) (512z:    0)
+TOTAL       :     3.633222 sec
+    10,032,348,170      cycles                           #    2.758 GHz                    
+    19,481,701,848      instructions                     #    1.94  insn per cycle         
+       3.651370321 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1964) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -147,26 +147,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282804e-02
-Avg ME (F77/C++)    = 1.2828039868165088E-002
-Relative difference = 1.0277089312025782e-08
+Avg ME (F77/C++)    = 1.2828039868165090E-002
+Relative difference = 1.0277089176796747e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_d_inl0_hrd0/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.197053e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.113568e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.113568e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.070263e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.978600e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.978600e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     3.253132 sec
-     9,700,425,131      cycles                           #    2.977 GHz                    
-    18,987,900,885      instructions                     #    1.96  insn per cycle         
-       3.272883868 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1690) (512y:  181) (512z:    0)
+TOTAL       :     3.453661 sec
+     9,572,367,477      cycles                           #    2.767 GHz                    
+    18,943,715,958      instructions                     #    1.98  insn per cycle         
+       3.473553059 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1655) (512y:  161) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -174,26 +174,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282804e-02
-Avg ME (F77/C++)    = 1.2828039868165088E-002
-Relative difference = 1.0277089312025782e-08
+Avg ME (F77/C++)    = 1.2828039868165090E-002
+Relative difference = 1.0277089176796747e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_d_inl0_hrd0/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.888824e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.524315e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.524315e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.819162e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.469996e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.469996e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     3.721051 sec
-     8,612,012,158      cycles                           #    2.311 GHz                    
-    15,727,858,115      instructions                     #    1.83  insn per cycle         
-       3.740799653 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:  901) (512y:  154) (512z: 1258)
+TOTAL       :     3.879359 sec
+     8,193,098,191      cycles                           #    2.110 GHz                    
+    15,513,331,501      instructions                     #    1.89  insn per cycle         
+       3.898953032 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:  920) (512y:   59) (512z: 1220)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -201,8 +201,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282804e-02
-Avg ME (F77/C++)    = 1.2828039868165088E-002
-Relative difference = 1.0277089312025782e-08
+Avg ME (F77/C++)    = 1.2828039868165090E-002
+Relative difference = 1.0277089176796747e-08
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_d_inl0_hrd0_bridge.txt b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_d_inl0_hrd0_bridge.txt
index c9fd7402fe..f78ea7251e 100644
--- a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_d_inl0_hrd0_bridge.txt
+++ b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_d_inl0_hrd0_bridge.txt
@@ -41,7 +41,7 @@ CUDACPP_BUILDDIR='build.512z_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum'
 
-DATE: 2023-11-24_15:09:46
+DATE: 2024-01-30_05:45:26
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
@@ -51,17 +51,17 @@ WARNING! Bridge selected: cannot use RamboDevice, will use RamboHost
 WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
 WARNING! Instantiate device Bridge (nevt=524288, gpublocks=2048, gputhreads=256, gpublocks*gputhreads=524288)
 WARNING! Set grid in Bridge (nevt=524288, gpublocks=2048, gputhreads=256, gpublocks*gputhreads=524288)
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 4.563802e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.504997e+07                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.504997e+07                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.460171e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.485962e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.485962e+07                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     2.276846 sec
-     7,462,184,879      cycles                           #    2.960 GHz                    
-    13,240,340,961      instructions                     #    1.77  insn per cycle         
-       2.578354941 seconds time elapsed
+TOTAL       :     2.319187 sec
+     7,341,770,811      cycles                           #    2.857 GHz                    
+    13,101,723,847      instructions                     #    1.78  insn per cycle         
+       2.628471382 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_d_inl0_hrd0/gcheck.exe -p 2048 256 1 --bridge
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Bridge selected: cannot use RamboDevice, will use RamboHost
@@ -81,20 +81,20 @@ OK (relative difference <= 5E-3)
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_d_inl0_hrd0/check.exe -p 2048 256 12 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=524288)
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.060089e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.241158e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.241158e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 9.576223e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.107198e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.107198e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     6.541091 sec
-    19,576,478,878      cycles                           #    2.990 GHz                    
-    44,199,334,845      instructions                     #    2.26  insn per cycle         
-       6.547828573 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  433) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     7.190455 sec
+    20,703,597,440      cycles                           #    2.877 GHz                    
+    47,160,901,733      instructions                     #    2.28  insn per cycle         
+       7.198222207 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  472) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -109,20 +109,20 @@ OK (relative difference <= 5E-3)
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_d_inl0_hrd0/check.exe -p 2048 256 12 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=524288)
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/sse4+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.520389e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.958155e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.958155e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.473769e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.897978e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.897978e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     4.763580 sec
-    13,999,932,501      cycles                           #    2.936 GHz                    
-    31,842,828,089      instructions                     #    2.27  insn per cycle         
-       4.771019993 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 1644) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     4.898106 sec
+    14,084,591,919      cycles                           #    2.873 GHz                    
+    32,028,151,491      instructions                     #    2.27  insn per cycle         
+       4.906157596 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 1626) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -137,20 +137,20 @@ OK (relative difference <= 5E-3)
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_d_inl0_hrd0/check.exe -p 2048 256 12 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=524288)
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/avx2+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.885630e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.569986e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.569986e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.834615e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.502061e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.502061e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     3.963354 sec
-    11,320,178,526      cycles                           #    2.852 GHz                    
-    20,728,383,013      instructions                     #    1.83  insn per cycle         
-       3.970490035 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1966) (512y:    0) (512z:    0)
+TOTAL       :     4.065584 sec
+    11,264,443,170      cycles                           #    2.767 GHz                    
+    20,844,723,129      instructions                     #    1.85  insn per cycle         
+       4.073296839 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1964) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -158,27 +158,27 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282804e-02
-Avg ME (F77/C++)    = 1.2828039868165088E-002
-Relative difference = 1.0277089312025782e-08
+Avg ME (F77/C++)    = 1.2828039868165090E-002
+Relative difference = 1.0277089176796747e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_d_inl0_hrd0/check.exe -p 2048 256 12 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=524288)
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/512y+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.971664e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.715052e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.715052e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.930005e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.695920e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.695920e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     3.811146 sec
-    10,942,360,524      cycles                           #    2.867 GHz                    
-    20,336,264,592      instructions                     #    1.86  insn per cycle         
-       3.818008523 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1690) (512y:  181) (512z:    0)
+TOTAL       :     3.900573 sec
+    10,821,072,419      cycles                           #    2.771 GHz                    
+    20,305,054,668      instructions                     #    1.88  insn per cycle         
+       3.908355042 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1655) (512y:  161) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -186,27 +186,27 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282804e-02
-Avg ME (F77/C++)    = 1.2828039868165088E-002
-Relative difference = 1.0277089312025782e-08
+Avg ME (F77/C++)    = 1.2828039868165090E-002
+Relative difference = 1.0277089176796747e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_d_inl0_hrd0/check.exe -p 2048 256 12 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=524288)
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/512z+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.699997e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.217210e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.217210e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.707724e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.274502e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.274502e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     4.325158 sec
-     9,912,314,209      cycles                           #    2.289 GHz                    
-    16,872,526,974      instructions                     #    1.70  insn per cycle         
-       4.332090544 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:  901) (512y:  154) (512z: 1258)
+TOTAL       :     4.333313 sec
+     9,497,951,325      cycles                           #    2.189 GHz                    
+    16,666,820,850      instructions                     #    1.75  insn per cycle         
+       4.341233179 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:  920) (512y:   59) (512z: 1220)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -214,8 +214,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282804e-02
-Avg ME (F77/C++)    = 1.2828039868165088E-002
-Relative difference = 1.0277089312025782e-08
+Avg ME (F77/C++)    = 1.2828039868165090E-002
+Relative difference = 1.0277089176796747e-08
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_d_inl0_hrd0_common.txt b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_d_inl0_hrd0_common.txt
index 4e10a04e19..f072467bfa 100644
--- a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_d_inl0_hrd0_common.txt
+++ b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_d_inl0_hrd0_common.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum'
 
-DATE: 2023-11-24_15:23:09
+DATE: 2024-01-30_05:59:18
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_d_inl0_hrd0/gcheck.exe -p 2048 256 12 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:COMMON+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 6.494137e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.568219e+08                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.090512e+08                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.483909e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.562012e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.071690e+08                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371879e-02 +- 3.270020e-06 )  GeV^0
-TOTAL       :     1.328549 sec
-     4,643,802,666      cycles                           #    2.984 GHz                    
-     7,163,037,263      instructions                     #    1.54  insn per cycle         
-       1.615810184 seconds time elapsed
+TOTAL       :     1.371489 sec
+     4,620,404,364      cycles                           #    2.861 GHz                    
+     7,153,271,516      instructions                     #    1.55  insn per cycle         
+       1.672602435 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_d_inl0_hrd0/gcheck.exe -p 2048 256 1 --common
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 166
@@ -72,20 +72,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_d_inl0_hrd0/check.exe -p 2048 256 12 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:COMMON+RMBHST+MESHST/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.098849e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.289504e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.289504e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 9.952512e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.155636e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.155636e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371879e-02 +- 3.270020e-06 )  GeV^0
-TOTAL       :     6.492604 sec
-    19,420,681,078      cycles                           #    2.994 GHz                    
-    44,081,202,964      instructions                     #    2.27  insn per cycle         
-       6.497991580 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  433) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     7.107803 sec
+    20,592,800,911      cycles                           #    2.895 GHz                    
+    47,037,031,319      instructions                     #    2.28  insn per cycle         
+       7.114495241 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  472) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -99,20 +99,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_d_inl0_hrd0/check.exe -p 2048 256 12 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:COMMON+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.642683e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.149460e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.149460e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.558277e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.038534e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.038534e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371879e-02 +- 3.270020e-06 )  GeV^0
-TOTAL       :     4.592487 sec
-    13,879,570,492      cycles                           #    3.021 GHz                    
-    31,002,445,042      instructions                     #    2.23  insn per cycle         
-       4.597888713 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 1644) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     4.822482 sec
+    13,870,774,877      cycles                           #    2.874 GHz                    
+    31,186,249,487      instructions                     #    2.25  insn per cycle         
+       4.828845646 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 1626) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -126,20 +126,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_d_inl0_hrd0/check.exe -p 2048 256 12 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:COMMON+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.041509e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.847095e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.847095e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.951724e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.730389e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.730389e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371879e-02 +- 3.270020e-06 )  GeV^0
-TOTAL       :     3.835804 sec
-    11,175,680,204      cycles                           #    2.910 GHz                    
-    19,267,089,581      instructions                     #    1.72  insn per cycle         
-       3.841361823 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1966) (512y:    0) (512z:    0)
+TOTAL       :     4.015384 sec
+    11,119,337,735      cycles                           #    2.766 GHz                    
+    19,381,852,554      instructions                     #    1.74  insn per cycle         
+       4.022009475 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1964) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -147,26 +147,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282804e-02
-Avg ME (F77/C++)    = 1.2828039868165088E-002
-Relative difference = 1.0277089312025782e-08
+Avg ME (F77/C++)    = 1.2828039868165090E-002
+Relative difference = 1.0277089176796747e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_d_inl0_hrd0/check.exe -p 2048 256 12 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:COMMON+RMBHST+MESHST/512y+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.097943e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.989766e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.989766e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.063314e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.951443e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.951443e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371879e-02 +- 3.270020e-06 )  GeV^0
-TOTAL       :     3.768859 sec
-    10,910,527,024      cycles                           #    2.891 GHz                    
-    18,677,308,293      instructions                     #    1.71  insn per cycle         
-       3.774878679 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1690) (512y:  181) (512z:    0)
+TOTAL       :     3.845408 sec
+    10,662,597,452      cycles                           #    2.769 GHz                    
+    18,643,141,459      instructions                     #    1.75  insn per cycle         
+       3.852109381 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1655) (512y:  161) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -174,26 +174,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282804e-02
-Avg ME (F77/C++)    = 1.2828039868165088E-002
-Relative difference = 1.0277089312025782e-08
+Avg ME (F77/C++)    = 1.2828039868165090E-002
+Relative difference = 1.0277089176796747e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_d_inl0_hrd0/check.exe -p 2048 256 12 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:COMMON+RMBHST+MESHST/512z+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.768127e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.357872e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.357872e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.811483e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.460421e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.460421e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371879e-02 +- 3.270020e-06 )  GeV^0
-TOTAL       :     4.339198 sec
-     9,800,415,487      cycles                           #    2.256 GHz                    
-    15,427,741,546      instructions                     #    1.57  insn per cycle         
-       4.345173089 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:  901) (512y:  154) (512z: 1258)
+TOTAL       :     4.272939 sec
+     9,279,488,955      cycles                           #    2.169 GHz                    
+    15,212,537,826      instructions                     #    1.64  insn per cycle         
+       4.279485071 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:  920) (512y:   59) (512z: 1220)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -201,8 +201,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282804e-02
-Avg ME (F77/C++)    = 1.2828039868165088E-002
-Relative difference = 1.0277089312025782e-08
+Avg ME (F77/C++)    = 1.2828039868165090E-002
+Relative difference = 1.0277089176796747e-08
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_d_inl0_hrd0_curhst.txt b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_d_inl0_hrd0_curhst.txt
index 6f403f72b4..a6db5de426 100644
--- a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_d_inl0_hrd0_curhst.txt
+++ b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_d_inl0_hrd0_curhst.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum'
 
-DATE: 2023-11-24_15:19:52
+DATE: 2024-01-30_05:55:54
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_d_inl0_hrd0/gcheck.exe -p 2048 256 12 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 6.511914e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.599963e+08                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.148106e+08                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.492089e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.565509e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.085712e+08                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     0.984391 sec
-     3,583,458,511      cycles                           #    2.950 GHz                    
-     7,069,877,265      instructions                     #    1.97  insn per cycle         
-       1.271392022 seconds time elapsed
+TOTAL       :     0.999229 sec
+     3,503,665,967      cycles                           #    2.851 GHz                    
+     7,040,796,455      instructions                     #    2.01  insn per cycle         
+       1.289089254 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_d_inl0_hrd0/gcheck.exe -p 2048 256 1 --curhst
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 166
@@ -72,20 +72,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_d_inl0_hrd0/check.exe -p 2048 256 12 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.096082e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.286319e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.286319e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 9.897604e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.152411e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.152411e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     6.151369 sec
-    18,323,138,260      cycles                           #    2.977 GHz                    
-    43,972,152,337      instructions                     #    2.40  insn per cycle         
-       6.156982875 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  433) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     6.777699 sec
+    19,525,012,140      cycles                           #    2.879 GHz                    
+    46,935,602,227      instructions                     #    2.40  insn per cycle         
+       6.784496054 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  472) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -99,20 +99,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_d_inl0_hrd0/check.exe -p 2048 256 12 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.603629e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.100551e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.100551e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.565929e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.046315e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.046315e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     4.336645 sec
-    12,767,737,304      cycles                           #    2.941 GHz                    
-    30,998,256,840      instructions                     #    2.43  insn per cycle         
-       4.342127866 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 1644) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     4.431371 sec
+    12,844,580,525      cycles                           #    2.895 GHz                    
+    31,183,505,413      instructions                     #    2.43  insn per cycle         
+       4.438022505 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 1626) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -126,20 +126,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_d_inl0_hrd0/check.exe -p 2048 256 12 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.019068e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.818538e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.818538e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.956069e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.738681e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.738681e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     3.519784 sec
-    10,072,637,753      cycles                           #    2.858 GHz                    
-    19,364,816,632      instructions                     #    1.92  insn per cycle         
-       3.525396454 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1966) (512y:    0) (512z:    0)
+TOTAL       :     3.632695 sec
+    10,040,197,478      cycles                           #    2.761 GHz                    
+    19,480,754,402      instructions                     #    1.94  insn per cycle         
+       3.639336589 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1964) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -147,26 +147,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282804e-02
-Avg ME (F77/C++)    = 1.2828039868165088E-002
-Relative difference = 1.0277089312025782e-08
+Avg ME (F77/C++)    = 1.2828039868165090E-002
+Relative difference = 1.0277089176796747e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_d_inl0_hrd0/check.exe -p 2048 256 12 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.119317e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.001533e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.001533e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.068909e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.973543e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.973543e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     3.367447 sec
-     9,690,886,611      cycles                           #    2.874 GHz                    
-    18,986,470,372      instructions                     #    1.96  insn per cycle         
-       3.372983732 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1690) (512y:  181) (512z:    0)
+TOTAL       :     3.456026 sec
+     9,583,252,780      cycles                           #    2.770 GHz                    
+    18,943,299,087      instructions                     #    1.98  insn per cycle         
+       3.462550493 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1655) (512y:  161) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -174,26 +174,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282804e-02
-Avg ME (F77/C++)    = 1.2828039868165088E-002
-Relative difference = 1.0277089312025782e-08
+Avg ME (F77/C++)    = 1.2828039868165090E-002
+Relative difference = 1.0277089176796747e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_d_inl0_hrd0/check.exe -p 2048 256 12 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.792369e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.378971e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.378971e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.820163e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.473451e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.473451e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     3.914886 sec
-     8,594,430,788      cycles                           #    2.193 GHz                    
-    15,726,409,723      instructions                     #    1.83  insn per cycle         
-       3.920405262 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:  901) (512y:  154) (512z: 1258)
+TOTAL       :     3.874228 sec
+     8,184,248,497      cycles                           #    2.110 GHz                    
+    15,512,168,002      instructions                     #    1.90  insn per cycle         
+       3.880483923 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:  920) (512y:   59) (512z: 1220)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -201,8 +201,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282804e-02
-Avg ME (F77/C++)    = 1.2828039868165088E-002
-Relative difference = 1.0277089312025782e-08
+Avg ME (F77/C++)    = 1.2828039868165090E-002
+Relative difference = 1.0277089176796747e-08
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_d_inl0_hrd0_rmbhst.txt b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_d_inl0_hrd0_rmbhst.txt
index d1ecb99501..4dded3e862 100644
--- a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_d_inl0_hrd0_rmbhst.txt
+++ b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_d_inl0_hrd0_rmbhst.txt
@@ -41,24 +41,24 @@ CUDACPP_BUILDDIR='build.512z_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum'
 
-DATE: 2023-11-24_15:16:32
+DATE: 2024-01-30_05:52:26
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_d_inl0_hrd0/gcheck.exe -p 2048 256 12 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 5.945904e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.504449e+08                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.028089e+08                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.831383e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.529080e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.990768e+08                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     1.911118 sec
-     6,389,730,073      cycles                           #    2.981 GHz                    
-    11,525,444,460      instructions                     #    1.80  insn per cycle         
-       2.200099911 seconds time elapsed
+TOTAL       :     1.936415 sec
+     6,196,996,673      cycles                           #    2.858 GHz                    
+    11,355,646,527      instructions                     #    1.83  insn per cycle         
+       2.226164304 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_d_inl0_hrd0/gcheck.exe -p 2048 256 1 --rmbhst
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
@@ -74,20 +74,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_d_inl0_hrd0/check.exe -p 2048 256 12 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.103959e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.297288e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.297288e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 9.923680e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.152570e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.152570e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     6.107673 sec
-    18,355,628,587      cycles                           #    3.004 GHz                    
-    43,972,892,726      instructions                     #    2.40  insn per cycle         
-       6.113444450 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  433) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     6.755774 sec
+    19,508,468,124      cycles                           #    2.886 GHz                    
+    46,934,079,079      instructions                     #    2.41  insn per cycle         
+       6.762162730 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  472) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -101,20 +101,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_d_inl0_hrd0/check.exe -p 2048 256 12 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.634128e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.138204e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.138204e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.560350e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.041132e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.041132e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     4.254868 sec
-    12,786,896,134      cycles                           #    3.002 GHz                    
-    30,998,062,799      instructions                     #    2.42  insn per cycle         
-       4.260594817 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 1644) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     4.445978 sec
+    12,824,682,223      cycles                           #    2.881 GHz                    
+    31,183,984,467      instructions                     #    2.43  insn per cycle         
+       4.452647644 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 1626) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -128,20 +128,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_d_inl0_hrd0/check.exe -p 2048 256 12 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.027528e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.835087e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.835087e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.945035e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.719021e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.719021e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     3.504925 sec
-    10,096,961,114      cycles                           #    2.877 GHz                    
-    19,366,201,509      instructions                     #    1.92  insn per cycle         
-       3.510626038 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1966) (512y:    0) (512z:    0)
+TOTAL       :     3.651562 sec
+    10,054,417,482      cycles                           #    2.750 GHz                    
+    19,480,651,159      instructions                     #    1.94  insn per cycle         
+       3.658175830 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1964) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -149,26 +149,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282804e-02
-Avg ME (F77/C++)    = 1.2828039868165088E-002
-Relative difference = 1.0277089312025782e-08
+Avg ME (F77/C++)    = 1.2828039868165090E-002
+Relative difference = 1.0277089176796747e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_d_inl0_hrd0/check.exe -p 2048 256 12 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.053655e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.901972e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.901972e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.065244e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.964476e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.964476e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     3.470906 sec
-     9,748,601,702      cycles                           #    2.805 GHz                    
-    18,987,159,627      instructions                     #    1.95  insn per cycle         
-       3.476589218 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1690) (512y:  181) (512z:    0)
+TOTAL       :     3.463334 sec
+     9,575,609,591      cycles                           #    2.761 GHz                    
+    18,944,249,093      instructions                     #    1.98  insn per cycle         
+       3.469928809 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1655) (512y:  161) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -176,26 +176,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282804e-02
-Avg ME (F77/C++)    = 1.2828039868165088E-002
-Relative difference = 1.0277089312025782e-08
+Avg ME (F77/C++)    = 1.2828039868165090E-002
+Relative difference = 1.0277089176796747e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_d_inl0_hrd0/check.exe -p 2048 256 12 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.791558e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.378176e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.378176e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.819790e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.476564e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.476564e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     3.917961 sec
-     8,615,779,869      cycles                           #    2.196 GHz                    
-    15,727,012,410      instructions                     #    1.83  insn per cycle         
-       3.924530110 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:  901) (512y:  154) (512z: 1258)
+TOTAL       :     3.875473 sec
+     8,194,000,405      cycles                           #    2.112 GHz                    
+    15,512,267,676      instructions                     #    1.89  insn per cycle         
+       3.882168596 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:  920) (512y:   59) (512z: 1220)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -203,8 +203,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282804e-02
-Avg ME (F77/C++)    = 1.2828039868165088E-002
-Relative difference = 1.0277089312025782e-08
+Avg ME (F77/C++)    = 1.2828039868165090E-002
+Relative difference = 1.0277089176796747e-08
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_d_inl0_hrd1.txt b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_d_inl0_hrd1.txt
index 0c748b5362..9238de7bbb 100644
--- a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_d_inl0_hrd1.txt
+++ b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_d_inl0_hrd1.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_d_inl0_hrd1'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum'
 
-DATE: 2023-11-24_14:28:26
+DATE: 2024-01-30_04:52:22
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_d_inl0_hrd1/gcheck.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 5.428371e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.291622e+08                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.195095e+08                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.433269e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.304294e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.211626e+08                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     0.698284 sec
-     2,831,920,965      cycles                           #    3.030 GHz                    
-     4,374,282,745      instructions                     #    1.54  insn per cycle         
-       1.019007238 seconds time elapsed
+TOTAL       :     0.708580 sec
+     2,678,035,833      cycles                           #    2.828 GHz                    
+     4,219,258,618      instructions                     #    1.58  insn per cycle         
+       1.025396427 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_d_inl0_hrd1/gcheck.exe -p 2048 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 154
@@ -72,20 +72,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_d_inl0_hrd1/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.203126e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.425801e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.425801e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.057712e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.240764e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.240764e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     5.626130 sec
-    17,496,349,043      cycles                           #    3.107 GHz                    
-    41,817,978,562      instructions                     #    2.39  insn per cycle         
-       5.639060633 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  392) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     6.363915 sec
+    18,420,155,453      cycles                           #    2.892 GHz                    
+    44,716,833,361      instructions                     #    2.43  insn per cycle         
+       6.376789264 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  486) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_d_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -99,20 +99,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_d_inl0_hrd1/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.748150e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.309667e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.309667e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.624136e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.147437e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.147437e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     3.992952 sec
-    12,464,968,903      cycles                           #    3.118 GHz                    
-    30,161,347,268      instructions                     #    2.42  insn per cycle         
-       4.011343884 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 1612) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     4.286124 sec
+    12,429,118,549      cycles                           #    2.897 GHz                    
+    30,107,231,858      instructions                     #    2.42  insn per cycle         
+       4.302706533 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 1569) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_d_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -126,20 +126,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_d_inl0_hrd1/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.129918e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.998050e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.998050e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.942189e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.705004e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.705004e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     3.345656 sec
-     9,985,760,194      cycles                           #    2.981 GHz                    
-    19,098,128,141      instructions                     #    1.91  insn per cycle         
-       3.367255257 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1931) (512y:    0) (512z:    0)
+TOTAL       :     3.656079 sec
+    10,127,428,804      cycles                           #    2.766 GHz                    
+    19,115,519,637      instructions                     #    1.89  insn per cycle         
+       3.673885868 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1902) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_d_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -153,20 +153,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_d_inl0_hrd1/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.214083e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.146132e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.146132e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.094903e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.039710e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.039710e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     3.227419 sec
-     9,634,388,874      cycles                           #    2.980 GHz                    
-    18,745,812,999      instructions                     #    1.95  insn per cycle         
-       3.242688774 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1662) (512y:  178) (512z:    0)
+TOTAL       :     3.417483 sec
+     9,477,381,758      cycles                           #    2.768 GHz                    
+    18,489,351,216      instructions                     #    1.95  insn per cycle         
+       3.434681568 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1576) (512y:  159) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_d_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -180,20 +180,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_d_inl0_hrd1/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.942471e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.616599e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.616599e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.183418e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.193735e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.193735e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     3.630500 sec
-     8,446,217,730      cycles                           #    2.323 GHz                    
-    15,603,353,222      instructions                     #    1.85  insn per cycle         
-       3.648277451 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:  887) (512y:  156) (512z: 1239)
+TOTAL       :     3.298580 sec
+     7,210,521,695      cycles                           #    2.182 GHz                    
+    13,864,693,183      instructions                     #    1.92  insn per cycle         
+       3.315590461 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:  818) (512y:   57) (512z:  898)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_d_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
diff --git a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_d_inl1_hrd0.txt b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_d_inl1_hrd0.txt
index 03ec5636b7..09e3552971 100644
--- a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_d_inl1_hrd0.txt
+++ b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_d_inl1_hrd0.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_d_inl1_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum'
 
-DATE: 2023-11-24_14:58:42
+DATE: 2024-01-30_05:33:56
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_d_inl1_hrd0/gcheck.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 6.462140e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.622718e+08                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.171485e+08                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.454720e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.590982e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.126095e+08                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     0.680938 sec
-     2,683,046,213      cycles                           #    2.937 GHz                    
-     4,147,768,853      instructions                     #    1.55  insn per cycle         
-       0.974174885 seconds time elapsed
+TOTAL       :     0.682889 sec
+     2,611,559,388      cycles                           #    2.831 GHz                    
+     3,986,840,129      instructions                     #    1.53  insn per cycle         
+       0.986209294 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_d_inl1_hrd0/gcheck.exe -p 2048 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 166
@@ -72,20 +72,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_d_inl1_hrd0/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.640951e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.107305e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.107305e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.350945e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.669369e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.669369e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     4.242929 sec
-    12,671,992,752      cycles                           #    2.983 GHz                    
-    32,514,493,053      instructions                     #    2.57  insn per cycle         
-       4.248758803 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  296) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     5.075739 sec
+    14,632,134,397      cycles                           #    2.880 GHz                    
+    36,697,212,873      instructions                     #    2.51  insn per cycle         
+       5.082665504 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  707) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_d_inl1_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -99,20 +99,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_d_inl1_hrd0/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.096760e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.006731e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.006731e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.975416e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.812212e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.812212e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     3.405789 sec
-    10,268,358,622      cycles                           #    3.010 GHz                    
-    24,473,723,768      instructions                     #    2.38  insn per cycle         
-       3.411753584 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 1251) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     3.599579 sec
+    10,391,716,980      cycles                           #    2.883 GHz                    
+    24,753,509,930      instructions                     #    2.38  insn per cycle         
+       3.606361950 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 2334) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_d_inl1_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -126,20 +126,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_d_inl1_hrd0/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.257873e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.333676e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.333676e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.206864e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.274609e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.274609e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     3.192822 sec
-     9,136,252,150      cycles                           #    2.858 GHz                    
-    16,922,200,539      instructions                     #    1.85  insn per cycle         
-       3.198278524 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1631) (512y:    0) (512z:    0)
+TOTAL       :     3.273737 sec
+     8,884,033,270      cycles                           #    2.722 GHz                    
+    16,960,441,009      instructions                     #    1.91  insn per cycle         
+       3.280558312 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1604) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_d_inl1_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -147,26 +147,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_d_inl1_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_d_inl1_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282804e-02
-Avg ME (F77/C++)    = 1.2828039868165088E-002
-Relative difference = 1.0277089312025782e-08
+Avg ME (F77/C++)    = 1.2828039868165090E-002
+Relative difference = 1.0277089176796747e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_d_inl1_hrd0/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.262662e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.358756e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.358756e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.436675e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.780065e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.780065e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     3.189189 sec
-     8,979,407,762      cycles                           #    2.811 GHz                    
-    16,334,341,152      instructions                     #    1.82  insn per cycle         
-       3.195303169 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1370) (512y:  139) (512z:    0)
+TOTAL       :     2.997375 sec
+     8,315,936,313      cycles                           #    2.769 GHz                    
+    16,298,181,743      instructions                     #    1.96  insn per cycle         
+       3.004046425 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2403) (512y:  292) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_d_inl1_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -174,26 +174,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_d_inl1_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_d_inl1_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282804e-02
-Avg ME (F77/C++)    = 1.2828039868165088E-002
-Relative difference = 1.0277089312025782e-08
+Avg ME (F77/C++)    = 1.2828039868165090E-002
+Relative difference = 1.0277089176796747e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_d_inl1_hrd0/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.031223e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.819033e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.819033e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.987391e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.794180e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.794180e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     3.500399 sec
-     7,927,835,353      cycles                           #    2.263 GHz                    
-    14,582,334,434      instructions                     #    1.84  insn per cycle         
-       3.505946160 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1015) (512y:  158) (512z:  955)
+TOTAL       :     3.583817 sec
+     7,670,874,044      cycles                           #    2.137 GHz                    
+    14,352,448,248      instructions                     #    1.87  insn per cycle         
+       3.590538974 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:  892) (512y:   63) (512z:  975)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_d_inl1_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -201,8 +201,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_d_inl1_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_d_inl1_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282804e-02
-Avg ME (F77/C++)    = 1.2828039868165088E-002
-Relative difference = 1.0277089312025782e-08
+Avg ME (F77/C++)    = 1.2828039868165090E-002
+Relative difference = 1.0277089176796747e-08
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_d_inl1_hrd1.txt b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_d_inl1_hrd1.txt
index ed59ed96d8..508008a0c5 100644
--- a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_d_inl1_hrd1.txt
+++ b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_d_inl1_hrd1.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_d_inl1_hrd1'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum'
 
-DATE: 2023-11-24_14:59:12
+DATE: 2024-01-30_05:34:28
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_d_inl1_hrd1/gcheck.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 6.466236e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.600756e+08                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.174175e+08                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.464301e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.594213e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.177261e+08                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     0.677095 sec
-     2,664,495,162      cycles                           #    2.925 GHz                    
-     4,015,268,500      instructions                     #    1.51  insn per cycle         
-       0.968065572 seconds time elapsed
+TOTAL       :     0.680513 sec
+     2,594,214,158      cycles                           #    2.833 GHz                    
+     3,992,420,158      instructions                     #    1.54  insn per cycle         
+       0.978034885 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_d_inl1_hrd1/gcheck.exe -p 2048 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 154
@@ -72,20 +72,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_d_inl1_hrd1/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.188114e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.096436e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.096436e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.895468e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.581482e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.581482e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     3.282515 sec
-     9,850,281,434      cycles                           #    2.996 GHz                    
-    25,394,191,492      instructions                     #    2.58  insn per cycle         
-       3.288415370 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  249) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     3.738704 sec
+    10,794,188,443      cycles                           #    2.885 GHz                    
+    28,356,720,092      instructions                     #    2.63  insn per cycle         
+       3.745371478 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  600) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_d_inl1_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -99,20 +99,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_d_inl1_hrd1/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.427078e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.740655e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.740655e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.231818e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.360148e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.360148e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     3.000969 sec
-     8,962,872,354      cycles                           #    2.983 GHz                    
-    21,484,742,690      instructions                     #    2.40  insn per cycle         
-       3.006873984 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 1119) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     3.232648 sec
+     9,331,358,518      cycles                           #    2.882 GHz                    
+    21,587,159,141      instructions                     #    2.31  insn per cycle         
+       3.239331570 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 2117) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_d_inl1_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -126,20 +126,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_d_inl1_hrd1/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.437736e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.701040e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.701040e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.406271e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.696326e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.696326e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     2.982872 sec
-     8,650,300,725      cycles                           #    2.896 GHz                    
-    15,811,446,227      instructions                     #    1.83  insn per cycle         
-       2.988774005 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1494) (512y:    0) (512z:    0)
+TOTAL       :     3.030114 sec
+     8,381,289,955      cycles                           #    2.761 GHz                    
+    15,943,872,727      instructions                     #    1.90  insn per cycle         
+       3.036686774 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1497) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_d_inl1_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -153,20 +153,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_d_inl1_hrd1/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.506471e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.822748e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.822748e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.611770e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.211566e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.211566e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     2.907315 sec
-     8,443,660,993      cycles                           #    2.899 GHz                    
-    15,514,495,232      instructions                     #    1.84  insn per cycle         
-       2.913215721 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1268) (512y:  139) (512z:    0)
+TOTAL       :     2.823652 sec
+     7,834,743,570      cycles                           #    2.770 GHz                    
+    15,370,444,400      instructions                     #    1.96  insn per cycle         
+       2.830226684 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2179) (512y:  307) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_d_inl1_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -180,20 +180,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_d_inl1_hrd1/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.094032e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.959241e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.959241e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.110110e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.044152e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.044152e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     3.415670 sec
-     7,621,535,351      cycles                           #    2.228 GHz                    
-    14,285,430,897      instructions                     #    1.87  insn per cycle         
-       3.421826468 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1041) (512y:  164) (512z:  874)
+TOTAL       :     3.399029 sec
+     7,342,854,469      cycles                           #    2.157 GHz                    
+    13,880,932,107      instructions                     #    1.89  insn per cycle         
+       3.405583219 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:  853) (512y:   69) (512z:  905)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_d_inl1_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
diff --git a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_f_inl0_hrd0.txt b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_f_inl0_hrd0.txt
index 5eeb6e403e..30054d0a8f 100644
--- a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_f_inl0_hrd0.txt
+++ b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_f_inl0_hrd0.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_f_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum'
 
-DATE: 2023-11-24_14:28:58
+DATE: 2024-01-30_04:52:57
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd0/gcheck.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 1.089717e+08                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.081644e+09                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.286206e+09                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.089125e+08                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.083340e+09                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.291553e+09                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371687e-02 +- 3.270220e-06 )  GeV^0
-TOTAL       :     0.586552 sec
-     2,420,869,304      cycles                           #    2.991 GHz                    
-     3,828,073,747      instructions                     #    1.58  insn per cycle         
-       0.888297566 seconds time elapsed
+TOTAL       :     0.592260 sec
+     2,336,196,912      cycles                           #    2.833 GHz                    
+     3,633,132,034      instructions                     #    1.56  insn per cycle         
+       0.902800684 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd0/gcheck.exe -p 2048 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 117
@@ -72,20 +72,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd0/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.174500e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.392497e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.392497e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.035118e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.220346e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.220346e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371707e-02 +- 3.270376e-06 )  GeV^0
-TOTAL       :     5.713477 sec
-    17,800,790,292      cycles                           #    3.113 GHz                    
-    43,512,722,868      instructions                     #    2.44  insn per cycle         
-       5.725545534 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  431) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     6.458158 sec
+    18,623,778,658      cycles                           #    2.882 GHz                    
+    47,047,597,520      instructions                     #    2.53  insn per cycle         
+       6.468376899 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  542) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -93,26 +93,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282804e-02
-Avg ME (F77/C++)    = 1.2828039854866802E-002
-Relative difference = 1.1313746984080878e-08
+Avg ME (F77/C++)    = 1.2828039441956207E-002
+Relative difference = 4.35018750695023e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_f_inl0_hrd0/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.407559e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.694429e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.694429e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.220597e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.402817e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.402817e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371707e-02 +- 3.270376e-06 )  GeV^0
-TOTAL       :     2.969731 sec
-     9,247,230,421      cycles                           #    3.108 GHz                    
-    21,907,456,003      instructions                     #    2.37  insn per cycle         
-       2.986090115 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 1938) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     3.207438 sec
+     9,259,856,985      cycles                           #    2.882 GHz                    
+    22,093,069,841      instructions                     #    2.39  insn per cycle         
+       3.223491423 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 1883) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -126,20 +126,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_f_inl0_hrd0/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.603940e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.990614e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.990614e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.440699e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.781387e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.781387e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270341e-06 )  GeV^0
-TOTAL       :     2.767987 sec
-     8,299,418,318      cycles                           #    2.993 GHz                    
-    15,591,106,391      instructions                     #    1.88  insn per cycle         
-       2.786468252 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2596) (512y:    0) (512z:    0)
+TOTAL       :     2.957121 sec
+     8,193,990,799      cycles                           #    2.766 GHz                    
+    15,625,791,555      instructions                     #    1.91  insn per cycle         
+       2.973833384 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2619) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -153,20 +153,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_f_inl0_hrd0/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.627112e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.068065e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.068065e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.532783e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.026282e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.026282e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270341e-06 )  GeV^0
-TOTAL       :     2.741024 sec
-     8,236,041,235      cycles                           #    2.999 GHz                    
-    15,428,889,211      instructions                     #    1.87  insn per cycle         
-       2.759001089 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2496) (512y:    9) (512z:    0)
+TOTAL       :     2.864857 sec
+     7,877,312,491      cycles                           #    2.746 GHz                    
+    15,298,553,606      instructions                     #    1.94  insn per cycle         
+       2.880238416 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2414) (512y:   13) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -180,20 +180,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_f_inl0_hrd0/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[16] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.613778e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.027225e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.027225e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.515538e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.925634e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.925634e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270342e-06 )  GeV^0
-TOTAL       :     2.766843 sec
-     6,634,803,505      cycles                           #    2.394 GHz                    
-    12,864,482,335      instructions                     #    1.94  insn per cycle         
-       2.786287293 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1736) (512y:   17) (512z: 1439)
+TOTAL       :     2.878384 sec
+     6,411,016,127      cycles                           #    2.223 GHz                    
+    12,624,518,195      instructions                     #    1.97  insn per cycle         
+       2.897065980 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1615) (512y:   12) (512z: 1404)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -201,8 +201,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282805e-02
-Avg ME (F77/C++)    = 1.2828052585973637E-002
-Relative difference = 2.0158743040564767e-07
+Avg ME (F77/C++)    = 1.2828052589611616E-002
+Relative difference = 2.0187102602673518e-07
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_f_inl0_hrd0_bridge.txt b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_f_inl0_hrd0_bridge.txt
index 77f863e39f..cb0960cef7 100644
--- a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_f_inl0_hrd0_bridge.txt
+++ b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_f_inl0_hrd0_bridge.txt
@@ -41,7 +41,7 @@ CUDACPP_BUILDDIR='build.512z_f_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum'
 
-DATE: 2023-11-24_15:10:24
+DATE: 2024-01-30_05:46:05
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
@@ -51,17 +51,17 @@ WARNING! Bridge selected: cannot use RamboDevice, will use RamboHost
 WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
 WARNING! Instantiate device Bridge (nevt=524288, gpublocks=2048, gputhreads=256, gpublocks*gputhreads=524288)
 WARNING! Set grid in Bridge (nevt=524288, gpublocks=2048, gputhreads=256, gpublocks*gputhreads=524288)
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 7.062157e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 6.459463e+07                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 6.459463e+07                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.896245e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.389243e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.389243e+07                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371710e-02 +- 3.270389e-06 )  GeV^0
-TOTAL       :     1.708612 sec
-     5,739,082,619      cycles                           #    2.966 GHz                    
-    10,314,110,992      instructions                     #    1.80  insn per cycle         
-       1.994013421 seconds time elapsed
+TOTAL       :     1.734031 sec
+     5,668,072,364      cycles                           #    2.868 GHz                    
+    10,146,395,921      instructions                     #    1.79  insn per cycle         
+       2.033339529 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd0/gcheck.exe -p 2048 256 1 --bridge
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Bridge selected: cannot use RamboDevice, will use RamboHost
@@ -81,20 +81,20 @@ OK (relative difference <= 5E-3)
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd0/check.exe -p 2048 256 12 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=524288)
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.106417e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.307176e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.307176e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.023723e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.199962e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.199962e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371707e-02 +- 3.270376e-06 )  GeV^0
-TOTAL       :     6.159436 sec
-    18,484,427,111      cycles                           #    2.999 GHz                    
-    43,663,224,156      instructions                     #    2.36  insn per cycle         
-       6.165948155 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  431) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     6.631109 sec
+    19,198,970,802      cycles                           #    2.893 GHz                    
+    47,195,604,267      instructions                     #    2.46  insn per cycle         
+       6.638520301 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  542) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -102,27 +102,27 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282804e-02
-Avg ME (F77/C++)    = 1.2828039854866802E-002
-Relative difference = 1.1313746984080878e-08
+Avg ME (F77/C++)    = 1.2828039441956207E-002
+Relative difference = 4.35018750695023e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_f_inl0_hrd0/check.exe -p 2048 256 12 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=524288)
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/sse4+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.217277e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.320268e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.320268e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.130711e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.183569e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.183569e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371707e-02 +- 3.270376e-06 )  GeV^0
-TOTAL       :     3.327491 sec
-     9,999,841,208      cycles                           #    3.000 GHz                    
-    23,242,551,966      instructions                     #    2.32  insn per cycle         
-       3.334092044 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 1938) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     3.452422 sec
+     9,989,387,225      cycles                           #    2.889 GHz                    
+    23,431,077,272      instructions                     #    2.35  insn per cycle         
+       3.459894158 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 1883) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -137,20 +137,20 @@ OK (relative difference <= 5E-3)
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_f_inl0_hrd0/check.exe -p 2048 256 12 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=524288)
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/avx2+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.386544e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.570204e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.570204e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.341081e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.547294e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.547294e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270341e-06 )  GeV^0
-TOTAL       :     3.123723 sec
-     9,026,914,064      cycles                           #    2.885 GHz                    
-    16,711,011,073      instructions                     #    1.85  insn per cycle         
-       3.130206343 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2596) (512y:    0) (512z:    0)
+TOTAL       :     3.196012 sec
+     8,906,176,925      cycles                           #    2.782 GHz                    
+    16,751,991,837      instructions                     #    1.88  insn per cycle         
+       3.203321936 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2619) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -165,20 +165,20 @@ OK (relative difference <= 5E-3)
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_f_inl0_hrd0/check.exe -p 2048 256 12 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=524288)
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/512y+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.374379e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.579230e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.579230e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.434021e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.786427e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.786427e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270341e-06 )  GeV^0
-TOTAL       :     3.141266 sec
-     8,951,033,856      cycles                           #    2.844 GHz                    
-    16,554,473,732      instructions                     #    1.85  insn per cycle         
-       3.147725016 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2496) (512y:    9) (512z:    0)
+TOTAL       :     3.093664 sec
+     8,635,370,178      cycles                           #    2.786 GHz                    
+    16,424,138,356      instructions                     #    1.90  insn per cycle         
+       3.101132741 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2414) (512y:   13) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -193,20 +193,20 @@ OK (relative difference <= 5E-3)
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_f_inl0_hrd0/check.exe -p 2048 256 12 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=524288)
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/512z+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[16] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.422417e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.623038e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.623038e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.383314e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.611676e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.611676e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270342e-06 )  GeV^0
-TOTAL       :     3.084070 sec
-     7,371,897,160      cycles                           #    2.386 GHz                    
-    14,070,343,022      instructions                     #    1.91  insn per cycle         
-       3.090398779 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1736) (512y:   17) (512z: 1439)
+TOTAL       :     3.145258 sec
+     7,151,980,153      cycles                           #    2.270 GHz                    
+    13,850,467,115      instructions                     #    1.94  insn per cycle         
+       3.152590479 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1615) (512y:   12) (512z: 1404)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -214,8 +214,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282805e-02
-Avg ME (F77/C++)    = 1.2828052585973637E-002
-Relative difference = 2.0158743040564767e-07
+Avg ME (F77/C++)    = 1.2828052589611616E-002
+Relative difference = 2.0187102602673518e-07
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_f_inl0_hrd0_common.txt b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_f_inl0_hrd0_common.txt
index 94aeb34d20..26c818590d 100644
--- a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_f_inl0_hrd0_common.txt
+++ b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_f_inl0_hrd0_common.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_f_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum'
 
-DATE: 2023-11-24_15:23:46
+DATE: 2024-01-30_05:59:57
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd0/gcheck.exe -p 2048 256 12 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:COMMON+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 1.278888e+08                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.156496e+09                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.253231e+09                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.303596e+08                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.175288e+09                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.243996e+09                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371863e-02 +- 3.269951e-06 )  GeV^0
-TOTAL       :     1.221248 sec
-     4,164,953,848      cycles                           #    2.869 GHz                    
-     6,574,190,110      instructions                     #    1.58  insn per cycle         
-       1.509194129 seconds time elapsed
+TOTAL       :     1.207260 sec
+     4,082,591,214      cycles                           #    2.858 GHz                    
+     6,515,356,659      instructions                     #    1.60  insn per cycle         
+       1.486873600 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd0/gcheck.exe -p 2048 256 1 --common
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 117
@@ -72,20 +72,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd0/check.exe -p 2048 256 12 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:COMMON+RMBHST+MESHST/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.130054e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.340325e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.340325e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.039099e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.222240e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.222240e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371887e-02 +- 3.270267e-06 )  GeV^0
-TOTAL       :     6.276129 sec
-    18,853,916,186      cycles                           #    3.003 GHz                    
-    43,696,977,700      instructions                     #    2.32  insn per cycle         
-       6.281870449 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  431) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     6.777770 sec
+    19,569,392,860      cycles                           #    2.885 GHz                    
+    47,229,099,277      instructions                     #    2.41  insn per cycle         
+       6.784024049 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  542) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -93,26 +93,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282804e-02
-Avg ME (F77/C++)    = 1.2828039854866802E-002
-Relative difference = 1.1313746984080878e-08
+Avg ME (F77/C++)    = 1.2828039441956207E-002
+Relative difference = 4.35018750695023e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_f_inl0_hrd0/check.exe -p 2048 256 12 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:COMMON+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.294770e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.514019e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.514019e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.224011e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.394362e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.394362e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371887e-02 +- 3.270266e-06 )  GeV^0
-TOTAL       :     3.444527 sec
-    10,256,624,459      cycles                           #    2.974 GHz                    
-    21,988,381,288      instructions                     #    2.14  insn per cycle         
-       3.450359347 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 1938) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     3.543713 sec
+    10,250,573,649      cycles                           #    2.890 GHz                    
+    22,173,775,935      instructions                     #    2.16  insn per cycle         
+       3.550219999 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 1883) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -126,20 +126,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_f_inl0_hrd0/check.exe -p 2048 256 12 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:COMMON+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.493233e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.822049e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.822049e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.458663e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.813529e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.813529e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371885e-02 +- 3.270112e-06 )  GeV^0
-TOTAL       :     3.224970 sec
-     9,324,398,209      cycles                           #    2.887 GHz                    
-    15,502,220,245      instructions                     #    1.66  insn per cycle         
-       3.230595711 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2596) (512y:    0) (512z:    0)
+TOTAL       :     3.280080 sec
+     9,161,776,432      cycles                           #    2.789 GHz                    
+    15,536,168,479      instructions                     #    1.70  insn per cycle         
+       3.286291256 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2619) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -153,20 +153,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_f_inl0_hrd0/check.exe -p 2048 256 12 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:COMMON+RMBHST+MESHST/512y+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.511667e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.890072e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.890072e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.554649e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.077981e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.077981e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371885e-02 +- 3.270112e-06 )  GeV^0
-TOTAL       :     3.207482 sec
-     9,275,437,455      cycles                           #    2.888 GHz                    
-    15,144,147,371      instructions                     #    1.63  insn per cycle         
-       3.213077967 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2496) (512y:    9) (512z:    0)
+TOTAL       :     3.189150 sec
+     8,891,496,493      cycles                           #    2.784 GHz                    
+    15,006,164,122      instructions                     #    1.69  insn per cycle         
+       3.195486341 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2414) (512y:   13) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -180,20 +180,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_f_inl0_hrd0/check.exe -p 2048 256 12 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:COMMON+RMBHST+MESHST/512z+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[16] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.556471e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.938543e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.938543e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.516232e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.934012e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.934012e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371885e-02 +- 3.270112e-06 )  GeV^0
-TOTAL       :     3.164113 sec
-     7,669,667,119      cycles                           #    2.420 GHz                    
-    12,572,972,519      instructions                     #    1.64  insn per cycle         
-       3.169922598 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1736) (512y:   17) (512z: 1439)
+TOTAL       :     3.229540 sec
+     7,432,998,054      cycles                           #    2.298 GHz                    
+    12,333,053,960      instructions                     #    1.66  insn per cycle         
+       3.235962697 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1615) (512y:   12) (512z: 1404)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -201,8 +201,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282805e-02
-Avg ME (F77/C++)    = 1.2828052585973637E-002
-Relative difference = 2.0158743040564767e-07
+Avg ME (F77/C++)    = 1.2828052589611616E-002
+Relative difference = 2.0187102602673518e-07
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_f_inl0_hrd0_curhst.txt b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_f_inl0_hrd0_curhst.txt
index 62bb3359c2..90d7f62db4 100644
--- a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_f_inl0_hrd0_curhst.txt
+++ b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_f_inl0_hrd0_curhst.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_f_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum'
 
-DATE: 2023-11-24_15:20:27
+DATE: 2024-01-30_05:56:29
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd0/gcheck.exe -p 2048 256 12 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 1.313617e+08                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.192806e+09                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.298454e+09                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.305141e+08                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.181296e+09                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.274552e+09                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371687e-02 +- 3.270220e-06 )  GeV^0
-TOTAL       :     0.858273 sec
-     3,147,417,725      cycles                           #    2.925 GHz                    
-     6,398,055,131      instructions                     #    2.03  insn per cycle         
-       1.135101998 seconds time elapsed
+TOTAL       :     0.867173 sec
+     3,085,877,327      cycles                           #    2.830 GHz                    
+     6,333,420,740      instructions                     #    2.05  insn per cycle         
+       1.147827940 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd0/gcheck.exe -p 2048 256 1 --curhst
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 117
@@ -72,20 +72,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd0/check.exe -p 2048 256 12 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.125924e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.336940e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.336940e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.039832e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.222763e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.222763e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371707e-02 +- 3.270376e-06 )  GeV^0
-TOTAL       :     5.957055 sec
-    17,830,147,444      cycles                           #    2.991 GHz                    
-    43,513,458,674      instructions                     #    2.44  insn per cycle         
-       5.962276638 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  431) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     6.429819 sec
+    18,561,263,651      cycles                           #    2.885 GHz                    
+    47,048,334,209      instructions                     #    2.53  insn per cycle         
+       6.436326918 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  542) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -93,26 +93,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282804e-02
-Avg ME (F77/C++)    = 1.2828039854866802E-002
-Relative difference = 1.1313746984080878e-08
+Avg ME (F77/C++)    = 1.2828039441956207E-002
+Relative difference = 4.35018750695023e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_f_inl0_hrd0/check.exe -p 2048 256 12 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.326147e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.566110e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.566110e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.222730e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.393980e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.393980e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371707e-02 +- 3.270376e-06 )  GeV^0
-TOTAL       :     3.071304 sec
-     9,223,399,519      cycles                           #    2.999 GHz                    
-    21,906,628,505      instructions                     #    2.38  insn per cycle         
-       3.076527254 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 1938) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     3.203768 sec
+     9,238,443,218      cycles                           #    2.879 GHz                    
+    22,092,244,938      instructions                     #    2.39  insn per cycle         
+       3.210105048 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 1883) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -126,20 +126,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_f_inl0_hrd0/check.exe -p 2048 256 12 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.495233e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.826120e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.826120e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.418509e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.733909e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.733909e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270341e-06 )  GeV^0
-TOTAL       :     2.884227 sec
-     8,304,746,805      cycles                           #    2.875 GHz                    
-    15,590,862,530      instructions                     #    1.88  insn per cycle         
-       2.889527315 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2596) (512y:    0) (512z:    0)
+TOTAL       :     2.982066 sec
+     8,185,679,734      cycles                           #    2.740 GHz                    
+    15,625,107,028      instructions                     #    1.91  insn per cycle         
+       2.988278371 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2619) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -153,20 +153,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_f_inl0_hrd0/check.exe -p 2048 256 12 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.543511e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.915375e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.915375e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.558846e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.085053e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.085053e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270341e-06 )  GeV^0
-TOTAL       :     2.832113 sec
-     8,194,558,683      cycles                           #    2.889 GHz                    
-    15,428,318,410      instructions                     #    1.88  insn per cycle         
-       2.837422327 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2496) (512y:    9) (512z:    0)
+TOTAL       :     2.831486 sec
+     7,894,514,850      cycles                           #    2.783 GHz                    
+    15,296,644,493      instructions                     #    1.94  insn per cycle         
+       2.837958999 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2414) (512y:   13) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -180,20 +180,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_f_inl0_hrd0/check.exe -p 2048 256 12 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[16] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.541346e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.894800e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.894800e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.525394e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.942507e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.942507e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270342e-06 )  GeV^0
-TOTAL       :     2.839095 sec
-     6,610,683,902      cycles                           #    2.326 GHz                    
-    12,864,156,943      instructions                     #    1.95  insn per cycle         
-       2.844410626 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1736) (512y:   17) (512z: 1439)
+TOTAL       :     2.867904 sec
+     6,407,267,092      cycles                           #    2.230 GHz                    
+    12,623,570,741      instructions                     #    1.97  insn per cycle         
+       2.874115235 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1615) (512y:   12) (512z: 1404)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -201,8 +201,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282805e-02
-Avg ME (F77/C++)    = 1.2828052585973637E-002
-Relative difference = 2.0158743040564767e-07
+Avg ME (F77/C++)    = 1.2828052589611616E-002
+Relative difference = 2.0187102602673518e-07
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_f_inl0_hrd0_rmbhst.txt b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_f_inl0_hrd0_rmbhst.txt
index 2afd300dfd..91671fa84d 100644
--- a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_f_inl0_hrd0_rmbhst.txt
+++ b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_f_inl0_hrd0_rmbhst.txt
@@ -41,24 +41,24 @@ CUDACPP_BUILDDIR='build.512z_f_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum'
 
-DATE: 2023-11-24_15:17:07
+DATE: 2024-01-30_05:53:04
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd0/gcheck.exe -p 2048 256 12 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 8.991991e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.144574e+09                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.162617e+09                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 8.674927e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.142204e+09                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.126513e+09                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371710e-02 +- 3.270389e-06 )  GeV^0
-TOTAL       :     1.488342 sec
-     5,031,725,760      cycles                           #    2.945 GHz                    
-     9,217,170,327      instructions                     #    1.83  insn per cycle         
-       1.765546553 seconds time elapsed
+TOTAL       :     1.522177 sec
+     5,014,296,377      cycles                           #    2.858 GHz                    
+     9,135,258,914      instructions                     #    1.82  insn per cycle         
+       1.813578794 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd0/gcheck.exe -p 2048 256 1 --rmbhst
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
@@ -74,20 +74,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd0/check.exe -p 2048 256 12 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.131966e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.341693e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.341693e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.043183e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.226572e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.226572e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371707e-02 +- 3.270376e-06 )  GeV^0
-TOTAL       :     5.927034 sec
-    17,813,349,259      cycles                           #    3.004 GHz                    
-    43,513,436,757      instructions                     #    2.44  insn per cycle         
-       5.932279783 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  431) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     6.408092 sec
+    18,567,709,150      cycles                           #    2.896 GHz                    
+    47,047,255,730      instructions                     #    2.53  insn per cycle         
+       6.414419955 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  542) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -95,26 +95,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282804e-02
-Avg ME (F77/C++)    = 1.2828039854866802E-002
-Relative difference = 1.1313746984080878e-08
+Avg ME (F77/C++)    = 1.2828039441956207E-002
+Relative difference = 4.35018750695023e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_f_inl0_hrd0/check.exe -p 2048 256 12 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.333420e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.574664e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.574664e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.231919e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.414648e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.414648e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371707e-02 +- 3.270376e-06 )  GeV^0
-TOTAL       :     3.064718 sec
-     9,236,786,325      cycles                           #    3.009 GHz                    
-    21,906,852,468      instructions                     #    2.37  insn per cycle         
-       3.070020479 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 1938) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     3.191678 sec
+     9,246,166,536      cycles                           #    2.894 GHz                    
+    22,093,449,321      instructions                     #    2.39  insn per cycle         
+       3.197919261 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 1883) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -128,20 +128,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_f_inl0_hrd0/check.exe -p 2048 256 12 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.506026e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.849776e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.849776e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.455778e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.806689e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.806689e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270341e-06 )  GeV^0
-TOTAL       :     2.874113 sec
-     8,320,499,209      cycles                           #    2.891 GHz                    
-    15,592,249,654      instructions                     #    1.87  insn per cycle         
-       2.879517495 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2596) (512y:    0) (512z:    0)
+TOTAL       :     2.938294 sec
+     8,179,243,825      cycles                           #    2.779 GHz                    
+    15,624,915,954      instructions                     #    1.91  insn per cycle         
+       2.944456642 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2619) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -155,20 +155,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_f_inl0_hrd0/check.exe -p 2048 256 12 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.509013e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.864492e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.864492e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.562111e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.082808e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.082808e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270341e-06 )  GeV^0
-TOTAL       :     2.870737 sec
-     8,241,489,655      cycles                           #    2.866 GHz                    
-    15,430,280,733      instructions                     #    1.87  insn per cycle         
-       2.876296254 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2496) (512y:    9) (512z:    0)
+TOTAL       :     2.828979 sec
+     7,880,998,863      cycles                           #    2.781 GHz                    
+    15,296,291,599      instructions                     #    1.94  insn per cycle         
+       2.835269816 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2414) (512y:   13) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -182,20 +182,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_f_inl0_hrd0/check.exe -p 2048 256 12 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[16] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.501831e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.838293e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.838293e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.528595e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.951135e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.951135e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270342e-06 )  GeV^0
-TOTAL       :     2.887603 sec
-     6,626,308,640      cycles                           #    2.293 GHz                    
-    12,864,606,095      instructions                     #    1.94  insn per cycle         
-       2.893234728 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1736) (512y:   17) (512z: 1439)
+TOTAL       :     2.864434 sec
+     6,402,503,393      cycles                           #    2.232 GHz                    
+    12,623,594,501      instructions                     #    1.97  insn per cycle         
+       2.870718249 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1615) (512y:   12) (512z: 1404)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -203,8 +203,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282805e-02
-Avg ME (F77/C++)    = 1.2828052585973637E-002
-Relative difference = 2.0158743040564767e-07
+Avg ME (F77/C++)    = 1.2828052589611616E-002
+Relative difference = 2.0187102602673518e-07
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_f_inl0_hrd1.txt b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_f_inl0_hrd1.txt
index efc03ad0b2..cc5700bb60 100644
--- a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_f_inl0_hrd1.txt
+++ b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_f_inl0_hrd1.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_f_inl0_hrd1'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum'
 
-DATE: 2023-11-24_14:29:28
+DATE: 2024-01-30_04:53:28
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd1/gcheck.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 1.089951e+08                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.092950e+09                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.329798e+09                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.091291e+08                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.093645e+09                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.338052e+09                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371687e-02 +- 3.270220e-06 )  GeV^0
-TOTAL       :     0.581630 sec
-     2,429,919,567      cycles                           #    3.010 GHz                    
-     3,742,142,294      instructions                     #    1.54  insn per cycle         
-       0.889170895 seconds time elapsed
+TOTAL       :     0.585723 sec
+     2,310,991,948      cycles                           #    2.835 GHz                    
+     3,567,792,024      instructions                     #    1.54  insn per cycle         
+       0.889438316 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd1/gcheck.exe -p 2048 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 95
@@ -72,20 +72,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd1/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.231336e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.478764e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.478764e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.092050e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.295990e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.295990e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371707e-02 +- 3.270376e-06 )  GeV^0
-TOTAL       :     5.469982 sec
-    16,730,831,486      cycles                           #    3.056 GHz                    
-    41,270,909,044      instructions                     #    2.47  insn per cycle         
-       5.482091928 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  375) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     6.138113 sec
+    17,749,278,373      cycles                           #    2.890 GHz                    
+    43,890,075,557      instructions                     #    2.47  insn per cycle         
+       6.149965364 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  467) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -93,26 +93,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl0_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282804e-02
-Avg ME (F77/C++)    = 1.2828039854866802E-002
-Relative difference = 1.1313746984080878e-08
+Avg ME (F77/C++)    = 1.2828039441956207E-002
+Relative difference = 4.35018750695023e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_f_inl0_hrd1/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.472788e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.845298e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.845298e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.281832e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.528866e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.528866e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371707e-02 +- 3.270376e-06 )  GeV^0
-TOTAL       :     2.902455 sec
-     8,999,277,217      cycles                           #    3.096 GHz                    
-    21,212,156,929      instructions                     #    2.36  insn per cycle         
-       2.920254424 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 1843) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     3.131918 sec
+     9,063,997,030      cycles                           #    2.890 GHz                    
+    21,583,444,087      instructions                     #    2.38  insn per cycle         
+       3.172631085 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 1827) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_f_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -126,20 +126,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_f_inl0_hrd1/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.606896e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.017656e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.017656e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.471429e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.850830e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.850830e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270341e-06 )  GeV^0
-TOTAL       :     2.765803 sec
-     8,250,996,001      cycles                           #    2.978 GHz                    
-    15,426,023,835      instructions                     #    1.87  insn per cycle         
-       2.784276443 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2537) (512y:    0) (512z:    0)
+TOTAL       :     2.922404 sec
+     8,130,490,307      cycles                           #    2.776 GHz                    
+    15,429,884,484      instructions                     #    1.90  insn per cycle         
+       2.941222784 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2542) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_f_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -153,20 +153,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_f_inl0_hrd1/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.659785e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.147153e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.147153e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.565898e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.093653e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.093653e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270341e-06 )  GeV^0
-TOTAL       :     2.718129 sec
-     8,116,789,845      cycles                           #    2.981 GHz                    
-    15,238,834,705      instructions                     #    1.88  insn per cycle         
-       2.738443955 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2423) (512y:    8) (512z:    0)
+TOTAL       :     2.826189 sec
+     7,861,694,964      cycles                           #    2.776 GHz                    
+    15,087,354,653      instructions                     #    1.92  insn per cycle         
+       2.844638276 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2323) (512y:   15) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_f_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -180,20 +180,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_f_inl0_hrd1/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[16] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.665676e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.125551e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.125551e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.637184e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.244046e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.244046e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270342e-06 )  GeV^0
-TOTAL       :     2.716332 sec
-     6,606,196,349      cycles                           #    2.427 GHz                    
-    12,842,263,926      instructions                     #    1.94  insn per cycle         
-       2.735947842 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1706) (512y:   18) (512z: 1427)
+TOTAL       :     2.766988 sec
+     6,178,543,208      cycles                           #    2.228 GHz                    
+    12,245,131,195      instructions                     #    1.98  insn per cycle         
+       2.787936795 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1538) (512y:    8) (512z: 1258)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_f_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -201,8 +201,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_f_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_f_inl0_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282805e-02
-Avg ME (F77/C++)    = 1.2828052564145764E-002
-Relative difference = 1.9988585667912256e-07
+Avg ME (F77/C++)    = 1.2828052431359538E-002
+Relative difference = 1.895346165094282e-07
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_f_inl1_hrd0.txt b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_f_inl1_hrd0.txt
index 6aac347ebc..df038945e7 100644
--- a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_f_inl1_hrd0.txt
+++ b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_f_inl1_hrd0.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_f_inl1_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum'
 
-DATE: 2023-11-24_14:59:40
+DATE: 2024-01-30_05:34:57
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl1_hrd0/gcheck.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 1.268443e+08                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.159820e+09                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.242861e+09                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.293279e+08                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.189438e+09                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.292426e+09                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371687e-02 +- 3.270220e-06 )  GeV^0
-TOTAL       :     0.585151 sec
-     2,376,619,364      cycles                           #    2.930 GHz                    
-     3,674,969,576      instructions                     #    1.55  insn per cycle         
-       0.869620998 seconds time elapsed
+TOTAL       :     0.574531 sec
+     2,278,103,742      cycles                           #    2.838 GHz                    
+     3,559,192,155      instructions                     #    1.56  insn per cycle         
+       0.862169679 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl1_hrd0/gcheck.exe -p 2048 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 117
@@ -72,20 +72,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl1_hrd0/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.681014e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.190697e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.190697e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.401205e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.755017e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.755017e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371707e-02 +- 3.270376e-06 )  GeV^0
-TOTAL       :     4.111208 sec
-    12,184,760,501      cycles                           #    2.960 GHz                    
-    32,432,932,194      instructions                     #    2.66  insn per cycle         
-       4.116807187 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  312) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     4.863284 sec
+    13,757,936,316      cycles                           #    2.826 GHz                    
+    37,850,126,745      instructions                     #    2.75  insn per cycle         
+       4.870249581 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  833) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl1_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -93,26 +93,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl1_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl1_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282804e-02
-Avg ME (F77/C++)    = 1.2828039840314887E-002
-Relative difference = 1.244813035273009e-08
+Avg ME (F77/C++)    = 1.2828039414671366E-002
+Relative difference = 4.562884388571957e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_f_inl1_hrd0/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.746012e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.644193e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.644193e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.651233e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.514070e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.514070e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371707e-02 +- 3.270376e-06 )  GeV^0
-TOTAL       :     2.661942 sec
-     8,002,880,266      cycles                           #    3.001 GHz                    
-    18,657,307,287      instructions                     #    2.33  insn per cycle         
-       2.667590652 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 1555) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     2.748031 sec
+     7,929,384,882      cycles                           #    2.881 GHz                    
+    18,604,713,730      instructions                     #    2.35  insn per cycle         
+       2.754502860 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 2808) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_f_inl1_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -120,26 +120,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_f_inl1_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_f_inl1_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282804e-02
-Avg ME (F77/C++)    = 1.2828039283704129E-002
-Relative difference = 5.583829420356249e-08
+Avg ME (F77/C++)    = 1.2828039280066150E-002
+Relative difference = 5.612189004572479e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_f_inl1_hrd0/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.839072e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.690688e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.690688e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.730630e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.541231e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.541231e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270341e-06 )  GeV^0
-TOTAL       :     2.575310 sec
-     7,484,630,621      cycles                           #    2.901 GHz                    
-    14,251,612,805      instructions                     #    1.90  insn per cycle         
-       2.581216132 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2237) (512y:    0) (512z:    0)
+TOTAL       :     2.679636 sec
+     7,420,774,430      cycles                           #    2.764 GHz                    
+    14,339,383,869      instructions                     #    1.93  insn per cycle         
+       2.686088553 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2251) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_f_inl1_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -147,26 +147,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_f_inl1_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_f_inl1_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282805e-02
-Avg ME (F77/C++)    = 1.2828053244447801E-002
-Relative difference = 2.5291823782248813e-07
+Avg ME (F77/C++)    = 1.2828053246266791E-002
+Relative difference = 2.5306003563303186e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_f_inl1_hrd0/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.852469e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.787188e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.787188e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.796396e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.739468e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.739468e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270341e-06 )  GeV^0
-TOTAL       :     2.574795 sec
-     7,331,036,803      cycles                           #    2.845 GHz                    
-    13,950,107,124      instructions                     #    1.90  insn per cycle         
-       2.580411131 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2096) (512y:    3) (512z:    0)
+TOTAL       :     2.625810 sec
+     7,304,334,176      cycles                           #    2.778 GHz                    
+    13,955,275,285      instructions                     #    1.91  insn per cycle         
+       2.632447793 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3875) (512y:    9) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_f_inl1_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -174,26 +174,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_f_inl1_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_f_inl1_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282805e-02
-Avg ME (F77/C++)    = 1.2828053244447801E-002
-Relative difference = 2.5291823782248813e-07
+Avg ME (F77/C++)    = 1.2828053277189611E-002
+Relative difference = 2.5547059841227576e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_f_inl1_hrd0/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[16] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.587347e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.044303e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.044303e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.601296e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.146430e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.146430e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270342e-06 )  GeV^0
-TOTAL       :     2.795609 sec
-     6,537,128,178      cycles                           #    2.335 GHz                    
-    13,422,893,916      instructions                     #    2.05  insn per cycle         
-       2.801293417 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2071) (512y:    1) (512z: 1198)
+TOTAL       :     2.796781 sec
+     6,273,154,150      cycles                           #    2.239 GHz                    
+    13,210,323,797      instructions                     #    2.11  insn per cycle         
+       2.803318258 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1734) (512y:    3) (512z: 1266)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_f_inl1_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -201,8 +201,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_f_inl1_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_f_inl1_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282805e-02
-Avg ME (F77/C++)    = 1.2828052562326775E-002
-Relative difference = 1.997440588685788e-07
+Avg ME (F77/C++)    = 1.2828052540498902E-002
+Relative difference = 1.980424851420537e-07
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_f_inl1_hrd1.txt b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_f_inl1_hrd1.txt
index 320612f062..784101060d 100644
--- a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_f_inl1_hrd1.txt
+++ b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_f_inl1_hrd1.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_f_inl1_hrd1'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum'
 
-DATE: 2023-11-24_15:00:08
+DATE: 2024-01-30_05:35:26
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl1_hrd1/gcheck.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 1.301880e+08                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.200050e+09                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.327102e+09                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.300997e+08                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.192378e+09                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.323768e+09                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371687e-02 +- 3.270220e-06 )  GeV^0
-TOTAL       :     0.577502 sec
-     2,367,605,467      cycles                           #    2.931 GHz                    
-     3,686,475,541      instructions                     #    1.56  insn per cycle         
-       0.865341028 seconds time elapsed
+TOTAL       :     0.574497 sec
+     2,274,789,999      cycles                           #    2.831 GHz                    
+     3,565,149,005      instructions                     #    1.57  insn per cycle         
+       0.863293975 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl1_hrd1/gcheck.exe -p 2048 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 95
@@ -72,20 +72,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl1_hrd1/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.231230e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.232983e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.232983e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.974769e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.758467e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.758467e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371707e-02 +- 3.270376e-06 )  GeV^0
-TOTAL       :     3.189007 sec
-     9,438,833,361      cycles                           #    2.955 GHz                    
-    25,269,338,727      instructions                     #    2.68  insn per cycle         
-       3.194786598 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  263) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     3.560083 sec
+    10,128,258,424      cycles                           #    2.841 GHz                    
+    28,399,859,483      instructions                     #    2.80  insn per cycle         
+       3.566485849 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  632) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl1_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -93,26 +93,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl1_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_f_inl1_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282804e-02
-Avg ME (F77/C++)    = 1.2828039838495897E-002
-Relative difference = 1.2589928273811243e-08
+Avg ME (F77/C++)    = 1.2828039441956207E-002
+Relative difference = 4.35018750695023e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_f_inl1_hrd1/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.111513e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 5.795769e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 5.795769e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.921662e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 5.360866e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.360866e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371707e-02 +- 3.270376e-06 )  GeV^0
-TOTAL       :     2.388364 sec
-     7,189,276,734      cycles                           #    3.004 GHz                    
-    16,868,514,931      instructions                     #    2.35  insn per cycle         
-       2.393974942 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 1360) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     2.529327 sec
+     7,292,501,410      cycles                           #    2.880 GHz                    
+    16,787,289,445      instructions                     #    2.30  insn per cycle         
+       2.535811154 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 2463) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_f_inl1_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -126,20 +126,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_f_inl1_hrd1/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.018803e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 5.200647e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 5.200647e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.902980e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 5.008268e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.008268e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270341e-06 )  GeV^0
-TOTAL       :     2.444205 sec
-     7,132,947,258      cycles                           #    2.912 GHz                    
-    13,616,405,803      instructions                     #    1.91  insn per cycle         
-       2.449801989 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2060) (512y:    0) (512z:    0)
+TOTAL       :     2.546290 sec
+     7,099,294,688      cycles                           #    2.783 GHz                    
+    13,729,465,706      instructions                     #    1.93  insn per cycle         
+       2.552602290 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2082) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_f_inl1_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -147,26 +147,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_f_inl1_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_f_inl1_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282805e-02
-Avg ME (F77/C++)    = 1.2828053220800939E-002
-Relative difference = 2.5107486628541925e-07
+Avg ME (F77/C++)    = 1.2828053198973066E-002
+Relative difference = 2.4937329255889414e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_f_inl1_hrd1/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.049548e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 5.286960e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 5.286960e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.894124e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 5.023412e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.023412e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270341e-06 )  GeV^0
-TOTAL       :     2.421039 sec
-     7,049,543,144      cycles                           #    2.906 GHz                    
-    13,426,454,258      instructions                     #    1.90  insn per cycle         
-       2.426713774 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1945) (512y:    4) (512z:    0)
+TOTAL       :     2.549860 sec
+     7,037,352,059      cycles                           #    2.755 GHz                    
+    13,462,222,302      instructions                     #    1.91  insn per cycle         
+       2.556338558 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3649) (512y:   12) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_f_inl1_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -174,26 +174,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_f_inl1_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_f_inl1_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282805e-02
-Avg ME (F77/C++)    = 1.2828053220800939E-002
-Relative difference = 2.5107486628541925e-07
+Avg ME (F77/C++)    = 1.2828053198973066E-002
+Relative difference = 2.4937329255889414e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_f_inl1_hrd1/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[16] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.688665e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.273438e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.273438e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.741921e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.505340e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.505340e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270342e-06 )  GeV^0
-TOTAL       :     2.707645 sec
-     6,353,817,019      cycles                           #    2.343 GHz                    
-    13,154,875,496      instructions                     #    2.07  insn per cycle         
-       2.713469152 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2029) (512y:    1) (512z: 1083)
+TOTAL       :     2.671598 sec
+     6,046,764,080      cycles                           #    2.259 GHz                    
+    12,911,501,907      instructions                     #    2.14  insn per cycle         
+       2.677952936 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1671) (512y:    3) (512z: 1155)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_f_inl1_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -201,8 +201,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_f_inl1_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_f_inl1_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.282805e-02
-Avg ME (F77/C++)    = 1.2828052536860923E-002
-Relative difference = 1.977588895209662e-07
+Avg ME (F77/C++)    = 1.2828052431359538E-002
+Relative difference = 1.895346165094282e-07
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_m_inl0_hrd0.txt b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_m_inl0_hrd0.txt
index 675f8002f0..7a09642823 100644
--- a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_m_inl0_hrd0.txt
+++ b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_m_inl0_hrd0.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_m_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum'
 
-DATE: 2023-11-24_14:29:57
+DATE: 2024-01-30_04:54:00
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_m_inl0_hrd0/gcheck.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 5.421403e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.263390e+08                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.120736e+08                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.434258e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.281519e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.171049e+08                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     0.699628 sec
-     2,791,107,487      cycles                           #    2.987 GHz                    
-     4,371,723,888      instructions                     #    1.57  insn per cycle         
-       1.020217670 seconds time elapsed
+TOTAL       :     0.704261 sec
+     2,701,570,097      cycles                           #    2.831 GHz                    
+     4,244,340,283      instructions                     #    1.57  insn per cycle         
+       1.033944641 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_m_inl0_hrd0/gcheck.exe -p 2048 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 166
@@ -72,20 +72,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_m_inl0_hrd0/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.110744e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.296406e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.296406e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 9.829628e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.139787e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.139787e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     6.059605 sec
-    18,785,537,380      cycles                           #    3.098 GHz                    
-    44,223,451,515      instructions                     #    2.35  insn per cycle         
-       6.072025963 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  439) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     6.819159 sec
+    19,690,827,956      cycles                           #    2.885 GHz                    
+    46,971,779,576      instructions                     #    2.39  insn per cycle         
+       6.832663552 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  474) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_m_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -99,20 +99,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_m_inl0_hrd0/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.746395e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.315536e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.315536e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.605344e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.116934e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.116934e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     3.998774 sec
-    12,365,852,528      cycles                           #    3.089 GHz                    
-    30,918,818,290      instructions                     #    2.50  insn per cycle         
-       4.019870048 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 1685) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     4.334479 sec
+    12,518,471,325      cycles                           #    2.884 GHz                    
+    30,922,888,427      instructions                     #    2.47  insn per cycle         
+       4.354467708 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 1667) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_m_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -126,20 +126,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_m_inl0_hrd0/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.045102e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.850310e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.850310e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.917239e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.660472e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.660472e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     3.473353 sec
-    10,119,124,866      cycles                           #    2.909 GHz                    
-    19,374,205,910      instructions                     #    1.91  insn per cycle         
-       3.489755488 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2130) (512y:    0) (512z:    0)
+TOTAL       :     3.702387 sec
+    10,174,876,030      cycles                           #    2.745 GHz                    
+    19,548,406,942      instructions                     #    1.92  insn per cycle         
+       3.720275920 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2119) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_m_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -153,20 +153,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_m_inl0_hrd0/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.189661e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.096081e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.096081e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.029293e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.888276e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.888276e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     3.261039 sec
-     9,715,236,596      cycles                           #    2.974 GHz                    
-    18,955,064,271      instructions                     #    1.95  insn per cycle         
-       3.284788216 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1860) (512y:  188) (512z:    0)
+TOTAL       :     3.515786 sec
+     9,723,051,646      cycles                           #    2.761 GHz                    
+    18,859,468,530      instructions                     #    1.94  insn per cycle         
+       3.531121351 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1850) (512y:  174) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_m_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -180,20 +180,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_m_inl0_hrd0/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.871346e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.512373e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.512373e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.839848e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.512898e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.512898e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     3.767041 sec
-     8,387,450,309      cycles                           #    2.224 GHz                    
-    15,057,184,265      instructions                     #    1.80  insn per cycle         
-       3.785820964 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1024) (512y:  155) (512z: 1316)
+TOTAL       :     3.838759 sec
+     8,110,381,366      cycles                           #    2.110 GHz                    
+    14,814,382,883      instructions                     #    1.83  insn per cycle         
+       3.856049832 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1023) (512y:   64) (512z: 1327)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_m_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
diff --git a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_m_inl0_hrd1.txt b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_m_inl0_hrd1.txt
index 486522e284..385e9ed225 100644
--- a/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_m_inl0_hrd1.txt
+++ b/epochX/cudacpp/tput/logs_eemumu_mad/log_eemumu_mad_m_inl0_hrd1.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_m_inl0_hrd1'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum'
 
-DATE: 2023-11-24_14:30:31
+DATE: 2024-01-30_04:54:36
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_m_inl0_hrd1/gcheck.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 5.608833e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.666599e+08                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.153828e+08                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.428632e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.291557e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.197877e+08                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     0.689454 sec
-     2,790,185,844      cycles                           #    3.012 GHz                    
-     4,322,858,919      instructions                     #    1.55  insn per cycle         
-       1.008572761 seconds time elapsed
+TOTAL       :     0.704173 sec
+     2,700,513,236      cycles                           #    2.833 GHz                    
+     4,160,757,344      instructions                     #    1.54  insn per cycle         
+       1.040080983 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_m_inl0_hrd1/gcheck.exe -p 2048 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 154
@@ -72,20 +72,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_m_inl0_hrd1/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.152239e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.354641e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.354641e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.048898e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.230601e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.230601e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     5.858273 sec
-    18,113,057,840      cycles                           #    3.089 GHz                    
-    42,472,824,521      instructions                     #    2.34  insn per cycle         
-       5.871069159 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  421) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     6.414952 sec
+    18,538,807,361      cycles                           #    2.888 GHz                    
+    44,591,647,960      instructions                     #    2.41  insn per cycle         
+       6.426389730 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  498) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.none_m_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -99,20 +99,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_m_inl0_hrd1/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.781270e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.374250e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.374250e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.655305e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.204388e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.204388e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     3.930434 sec
-    12,144,033,984      cycles                           #    3.086 GHz                    
-    30,226,379,703      instructions                     #    2.49  insn per cycle         
-       3.949543769 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 1692) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     4.214890 sec
+    12,207,966,974      cycles                           #    2.892 GHz                    
+    30,217,340,923      instructions                     #    2.48  insn per cycle         
+       4.236133486 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 1650) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.sse4_m_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -126,20 +126,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_m_inl0_hrd1/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.103753e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.937318e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.937318e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.899712e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.627205e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.627205e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     3.379502 sec
-    10,060,863,868      cycles                           #    2.972 GHz                    
-    19,256,812,816      instructions                     #    1.91  insn per cycle         
-       3.401404400 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2146) (512y:    0) (512z:    0)
+TOTAL       :     3.730288 sec
+    10,158,219,608      cycles                           #    2.719 GHz                    
+    19,037,132,874      instructions                     #    1.87  insn per cycle         
+       3.746558078 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2072) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.avx2_m_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -153,20 +153,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_m_inl0_hrd1/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.196753e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.119301e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.119301e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.048047e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.931283e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.931283e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     3.254342 sec
-     9,672,428,613      cycles                           #    2.968 GHz                    
-    18,746,035,376      instructions                     #    1.94  insn per cycle         
-       3.271198287 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1834) (512y:  191) (512z:    0)
+TOTAL       :     3.492411 sec
+     9,571,391,969      cycles                           #    2.738 GHz                    
+    18,453,150,608      instructions                     #    1.93  insn per cycle         
+       3.509341045 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1775) (512y:  174) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512y_m_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -180,20 +180,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_m_inl0_hrd1/check.exe -p 2048 256 12 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_EPEM_MUPMUM_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.930945e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.613686e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.613686e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.170414e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.170487e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.170487e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.371706e-02 +- 3.270315e-06 )  GeV^0
-TOTAL       :     3.660732 sec
-     8,282,994,053      cycles                           #    2.260 GHz                    
-    14,980,698,691      instructions                     #    1.81  insn per cycle         
-       3.677125812 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1021) (512y:  156) (512z: 1305)
+TOTAL       :     3.317793 sec
+     7,240,072,684      cycles                           #    2.179 GHz                    
+    13,244,781,040      instructions                     #    1.83  insn per cycle         
+       3.341198784 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:  911) (512y:   56) (512z:  993)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/ee_mumu.mad/SubProcesses/P1_epem_mupmum/build.512z_m_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
diff --git a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_d_inl0_hrd0.txt b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_d_inl0_hrd0.txt
index 8af0b3625a..2453732bed 100644
--- a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_d_inl0_hrd0.txt
+++ b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_d_inl0_hrd0.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
 
-DATE: 2023-11-24_14:31:04
+DATE: 2024-01-30_04:55:10
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl0_hrd0/gcheck.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 4.273933e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.156544e+08                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.270780e+08                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.010275e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.133419e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.272295e+08                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     0.525661 sec
-     2,285,041,869      cycles                           #    2.996 GHz                    
-     3,259,226,119      instructions                     #    1.43  insn per cycle         
-       0.837415991 seconds time elapsed
+TOTAL       :     0.538830 sec
+     2,187,358,219      cycles                           #    2.824 GHz                    
+     3,139,905,445      instructions                     #    1.44  insn per cycle         
+       0.856073288 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl0_hrd0/gcheck.exe -p 2048 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 214
@@ -72,20 +72,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl0_hrd0/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.147376e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.208578e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.208578e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.073581e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.135755e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.135755e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     4.988043 sec
-    15,215,107,476      cycles                           #    3.051 GHz                    
-    38,378,066,668      instructions                     #    2.52  insn per cycle         
-       4.996465245 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  668) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     5.168677 sec
+    14,980,961,047      cycles                           #    2.896 GHz                    
+    38,724,485,120      instructions                     #    2.58  insn per cycle         
+       5.178651966 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  719) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -93,26 +93,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028807e+00
-Avg ME (F77/C++)    = 2.0288063388515645
-Relative difference = 3.258803994438787e-07
+Avg ME (F77/C++)    = 2.0288063388515649
+Relative difference = 3.258803992249869e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_d_inl0_hrd0/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.686108e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.886642e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.886642e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.523460e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.721558e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.721558e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     2.950305 sec
-     9,139,887,426      cycles                           #    3.093 GHz                    
-    24,578,745,754      instructions                     #    2.69  insn per cycle         
-       2.962852368 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 2159) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     3.090012 sec
+     8,952,192,290      cycles                           #    2.893 GHz                    
+    24,430,503,496      instructions                     #    2.73  insn per cycle         
+       3.108451490 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 2067) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -126,20 +126,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_d_inl0_hrd0/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 5.966833e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 6.492744e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 6.492744e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.390626e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 5.850527e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.850527e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     1.861721 sec
-     5,467,401,485      cycles                           #    2.927 GHz                    
-    11,252,306,729      instructions                     #    2.06  insn per cycle         
-       1.878450325 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2376) (512y:    0) (512z:    0)
+TOTAL       :     2.056967 sec
+     5,535,228,908      cycles                           #    2.683 GHz                    
+    11,562,552,185      instructions                     #    2.09  insn per cycle         
+       2.068379535 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2396) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -153,20 +153,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_d_inl0_hrd0/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 6.607385e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.251361e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.251361e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.323214e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.965355e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.965355e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     1.691034 sec
-     4,950,324,717      cycles                           #    2.917 GHz                    
-    10,558,681,133      instructions                     #    2.13  insn per cycle         
-       1.704227686 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2074) (512y:  144) (512z:    0)
+TOTAL       :     1.769440 sec
+     4,825,692,035      cycles                           #    2.719 GHz                    
+    10,341,008,591      instructions                     #    2.14  insn per cycle         
+       1.786949030 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1972) (512y:  131) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -180,20 +180,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_d_inl0_hrd0/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.997999e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.223536e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.223536e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.039053e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.289363e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.289363e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     2.729108 sec
-     5,392,071,178      cycles                           #    1.972 GHz                    
-     7,793,346,651      instructions                     #    1.45  insn per cycle         
-       2.742252965 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1445) (512y:  122) (512z: 1542)
+TOTAL       :     2.707049 sec
+     4,944,236,176      cycles                           #    1.822 GHz                    
+     7,554,838,116      instructions                     #    1.53  insn per cycle         
+       2.726854934 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1212) (512y:   65) (512z: 1543)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
diff --git a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_d_inl0_hrd0_bridge.txt b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_d_inl0_hrd0_bridge.txt
index abc9cb4db6..adcfa48462 100644
--- a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_d_inl0_hrd0_bridge.txt
+++ b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_d_inl0_hrd0_bridge.txt
@@ -41,7 +41,7 @@ CUDACPP_BUILDDIR='build.512z_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
 
-DATE: 2023-11-24_15:10:57
+DATE: 2024-01-30_05:46:39
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
@@ -51,17 +51,17 @@ WARNING! Bridge selected: cannot use RamboDevice, will use RamboHost
 WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
 WARNING! Instantiate device Bridge (nevt=524288, gpublocks=2048, gputhreads=256, gpublocks*gputhreads=524288)
 WARNING! Set grid in Bridge (nevt=524288, gpublocks=2048, gputhreads=256, gpublocks*gputhreads=524288)
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 4.408166e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.841952e+07                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.841952e+07                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.344134e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.848581e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.848581e+07                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     0.827206 sec
-     3,081,034,031      cycles                           #    2.876 GHz                    
-     4,780,745,395      instructions                     #    1.55  insn per cycle         
-       1.128942393 seconds time elapsed
+TOTAL       :     0.830257 sec
+     3,050,711,174      cycles                           #    2.837 GHz                    
+     4,744,287,151      instructions                     #    1.56  insn per cycle         
+       1.134543078 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl0_hrd0/gcheck.exe -p 2048 256 1 --bridge
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Bridge selected: cannot use RamboDevice, will use RamboHost
@@ -81,20 +81,20 @@ OK (relative difference <= 5E-3)
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl0_hrd0/check.exe -p 2048 256 2 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=524288)
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.108776e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.169543e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.169543e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.051768e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.112380e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.112380e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     5.156626 sec
-    15,517,564,279      cycles                           #    3.006 GHz                    
-    38,434,109,341      instructions                     #    2.48  insn per cycle         
-       5.163690036 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  668) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     5.299158 sec
+    15,311,911,023      cycles                           #    2.886 GHz                    
+    38,783,796,929      instructions                     #    2.53  insn per cycle         
+       5.307164517 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  719) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -102,27 +102,27 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028807e+00
-Avg ME (F77/C++)    = 2.0288063388515645
-Relative difference = 3.258803994438787e-07
+Avg ME (F77/C++)    = 2.0288063388515649
+Relative difference = 3.258803992249869e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_d_inl0_hrd0/check.exe -p 2048 256 2 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=524288)
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/sse4+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.545882e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.738725e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.738725e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.466739e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.657869e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.657869e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     3.141184 sec
-     9,467,320,162      cycles                           #    3.008 GHz                    
-    24,761,393,280      instructions                     #    2.62  insn per cycle         
-       3.148330326 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 2159) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     3.216841 sec
+     9,297,524,138      cycles                           #    2.885 GHz                    
+    24,613,723,387      instructions                     #    2.65  insn per cycle         
+       3.224967553 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 2067) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -137,20 +137,20 @@ OK (relative difference <= 5E-3)
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_d_inl0_hrd0/check.exe -p 2048 256 2 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=524288)
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/avx2+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 5.651058e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 6.129257e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 6.129257e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.363369e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 5.815752e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.815752e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     2.039696 sec
-     5,802,587,355      cycles                           #    2.836 GHz                    
-    11,538,014,597      instructions                     #    1.99  insn per cycle         
-       2.046642189 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2376) (512y:    0) (512z:    0)
+TOTAL       :     2.149413 sec
+     5,860,102,645      cycles                           #    2.720 GHz                    
+    11,849,599,468      instructions                     #    2.02  insn per cycle         
+       2.157292568 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2396) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -165,20 +165,20 @@ OK (relative difference <= 5E-3)
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_d_inl0_hrd0/check.exe -p 2048 256 2 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=524288)
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/512y+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 6.265451e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 6.861082e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 6.861082e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.162124e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.773170e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.773170e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     1.859965 sec
-     5,312,241,459      cycles                           #    2.848 GHz                    
-    10,843,960,204      instructions                     #    2.04  insn per cycle         
-       1.867027334 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2074) (512y:  144) (512z:    0)
+TOTAL       :     1.893501 sec
+     5,161,881,245      cycles                           #    2.717 GHz                    
+    10,626,023,875      instructions                     #    2.06  insn per cycle         
+       1.901369932 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1972) (512y:  131) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -193,20 +193,20 @@ OK (relative difference <= 5E-3)
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_d_inl0_hrd0/check.exe -p 2048 256 2 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=524288)
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/512z+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.763993e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.974534e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.974534e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.945106e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.186618e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.186618e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     2.977042 sec
-     5,732,583,120      cycles                           #    1.922 GHz                    
-     8,038,189,975      instructions                     #    1.40  insn per cycle         
-       2.984163935 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1445) (512y:  122) (512z: 1542)
+TOTAL       :     2.853446 sec
+     5,298,812,686      cycles                           #    1.853 GHz                    
+     7,800,536,018      instructions                     #    1.47  insn per cycle         
+       2.861501356 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1212) (512y:   65) (512z: 1543)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
diff --git a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_d_inl0_hrd0_common.txt b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_d_inl0_hrd0_common.txt
index 44f3cc8c52..b23b4b948e 100644
--- a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_d_inl0_hrd0_common.txt
+++ b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_d_inl0_hrd0_common.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
 
-DATE: 2023-11-24_15:24:19
+DATE: 2024-01-30_06:00:31
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl0_hrd0/gcheck.exe -p 2048 256 2 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:COMMON+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 4.501067e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.156482e+08                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.276049e+08                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.565155e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.155605e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.269580e+08                 )  sec^-1
 MeanMatrixElemValue         = ( 2.079401e+00 +- 3.402993e-03 )  GeV^0
-TOTAL       :     0.626850 sec
-     2,536,021,563      cycles                           #    2.961 GHz                    
-     3,670,883,227      instructions                     #    1.45  insn per cycle         
-       0.915863654 seconds time elapsed
+TOTAL       :     0.625620 sec
+     2,433,553,514      cycles                           #    2.839 GHz                    
+     3,531,317,325      instructions                     #    1.45  insn per cycle         
+       0.914840106 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl0_hrd0/gcheck.exe -p 2048 256 1 --common
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 214
@@ -72,20 +72,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl0_hrd0/check.exe -p 2048 256 2 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:COMMON+RMBHST+MESHST/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.139594e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.203265e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.203265e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.073060e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.134707e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.134707e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.079401e+00 +- 3.402993e-03 )  GeV^0
-TOTAL       :     5.067424 sec
-    15,399,016,558      cycles                           #    3.036 GHz                    
-    38,390,017,774      instructions                     #    2.49  insn per cycle         
-       5.073191640 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  668) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     5.229961 sec
+    15,157,435,498      cycles                           #    2.896 GHz                    
+    38,739,723,091      instructions                     #    2.56  insn per cycle         
+       5.236486145 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  719) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -93,26 +93,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028807e+00
-Avg ME (F77/C++)    = 2.0288063388515645
-Relative difference = 3.258803994438787e-07
+Avg ME (F77/C++)    = 2.0288063388515649
+Relative difference = 3.258803992249869e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_d_inl0_hrd0/check.exe -p 2048 256 2 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:COMMON+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.625228e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.825269e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.825269e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.526696e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.723798e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.723798e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.079401e+00 +- 3.402993e-03 )  GeV^0
-TOTAL       :     3.057820 sec
-     9,317,797,118      cycles                           #    3.043 GHz                    
-    24,577,704,184      instructions                     #    2.64  insn per cycle         
-       3.063889144 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 2159) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     3.143207 sec
+     9,122,833,846      cycles                           #    2.898 GHz                    
+    24,428,638,513      instructions                     #    2.68  insn per cycle         
+       3.149727451 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 2067) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -126,20 +126,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_d_inl0_hrd0/check.exe -p 2048 256 2 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:COMMON+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 5.671770e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 6.193648e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 6.193648e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.453278e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 5.923487e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.923487e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.079401e+00 +- 3.402993e-03 )  GeV^0
-TOTAL       :     2.021111 sec
-     5,663,540,101      cycles                           #    2.803 GHz                    
-    11,235,902,290      instructions                     #    1.98  insn per cycle         
-       2.027218591 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2376) (512y:    0) (512z:    0)
+TOTAL       :     2.094012 sec
+     5,713,399,327      cycles                           #    2.721 GHz                    
+    11,544,398,198      instructions                     #    2.02  insn per cycle         
+       2.100575275 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2396) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -153,20 +153,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_d_inl0_hrd0/check.exe -p 2048 256 2 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:COMMON+RMBHST+MESHST/512y+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 6.440843e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.078509e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.078509e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.340982e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.000324e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.000324e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.079401e+00 +- 3.402993e-03 )  GeV^0
-TOTAL       :     1.794383 sec
-     5,172,944,461      cycles                           #    2.875 GHz                    
-    10,505,659,530      instructions                     #    2.03  insn per cycle         
-       1.800280603 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2074) (512y:  144) (512z:    0)
+TOTAL       :     1.826424 sec
+     5,007,819,577      cycles                           #    2.734 GHz                    
+    10,288,512,439      instructions                     #    2.05  insn per cycle         
+       1.833139039 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1972) (512y:  131) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -180,20 +180,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_d_inl0_hrd0/check.exe -p 2048 256 2 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:COMMON+RMBHST+MESHST/512z+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.980860e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.211163e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.211163e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.024689e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.274198e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.274198e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.079401e+00 +- 3.402993e-03 )  GeV^0
-TOTAL       :     2.800543 sec
-     5,575,018,302      cycles                           #    1.987 GHz                    
-     7,742,183,831      instructions                     #    1.39  insn per cycle         
-       2.806534120 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1445) (512y:  122) (512z: 1542)
+TOTAL       :     2.778758 sec
+     5,115,298,192      cycles                           #    1.837 GHz                    
+     7,503,411,062      instructions                     #    1.47  insn per cycle         
+       2.785395708 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1212) (512y:   65) (512z: 1543)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
diff --git a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_d_inl0_hrd0_curhst.txt b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_d_inl0_hrd0_curhst.txt
index af2440539e..66a621d02a 100644
--- a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_d_inl0_hrd0_curhst.txt
+++ b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_d_inl0_hrd0_curhst.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
 
-DATE: 2023-11-24_15:20:58
+DATE: 2024-01-30_05:57:01
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl0_hrd0/gcheck.exe -p 2048 256 2 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 4.564058e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.155046e+08                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.272805e+08                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.578143e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.159887e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.277521e+08                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     0.562322 sec
-     2,282,427,973      cycles                           #    2.881 GHz                    
-     3,556,665,675      instructions                     #    1.56  insn per cycle         
-       0.849543273 seconds time elapsed
+TOTAL       :     0.567703 sec
+     2,256,406,335      cycles                           #    2.832 GHz                    
+     3,552,290,336      instructions                     #    1.57  insn per cycle         
+       0.856591173 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl0_hrd0/gcheck.exe -p 2048 256 1 --curhst
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 214
@@ -72,20 +72,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl0_hrd0/check.exe -p 2048 256 2 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.099614e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.160783e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.160783e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.061569e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.123242e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.123242e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     5.101479 sec
-    15,194,748,744      cycles                           #    2.976 GHz                    
-    38,374,370,448      instructions                     #    2.53  insn per cycle         
-       5.107016878 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  668) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     5.196625 sec
+    14,980,592,489      cycles                           #    2.880 GHz                    
+    38,723,298,937      instructions                     #    2.58  insn per cycle         
+       5.203366404 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  719) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -93,26 +93,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028807e+00
-Avg ME (F77/C++)    = 2.0288063388515645
-Relative difference = 3.258803994438787e-07
+Avg ME (F77/C++)    = 2.0288063388515649
+Relative difference = 3.258803992249869e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_d_inl0_hrd0/check.exe -p 2048 256 2 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.574526e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.771653e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.771653e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.518700e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.715553e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.715553e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     3.040842 sec
-     9,128,651,326      cycles                           #    2.997 GHz                    
-    24,578,551,986      instructions                     #    2.69  insn per cycle         
-       3.046498874 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 2159) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     3.090850 sec
+     8,946,489,145      cycles                           #    2.890 GHz                    
+    24,429,263,818      instructions                     #    2.73  insn per cycle         
+       3.097198356 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 2067) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -126,20 +126,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_d_inl0_hrd0/check.exe -p 2048 256 2 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 5.714772e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 6.221583e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 6.221583e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.476437e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 5.948509e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.948509e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     1.940536 sec
-     5,479,977,577      cycles                           #    2.817 GHz                    
-    11,251,107,887      instructions                     #    2.05  insn per cycle         
-       1.946094154 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2376) (512y:    0) (512z:    0)
+TOTAL       :     2.025257 sec
+     5,523,468,825      cycles                           #    2.720 GHz                    
+    11,561,737,650      instructions                     #    2.09  insn per cycle         
+       2.031752517 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2396) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -153,20 +153,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_d_inl0_hrd0/check.exe -p 2048 256 2 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 6.393849e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.021326e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.021326e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.358069e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.007551e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.007551e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     1.745575 sec
-     4,952,345,278      cycles                           #    2.829 GHz                    
-    10,558,643,656      instructions                     #    2.13  insn per cycle         
-       1.751295687 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2074) (512y:  144) (512z:    0)
+TOTAL       :     1.759129 sec
+     4,801,841,802      cycles                           #    2.722 GHz                    
+    10,338,992,386      instructions                     #    2.15  insn per cycle         
+       1.765685267 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1972) (512y:  131) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -180,20 +180,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_d_inl0_hrd0/check.exe -p 2048 256 2 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.913352e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.137894e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.137894e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.036811e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.287808e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.287808e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     2.786489 sec
-     5,383,532,315      cycles                           #    1.930 GHz                    
-     7,793,484,712      instructions                     #    1.45  insn per cycle         
-       2.792074852 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1445) (512y:  122) (512z: 1542)
+TOTAL       :     2.707972 sec
+     4,942,835,417      cycles                           #    1.822 GHz                    
+     7,554,452,946      instructions                     #    1.53  insn per cycle         
+       2.714536601 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1212) (512y:   65) (512z: 1543)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
diff --git a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_d_inl0_hrd0_rmbhst.txt b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_d_inl0_hrd0_rmbhst.txt
index e2193a8ad7..defb46a739 100644
--- a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_d_inl0_hrd0_rmbhst.txt
+++ b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_d_inl0_hrd0_rmbhst.txt
@@ -41,24 +41,24 @@ CUDACPP_BUILDDIR='build.512z_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
 
-DATE: 2023-11-24_15:17:39
+DATE: 2024-01-30_05:53:36
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl0_hrd0/gcheck.exe -p 2048 256 2 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 5.736010e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.153457e+08                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.268173e+08                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.688012e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.154108e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.269539e+08                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     0.714625 sec
-     2,774,816,362      cycles                           #    2.935 GHz                    
-     4,339,252,047      instructions                     #    1.56  insn per cycle         
-       1.002590963 seconds time elapsed
+TOTAL       :     0.720498 sec
+     2,707,766,521      cycles                           #    2.848 GHz                    
+     4,278,662,893      instructions                     #    1.58  insn per cycle         
+       1.009865256 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl0_hrd0/gcheck.exe -p 2048 256 1 --rmbhst
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
@@ -74,20 +74,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl0_hrd0/check.exe -p 2048 256 2 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.123090e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.185367e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.185367e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.066193e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.127901e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.127901e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     5.045686 sec
-    15,189,366,713      cycles                           #    3.007 GHz                    
-    38,373,841,814      instructions                     #    2.53  insn per cycle         
-       5.051730938 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  668) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     5.184582 sec
+    14,984,631,159      cycles                           #    2.888 GHz                    
+    38,723,388,390      instructions                     #    2.58  insn per cycle         
+       5.191155299 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  719) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -95,26 +95,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028807e+00
-Avg ME (F77/C++)    = 2.0288063388515645
-Relative difference = 3.258803994438787e-07
+Avg ME (F77/C++)    = 2.0288063388515649
+Relative difference = 3.258803992249869e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_d_inl0_hrd0/check.exe -p 2048 256 2 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.586999e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.782954e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.782954e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.511860e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.708377e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.708377e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     3.030197 sec
-     9,119,166,751      cycles                           #    3.005 GHz                    
-    24,577,924,210      instructions                     #    2.70  insn per cycle         
-       3.035979194 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 2159) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     3.095761 sec
+     8,950,231,816      cycles                           #    2.886 GHz                    
+    24,430,052,071      instructions                     #    2.73  insn per cycle         
+       3.102564983 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 2067) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -128,20 +128,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_d_inl0_hrd0/check.exe -p 2048 256 2 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 5.732847e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 6.231342e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 6.231342e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.454029e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 5.925499e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.925499e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     1.935219 sec
-     5,468,637,644      cycles                           #    2.819 GHz                    
-    11,251,012,735      instructions                     #    2.06  insn per cycle         
-       1.941019031 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2376) (512y:    0) (512z:    0)
+TOTAL       :     2.033615 sec
+     5,531,582,240      cycles                           #    2.713 GHz                    
+    11,562,288,179      instructions                     #    2.09  insn per cycle         
+       2.040383969 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2396) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -155,20 +155,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_d_inl0_hrd0/check.exe -p 2048 256 2 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 6.292655e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 6.897577e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 6.897577e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.327959e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.977069e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.977069e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     1.772120 sec
-     4,969,624,681      cycles                           #    2.796 GHz                    
-    10,556,732,876      instructions                     #    2.12  insn per cycle         
-       1.777837624 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2074) (512y:  144) (512z:    0)
+TOTAL       :     1.768254 sec
+     4,816,907,251      cycles                           #    2.716 GHz                    
+    10,339,308,595      instructions                     #    2.15  insn per cycle         
+       1.774968996 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1972) (512y:  131) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -182,20 +182,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_d_inl0_hrd0/check.exe -p 2048 256 2 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.866239e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.092519e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.092519e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.992436e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.241387e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.241387e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     2.846913 sec
-     5,391,039,174      cycles                           #    1.904 GHz                    
-     7,794,494,031      instructions                     #    1.45  insn per cycle         
-       2.856378481 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1445) (512y:  122) (512z: 1542)
+TOTAL       :     2.737995 sec
+     4,943,973,305      cycles                           #    1.803 GHz                    
+     7,555,690,658      instructions                     #    1.53  insn per cycle         
+       2.744582139 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1212) (512y:   65) (512z: 1543)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
diff --git a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_d_inl0_hrd1.txt b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_d_inl0_hrd1.txt
index 977d853d64..fe6f195aa6 100644
--- a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_d_inl0_hrd1.txt
+++ b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_d_inl0_hrd1.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_d_inl0_hrd1'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
 
-DATE: 2023-11-24_14:31:31
+DATE: 2024-01-30_04:55:39
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl0_hrd1/gcheck.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 4.446963e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.163595e+08                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.279925e+08                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.125481e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.158117e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.273663e+08                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     0.520650 sec
-     2,264,469,596      cycles                           #    3.001 GHz                    
-     3,268,925,935      instructions                     #    1.44  insn per cycle         
-       0.824307613 seconds time elapsed
+TOTAL       :     0.534485 sec
+     2,191,778,134      cycles                           #    2.834 GHz                    
+     3,140,951,827      instructions                     #    1.43  insn per cycle         
+       0.850685752 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl0_hrd1/gcheck.exe -p 2048 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 208
@@ -72,20 +72,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl0_hrd1/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.178244e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.243144e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.243144e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.109309e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.173415e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.173415e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     4.920762 sec
-    15,120,754,616      cycles                           #    3.070 GHz                    
-    40,101,358,095      instructions                     #    2.65  insn per cycle         
-       4.928885973 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  669) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     5.081724 sec
+    14,685,294,357      cycles                           #    2.887 GHz                    
+    39,544,026,748      instructions                     #    2.69  insn per cycle         
+       5.093038112 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  596) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -99,20 +99,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_d_inl0_hrd1/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.908591e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.133379e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.133379e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.661768e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.875473e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.875473e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     2.787534 sec
-     8,670,203,341      cycles                           #    3.104 GHz                    
-    23,670,756,212      instructions                     #    2.73  insn per cycle         
-       2.801209294 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 2072) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     2.973090 sec
+     8,600,238,365      cycles                           #    2.886 GHz                    
+    23,576,508,735      instructions                     #    2.74  insn per cycle         
+       2.991032269 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 1948) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_d_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -126,20 +126,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_d_inl0_hrd1/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 5.339479e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 5.760864e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 5.760864e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.966204e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 5.352181e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.352181e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     2.067737 sec
-     6,095,582,988      cycles                           #    2.940 GHz                    
-    13,061,573,034      instructions                     #    2.14  insn per cycle         
-       2.079447172 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2546) (512y:    0) (512z:    0)
+TOTAL       :     2.222703 sec
+     5,964,350,122      cycles                           #    2.676 GHz                    
+    13,193,903,385      instructions                     #    2.21  insn per cycle         
+       2.290428549 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2560) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_d_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -153,20 +153,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_d_inl0_hrd1/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 5.617733e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 6.075014e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 6.075014e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.425705e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 5.897406e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.897406e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     1.970361 sec
-     5,800,772,939      cycles                           #    2.935 GHz                    
-    12,319,856,699      instructions                     #    2.12  insn per cycle         
-       1.986108854 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2093) (512y:  294) (512z:    0)
+TOTAL       :     2.043603 sec
+     5,539,021,528      cycles                           #    2.702 GHz                    
+    12,103,311,893      instructions                     #    2.19  insn per cycle         
+       2.060365335 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2030) (512y:  278) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_d_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -180,20 +180,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_d_inl0_hrd1/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.793878e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.997441e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.997441e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.662802e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.870728e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.870728e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     2.868253 sec
-     5,826,829,144      cycles                           #    2.028 GHz                    
-     9,602,061,917      instructions                     #    1.65  insn per cycle         
-       2.882159739 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1509) (512y:  209) (512z: 1971)
+TOTAL       :     2.974749 sec
+     5,366,303,915      cycles                           #    1.800 GHz                    
+     9,381,926,109      instructions                     #    1.75  insn per cycle         
+       2.994553633 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1350) (512y:   88) (512z: 1989)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_d_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
diff --git a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_d_inl1_hrd0.txt b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_d_inl1_hrd0.txt
index eaf53289b1..8cd37966a9 100644
--- a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_d_inl1_hrd0.txt
+++ b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_d_inl1_hrd0.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_d_inl1_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
 
-DATE: 2023-11-24_15:00:34
+DATE: 2024-01-30_05:35:53
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl1_hrd0/gcheck.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 4.552166e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.156300e+08                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.271682e+08                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.561376e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.154966e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.270589e+08                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     0.524921 sec
-     2,220,579,587      cycles                           #    2.921 GHz                    
-     3,192,850,808      instructions                     #    1.44  insn per cycle         
-       0.818936986 seconds time elapsed
+TOTAL       :     0.529335 sec
+     2,159,762,911      cycles                           #    2.829 GHz                    
+     3,107,803,545      instructions                     #    1.44  insn per cycle         
+       0.822533200 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl1_hrd0/gcheck.exe -p 2048 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 214
@@ -72,20 +72,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl1_hrd0/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.433349e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.518013e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.518013e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.227004e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.298943e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.298943e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     4.419477 sec
-    13,028,620,235      cycles                           #    2.950 GHz                    
-    34,390,570,196      instructions                     #    2.64  insn per cycle         
-       4.425198715 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  686) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     4.818572 sec
+    13,907,927,893      cycles                           #    2.883 GHz                    
+    35,849,684,316      instructions                     #    2.58  insn per cycle         
+       4.825096940 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 1078) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl1_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -99,20 +99,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_d_inl1_hrd0/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.068689e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.210680e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.210680e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.848483e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.087109e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.087109e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     3.525285 sec
-    10,592,983,737      cycles                           #    3.000 GHz                    
-    24,007,204,141      instructions                     #    2.27  insn per cycle         
-       3.531288478 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 2582) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     2.835129 sec
+     8,213,185,511      cycles                           #    2.892 GHz                    
+    21,908,282,308      instructions                     #    2.67  insn per cycle         
+       2.841971377 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 2334) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_d_inl1_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -126,20 +126,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_d_inl1_hrd0/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 4.591275e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.918287e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.918287e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.473983e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 5.948336e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.948336e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     2.392081 sec
-     6,643,133,250      cycles                           #    2.771 GHz                    
-    12,401,445,774      instructions                     #    1.87  insn per cycle         
-       2.398243451 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3154) (512y:    0) (512z:    0)
+TOTAL       :     2.025782 sec
+     5,530,364,572      cycles                           #    2.723 GHz                    
+    12,076,349,288      instructions                     #    2.18  insn per cycle         
+       2.032542267 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3062) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_d_inl1_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -147,26 +147,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_d_inl1_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_d_inl1_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028807e+00
-Avg ME (F77/C++)    = 2.0288063388516200
-Relative difference = 3.2588037208240405e-07
+Avg ME (F77/C++)    = 2.0288063388516204
+Relative difference = 3.2588037186351226e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_d_inl1_hrd0/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 5.018329e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 5.394522e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 5.394522e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.936500e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.499652e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.499652e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     2.195662 sec
-     6,234,944,949      cycles                           #    2.833 GHz                    
-    11,573,018,455      instructions                     #    1.86  insn per cycle         
-       2.201415670 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2690) (512y:  239) (512z:    0)
+TOTAL       :     1.876359 sec
+     5,112,015,535      cycles                           #    2.716 GHz                    
+    11,141,551,976      instructions                     #    2.18  insn per cycle         
+       1.883163972 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2527) (512y:  224) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_d_inl1_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -174,26 +174,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_d_inl1_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_d_inl1_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028807e+00
-Avg ME (F77/C++)    = 2.0288063388516200
-Relative difference = 3.2588037208240405e-07
+Avg ME (F77/C++)    = 2.0288063388516204
+Relative difference = 3.2588037186351226e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_d_inl1_hrd0/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.926324e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.149997e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.149997e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.149105e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.416003e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.416003e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     2.778213 sec
-     5,362,353,364      cycles                           #    1.927 GHz                    
-     9,296,610,904      instructions                     #    1.73  insn per cycle         
-       2.784120643 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2115) (512y:  282) (512z: 1958)
+TOTAL       :     2.637284 sec
+     4,829,728,502      cycles                           #    1.827 GHz                    
+     8,842,382,666      instructions                     #    1.83  insn per cycle         
+       2.644418009 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1821) (512y:   97) (512z: 2034)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_d_inl1_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
diff --git a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_d_inl1_hrd1.txt b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_d_inl1_hrd1.txt
index 223a6bbd07..8eec31c0d3 100644
--- a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_d_inl1_hrd1.txt
+++ b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_d_inl1_hrd1.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_d_inl1_hrd1'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
 
-DATE: 2023-11-24_15:01:02
+DATE: 2024-01-30_05:36:21
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl1_hrd1/gcheck.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 4.545579e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.156991e+08                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.274409e+08                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.565410e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.157958e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.274503e+08                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     0.524046 sec
-     2,220,143,747      cycles                           #    2.921 GHz                    
-     3,161,876,071      instructions                     #    1.42  insn per cycle         
-       0.819384909 seconds time elapsed
+TOTAL       :     0.528794 sec
+     2,178,979,969      cycles                           #    2.840 GHz                    
+     3,111,172,536      instructions                     #    1.43  insn per cycle         
+       0.825662442 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl1_hrd1/gcheck.exe -p 2048 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 208
@@ -72,20 +72,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl1_hrd1/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.624216e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.719163e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.719163e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.483554e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.573797e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.573797e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     4.103081 sec
-    12,357,400,963      cycles                           #    3.008 GHz                    
-    35,038,305,221      instructions                     #    2.84  insn per cycle         
-       4.108989255 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  457) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     4.331634 sec
+    12,513,147,299      cycles                           #    2.885 GHz                    
+    35,729,824,625      instructions                     #    2.86  insn per cycle         
+       4.338115382 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  469) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_d_inl1_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -99,20 +99,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_d_inl1_hrd1/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.981089e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.114332e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.114332e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.944859e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.193242e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.193242e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     3.624381 sec
-    10,699,873,531      cycles                           #    2.953 GHz                    
-    23,087,357,159      instructions                     #    2.16  insn per cycle         
-       3.630253133 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 2363) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     2.766913 sec
+     8,026,265,535      cycles                           #    2.895 GHz                    
+    21,260,291,484      instructions                     #    2.65  insn per cycle         
+       2.773559046 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 2088) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_d_inl1_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -126,20 +126,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_d_inl1_hrd1/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 5.093491e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 5.483832e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 5.483832e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.719292e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.240372e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.240372e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     2.165104 sec
-     6,157,585,167      cycles                           #    2.837 GHz                    
-    11,956,600,181      instructions                     #    1.94  insn per cycle         
-       2.170909922 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2509) (512y:    0) (512z:    0)
+TOTAL       :     1.943097 sec
+     5,300,809,350      cycles                           #    2.722 GHz                    
+    11,405,959,044      instructions                     #    2.15  insn per cycle         
+       1.950186269 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2370) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_d_inl1_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -153,20 +153,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_d_inl1_hrd1/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 5.214844e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 5.619393e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 5.619393e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.116224e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.720108e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.720108e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     2.116517 sec
-     6,017,617,016      cycles                           #    2.837 GHz                    
-    11,127,934,643      instructions                     #    1.85  insn per cycle         
-       2.122364748 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2126) (512y:  174) (512z:    0)
+TOTAL       :     1.828206 sec
+     4,977,318,735      cycles                           #    2.718 GHz                    
+    10,599,506,112      instructions                     #    2.13  insn per cycle         
+       1.834822870 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1970) (512y:  162) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_d_inl1_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -180,20 +180,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_d_inl1_hrd1/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 4.073709e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.313809e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.313809e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.275159e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.556705e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.556705e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     2.680169 sec
-     5,188,375,015      cycles                           #    1.932 GHz                    
-     9,020,642,676      instructions                     #    1.74  insn per cycle         
-       2.686120218 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1650) (512y:  208) (512z: 1567)
+TOTAL       :     2.563497 sec
+     4,703,376,134      cycles                           #    1.831 GHz                    
+     8,567,908,292      instructions                     #    1.82  insn per cycle         
+       2.570320519 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1392) (512y:   70) (512z: 1630)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_d_inl1_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
diff --git a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_f_inl0_hrd0.txt b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_f_inl0_hrd0.txt
index 5b41a4d066..03334a40e8 100644
--- a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_f_inl0_hrd0.txt
+++ b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_f_inl0_hrd0.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_f_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
 
-DATE: 2023-11-24_14:31:58
+DATE: 2024-01-30_04:56:08
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd0/gcheck.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 9.016432e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.646338e+08                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.972972e+08                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 8.266078e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.583524e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.962786e+08                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086718e+00 +- 3.413389e-03 )  GeV^0
-TOTAL       :     0.480351 sec
-     2,093,260,903      cycles                           #    2.971 GHz                    
-     2,988,705,515      instructions                     #    1.43  insn per cycle         
-       0.774971081 seconds time elapsed
+TOTAL       :     0.486964 sec
+     2,022,378,491      cycles                           #    2.826 GHz                    
+     2,872,554,108      instructions                     #    1.42  insn per cycle         
+       0.794836465 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd0/gcheck.exe -p 2048 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 128
@@ -72,20 +72,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd0/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.337300e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.415066e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.415066e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.220233e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.293728e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.293728e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086780e+00 +- 3.413794e-03 )  GeV^0
-TOTAL       :     4.572712 sec
-    14,005,022,680      cycles                           #    3.059 GHz                    
-    38,340,620,868      instructions                     #    2.74  insn per cycle         
-       4.581007114 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  587) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     4.812282 sec
+    13,901,639,181      cycles                           #    2.885 GHz                    
+    37,078,732,469      instructions                     #    2.67  insn per cycle         
+       4.824222975 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  578) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -93,26 +93,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028820e+00
-Avg ME (F77/C++)    = 2.0288199022179469
-Relative difference = 4.819651478256564e-08
+Avg ME (F77/C++)    = 2.0288197983754799
+Relative difference = 9.938019153537065e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_f_inl0_hrd0/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 5.261983e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 5.699946e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 5.699946e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.150516e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 5.595808e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.595808e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086779e+00 +- 3.413793e-03 )  GeV^0
-TOTAL       :     2.076354 sec
-     6,472,095,043      cycles                           #    3.109 GHz                    
-    15,815,873,405      instructions                     #    2.44  insn per cycle         
-       2.093110720 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 2690) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     2.124737 sec
+     6,168,101,005      cycles                           #    2.895 GHz                    
+    15,212,489,109      instructions                     #    2.47  insn per cycle         
+       2.142108549 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 2459) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -120,26 +120,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028819e+00
-Avg ME (F77/C++)    = 2.0288193548331037
-Relative difference = 1.748963824709674e-07
+Avg ME (F77/C++)    = 2.0288191968575120
+Relative difference = 9.703059369476286e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_f_inl0_hrd0/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 9.625847e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.106546e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.106546e+06                 )  sec^-1
-MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414231e-03 )  GeV^0
-TOTAL       :     1.173558 sec
-     3,465,806,997      cycles                           #    2.939 GHz                    
-     7,594,629,416      instructions                     #    2.19  insn per cycle         
-       1.191325878 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3049) (512y:    0) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 8.954385e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.029179e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.029179e+06                 )  sec^-1
+MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414230e-03 )  GeV^0
+TOTAL       :     1.259990 sec
+     3,437,290,204      cycles                           #    2.715 GHz                    
+     7,715,643,345      instructions                     #    2.24  insn per cycle         
+       1.287994689 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3071) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -147,26 +147,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028818e+00
-Avg ME (F77/C++)    = 2.0288181684445590
-Relative difference = 8.302595855806234e-08
+Avg ME (F77/C++)    = 2.0288179996423423
+Relative difference = 1.7628858734720142e-10
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_f_inl0_hrd0/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.036192e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.205283e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.205283e+06                 )  sec^-1
-MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414231e-03 )  GeV^0
-TOTAL       :     1.095666 sec
-     3,249,594,094      cycles                           #    2.950 GHz                    
-     7,203,151,855      instructions                     #    2.22  insn per cycle         
-       1.112919323 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2849) (512y:   23) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 9.805420e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.144112e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.144112e+06                 )  sec^-1
+MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414230e-03 )  GeV^0
+TOTAL       :     1.162105 sec
+     3,179,163,625      cycles                           #    2.727 GHz                    
+     7,109,925,739      instructions                     #    2.24  insn per cycle         
+       1.178171652 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2733) (512y:   13) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -174,26 +174,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028818e+00
-Avg ME (F77/C++)    = 2.0288181684445590
-Relative difference = 8.302595855806234e-08
+Avg ME (F77/C++)    = 2.0288179996423423
+Relative difference = 1.7628858734720142e-10
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_f_inl0_hrd0/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[16] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.709924e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 8.585781e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 8.585781e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 7.071814e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.862424e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.862424e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414231e-03 )  GeV^0
-TOTAL       :     1.443704 sec
-     3,056,175,378      cycles                           #    2.109 GHz                    
-     5,834,850,495      instructions                     #    1.91  insn per cycle         
-       1.456542629 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2363) (512y:   24) (512z: 1889)
+TOTAL       :     1.572617 sec
+     2,980,157,633      cycles                           #    1.888 GHz                    
+     5,763,820,562      instructions                     #    1.93  insn per cycle         
+       1.590552097 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2088) (512y:   20) (512z: 1914)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -201,8 +201,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028818e+00
-Avg ME (F77/C++)    = 2.0288183349184692
-Relative difference = 1.6508058850146622e-07
+Avg ME (F77/C++)    = 2.0288183195516467
+Relative difference = 1.5750631496822894e-07
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_f_inl0_hrd0_bridge.txt b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_f_inl0_hrd0_bridge.txt
index e2569bda32..3a80a864ae 100644
--- a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_f_inl0_hrd0_bridge.txt
+++ b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_f_inl0_hrd0_bridge.txt
@@ -41,7 +41,7 @@ CUDACPP_BUILDDIR='build.512z_f_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
 
-DATE: 2023-11-24_15:11:25
+DATE: 2024-01-30_05:47:08
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
@@ -51,17 +51,17 @@ WARNING! Bridge selected: cannot use RamboDevice, will use RamboHost
 WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
 WARNING! Instantiate device Bridge (nevt=524288, gpublocks=2048, gputhreads=256, gpublocks*gputhreads=524288)
 WARNING! Set grid in Bridge (nevt=524288, gpublocks=2048, gputhreads=256, gpublocks*gputhreads=524288)
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 7.002389e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 5.469916e+07                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 5.469916e+07                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.753522e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 5.358863e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.358863e+07                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086805e+00 +- 3.414078e-03 )  GeV^0
-TOTAL       :     0.673572 sec
-     2,669,669,937      cycles                           #    2.942 GHz                    
-     4,126,698,700      instructions                     #    1.55  insn per cycle         
-       0.966897204 seconds time elapsed
+TOTAL       :     0.684539 sec
+     2,591,525,623      cycles                           #    2.839 GHz                    
+     3,989,244,311      instructions                     #    1.54  insn per cycle         
+       0.972564077 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd0/gcheck.exe -p 2048 256 1 --bridge
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Bridge selected: cannot use RamboDevice, will use RamboHost
@@ -81,20 +81,20 @@ OK (relative difference <= 5E-3)
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd0/check.exe -p 2048 256 2 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=524288)
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.271859e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.346202e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.346202e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.212668e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.285744e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.285744e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086780e+00 +- 3.413794e-03 )  GeV^0
-TOTAL       :     4.744076 sec
-    14,203,194,064      cycles                           #    2.991 GHz                    
-    38,386,589,589      instructions                     #    2.70  insn per cycle         
-       4.750636160 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  587) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     4.870798 sec
+    14,070,285,227      cycles                           #    2.885 GHz                    
+    37,122,197,019      instructions                     #    2.64  insn per cycle         
+       4.878379515 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  578) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -102,27 +102,27 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028820e+00
-Avg ME (F77/C++)    = 2.0288199022179469
-Relative difference = 4.819651478256564e-08
+Avg ME (F77/C++)    = 2.0288197983754799
+Relative difference = 9.938019153537065e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_f_inl0_hrd0/check.exe -p 2048 256 2 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=524288)
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/sse4+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 5.023528e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 5.430819e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 5.430819e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.080420e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 5.515170e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.515170e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086779e+00 +- 3.413793e-03 )  GeV^0
-TOTAL       :     2.219353 sec
-     6,668,132,532      cycles                           #    2.997 GHz                    
-    16,095,906,246      instructions                     #    2.41  insn per cycle         
-       2.225914329 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 2690) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     2.198766 sec
+     6,358,773,769      cycles                           #    2.884 GHz                    
+    15,492,113,204      instructions                     #    2.44  insn per cycle         
+       2.206392318 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 2459) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -130,27 +130,27 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028819e+00
-Avg ME (F77/C++)    = 2.0288193548331037
-Relative difference = 1.748963824709674e-07
+Avg ME (F77/C++)    = 2.0288191968575120
+Relative difference = 9.703059369476286e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_f_inl0_hrd0/check.exe -p 2048 256 2 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=524288)
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/avx2+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 9.036698e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.035040e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.035040e+06                 )  sec^-1
-MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414231e-03 )  GeV^0
-TOTAL       :     1.291419 sec
-     3,685,819,556      cycles                           #    2.841 GHz                    
-     7,831,220,798      instructions                     #    2.12  insn per cycle         
-       1.298093181 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3049) (512y:    0) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 8.787706e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.007873e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.007873e+06                 )  sec^-1
+MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414230e-03 )  GeV^0
+TOTAL       :     1.328722 sec
+     3,633,771,509      cycles                           #    2.722 GHz                    
+     7,954,097,743      instructions                     #    2.19  insn per cycle         
+       1.336366634 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3071) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -158,27 +158,27 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028818e+00
-Avg ME (F77/C++)    = 2.0288181684445590
-Relative difference = 8.302595855806234e-08
+Avg ME (F77/C++)    = 2.0288179996423423
+Relative difference = 1.7628858734720142e-10
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_f_inl0_hrd0/check.exe -p 2048 256 2 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=524288)
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/512y+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 9.776675e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.132413e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.132413e+06                 )  sec^-1
-MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414231e-03 )  GeV^0
-TOTAL       :     1.202817 sec
-     3,439,533,821      cycles                           #    2.846 GHz                    
-     7,440,100,221      instructions                     #    2.16  insn per cycle         
-       1.209376262 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2849) (512y:   23) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 9.612179e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.118037e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.118037e+06                 )  sec^-1
+MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414230e-03 )  GeV^0
+TOTAL       :     1.225517 sec
+     3,366,927,421      cycles                           #    2.733 GHz                    
+     7,347,508,752      instructions                     #    2.18  insn per cycle         
+       1.232992993 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2733) (512y:   13) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -186,27 +186,27 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028818e+00
-Avg ME (F77/C++)    = 2.0288181684445590
-Relative difference = 8.302595855806234e-08
+Avg ME (F77/C++)    = 2.0288179996423423
+Relative difference = 1.7628858734720142e-10
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_f_inl0_hrd0/check.exe -p 2048 256 2 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=524288)
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/512z+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[16] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.099551e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.865069e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.865069e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.960005e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.722467e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.722467e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414231e-03 )  GeV^0
-TOTAL       :     1.609144 sec
-     3,270,738,224      cycles                           #    2.025 GHz                    
-     6,088,987,367      instructions                     #    1.86  insn per cycle         
-       1.615635336 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2363) (512y:   24) (512z: 1889)
+TOTAL       :     1.642570 sec
+     3,181,631,608      cycles                           #    1.930 GHz                    
+     6,021,725,956      instructions                     #    1.89  insn per cycle         
+       1.650041277 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2088) (512y:   20) (512z: 1914)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -214,8 +214,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028818e+00
-Avg ME (F77/C++)    = 2.0288183349184692
-Relative difference = 1.6508058850146622e-07
+Avg ME (F77/C++)    = 2.0288183195516467
+Relative difference = 1.5750631496822894e-07
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_f_inl0_hrd0_common.txt b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_f_inl0_hrd0_common.txt
index 1279e79222..38a7216065 100644
--- a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_f_inl0_hrd0_common.txt
+++ b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_f_inl0_hrd0_common.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_f_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
 
-DATE: 2023-11-24_15:24:46
+DATE: 2024-01-30_06:00:59
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd0/gcheck.exe -p 2048 256 2 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:COMMON+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 9.310797e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.615494e+08                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.945034e+08                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 9.412461e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.631522e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.951868e+08                 )  sec^-1
 MeanMatrixElemValue         = ( 2.079446e+00 +- 3.403306e-03 )  GeV^0
-TOTAL       :     0.567939 sec
-     2,336,393,305      cycles                           #    2.964 GHz                    
-     3,415,377,807      instructions                     #    1.46  insn per cycle         
-       0.847416380 seconds time elapsed
+TOTAL       :     0.573986 sec
+     2,244,851,144      cycles                           #    2.822 GHz                    
+     3,300,445,554      instructions                     #    1.47  insn per cycle         
+       0.853607464 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd0/gcheck.exe -p 2048 256 1 --common
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 128
@@ -72,20 +72,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd0/check.exe -p 2048 256 2 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:COMMON+RMBHST+MESHST/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.285858e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.359466e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.359466e+05                 )  sec^-1
-MeanMatrixElemValue         = ( 2.079573e+00 +- 3.404712e-03 )  GeV^0
-TOTAL       :     4.728147 sec
-    14,182,471,733      cycles                           #    2.997 GHz                    
-    38,371,003,858      instructions                     #    2.71  insn per cycle         
-       4.733810751 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  587) (avx2:    0) (512y:    0) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 2.218192e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.291861e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.291861e+05                 )  sec^-1
+MeanMatrixElemValue         = ( 2.079572e+00 +- 3.404712e-03 )  GeV^0
+TOTAL       :     4.876695 sec
+    14,064,697,494      cycles                           #    2.884 GHz                    
+    37,110,369,611      instructions                     #    2.64  insn per cycle         
+       4.882981134 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  578) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -93,26 +93,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028820e+00
-Avg ME (F77/C++)    = 2.0288199022179469
-Relative difference = 4.819651478256564e-08
+Avg ME (F77/C++)    = 2.0288197983754799
+Relative difference = 9.938019153537065e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_f_inl0_hrd0/check.exe -p 2048 256 2 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:COMMON+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 4.888028e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 5.282207e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 5.282207e+05                 )  sec^-1
-MeanMatrixElemValue         = ( 2.079572e+00 +- 3.404712e-03 )  GeV^0
-TOTAL       :     2.288230 sec
-     6,645,756,806      cycles                           #    2.898 GHz                    
-    15,828,500,466      instructions                     #    2.38  insn per cycle         
-       2.293948357 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 2690) (avx2:    0) (512y:    0) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 5.131220e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 5.575839e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.575839e+05                 )  sec^-1
+MeanMatrixElemValue         = ( 2.079572e+00 +- 3.404711e-03 )  GeV^0
+TOTAL       :     2.187823 sec
+     6,322,431,284      cycles                           #    2.883 GHz                    
+    15,223,876,723      instructions                     #    2.41  insn per cycle         
+       2.194184928 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 2459) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -120,26 +120,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028819e+00
-Avg ME (F77/C++)    = 2.0288193548331037
-Relative difference = 1.748963824709674e-07
+Avg ME (F77/C++)    = 2.0288191968575120
+Relative difference = 9.703059369476286e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_f_inl0_hrd0/check.exe -p 2048 256 2 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:COMMON+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 9.344568e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.074101e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.074101e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 8.948892e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.027773e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.027773e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 2.079550e+00 +- 3.404207e-03 )  GeV^0
-TOTAL       :     1.263326 sec
-     3,638,417,018      cycles                           #    2.869 GHz                    
-     7,579,208,332      instructions                     #    2.08  insn per cycle         
-       1.269045825 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3049) (512y:    0) (512z:    0)
+TOTAL       :     1.320835 sec
+     3,601,071,923      cycles                           #    2.719 GHz                    
+     7,699,828,133      instructions                     #    2.14  insn per cycle         
+       1.327138068 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3071) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -147,26 +147,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028818e+00
-Avg ME (F77/C++)    = 2.0288181684445590
-Relative difference = 8.302595855806234e-08
+Avg ME (F77/C++)    = 2.0288179996423423
+Relative difference = 1.7628858734720142e-10
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_f_inl0_hrd0/check.exe -p 2048 256 2 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:COMMON+RMBHST+MESHST/512y+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.006852e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.170884e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.170884e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 9.790537e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.142626e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.142626e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 2.079550e+00 +- 3.404207e-03 )  GeV^0
-TOTAL       :     1.182767 sec
-     3,425,925,806      cycles                           #    2.885 GHz                    
-     7,153,954,116      instructions                     #    2.09  insn per cycle         
-       1.188401535 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2849) (512y:   23) (512z:    0)
+TOTAL       :     1.218690 sec
+     3,342,798,362      cycles                           #    2.731 GHz                    
+     7,059,572,278      instructions                     #    2.11  insn per cycle         
+       1.225217680 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2733) (512y:   13) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -174,26 +174,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028818e+00
-Avg ME (F77/C++)    = 2.0288181684445590
-Relative difference = 8.302595855806234e-08
+Avg ME (F77/C++)    = 2.0288179996423423
+Relative difference = 1.7628858734720142e-10
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_f_inl0_hrd0/check.exe -p 2048 256 2 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:COMMON+RMBHST+MESHST/512z+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[16] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.384551e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 8.223950e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 8.223950e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 7.022088e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.806836e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.806836e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.079550e+00 +- 3.404208e-03 )  GeV^0
-TOTAL       :     1.561958 sec
-     3,241,666,935      cycles                           #    2.069 GHz                    
-     5,785,669,906      instructions                     #    1.78  insn per cycle         
-       1.567670740 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2363) (512y:   24) (512z: 1889)
+TOTAL       :     1.641075 sec
+     3,147,503,652      cycles                           #    1.912 GHz                    
+     5,713,849,148      instructions                     #    1.82  insn per cycle         
+       1.647331874 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2088) (512y:   20) (512z: 1914)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -201,8 +201,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028818e+00
-Avg ME (F77/C++)    = 2.0288183349184692
-Relative difference = 1.6508058850146622e-07
+Avg ME (F77/C++)    = 2.0288183195516467
+Relative difference = 1.5750631496822894e-07
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_f_inl0_hrd0_curhst.txt b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_f_inl0_hrd0_curhst.txt
index c75f243f6b..cb54d3236b 100644
--- a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_f_inl0_hrd0_curhst.txt
+++ b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_f_inl0_hrd0_curhst.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_f_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
 
-DATE: 2023-11-24_15:21:25
+DATE: 2024-01-30_05:57:29
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd0/gcheck.exe -p 2048 256 2 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 9.425944e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.638582e+08                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.958027e+08                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 9.414196e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.655173e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.981062e+08                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086718e+00 +- 3.413389e-03 )  GeV^0
-TOTAL       :     0.512465 sec
-     2,128,740,635      cycles                           #    2.913 GHz                    
-     3,349,486,673      instructions                     #    1.57  insn per cycle         
-       0.789824764 seconds time elapsed
+TOTAL       :     0.514856 sec
+     2,089,539,478      cycles                           #    2.840 GHz                    
+     3,296,506,746      instructions                     #    1.58  insn per cycle         
+       0.794530995 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd0/gcheck.exe -p 2048 256 1 --curhst
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 128
@@ -72,20 +72,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd0/check.exe -p 2048 256 2 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.258814e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.334265e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.334265e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.227183e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.300870e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.300870e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086780e+00 +- 3.413794e-03 )  GeV^0
-TOTAL       :     4.729033 sec
-    14,008,183,941      cycles                           #    2.959 GHz                    
-    38,341,321,376      instructions                     #    2.74  insn per cycle         
-       4.734430894 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  587) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     4.796202 sec
+    13,896,514,461      cycles                           #    2.894 GHz                    
+    37,078,595,071      instructions                     #    2.67  insn per cycle         
+       4.803618427 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  578) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -93,26 +93,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028820e+00
-Avg ME (F77/C++)    = 2.0288199022179469
-Relative difference = 4.819651478256564e-08
+Avg ME (F77/C++)    = 2.0288197983754799
+Relative difference = 9.938019153537065e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_f_inl0_hrd0/check.exe -p 2048 256 2 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 4.915242e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 5.303544e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 5.303544e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.077190e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 5.527451e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.527451e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086779e+00 +- 3.413793e-03 )  GeV^0
-TOTAL       :     2.218605 sec
-     6,466,211,293      cycles                           #    2.909 GHz                    
-    15,815,641,765      instructions                     #    2.45  insn per cycle         
-       2.223961450 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 2690) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     2.154367 sec
+     6,177,704,022      cycles                           #    2.870 GHz                    
+    15,215,532,210      instructions                     #    2.46  insn per cycle         
+       2.160620609 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 2459) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -120,26 +120,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028819e+00
-Avg ME (F77/C++)    = 2.0288193548331037
-Relative difference = 1.748963824709674e-07
+Avg ME (F77/C++)    = 2.0288191968575120
+Relative difference = 9.703059369476286e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_f_inl0_hrd0/check.exe -p 2048 256 2 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 9.288130e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.067419e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.067419e+06                 )  sec^-1
-MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414231e-03 )  GeV^0
-TOTAL       :     1.215145 sec
-     3,455,872,602      cycles                           #    2.833 GHz                    
-     7,593,852,686      instructions                     #    2.20  insn per cycle         
-       1.220604147 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3049) (512y:    0) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 8.911398e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.023141e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.023141e+06                 )  sec^-1
+MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414230e-03 )  GeV^0
+TOTAL       :     1.265022 sec
+     3,447,761,650      cycles                           #    2.714 GHz                    
+     7,715,058,636      instructions                     #    2.24  insn per cycle         
+       1.271511064 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3071) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -147,26 +147,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028818e+00
-Avg ME (F77/C++)    = 2.0288181684445590
-Relative difference = 8.302595855806234e-08
+Avg ME (F77/C++)    = 2.0288179996423423
+Relative difference = 1.7628858734720142e-10
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_f_inl0_hrd0/check.exe -p 2048 256 2 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 9.949365e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.155182e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.155182e+06                 )  sec^-1
-MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414231e-03 )  GeV^0
-TOTAL       :     1.138566 sec
-     3,249,226,385      cycles                           #    2.842 GHz                    
-     7,201,814,152      instructions                     #    2.22  insn per cycle         
-       1.144215776 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2849) (512y:   23) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 9.829060e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.147412e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.147412e+06                 )  sec^-1
+MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414230e-03 )  GeV^0
+TOTAL       :     1.155347 sec
+     3,170,001,813      cycles                           #    2.731 GHz                    
+     7,109,524,161      instructions                     #    2.24  insn per cycle         
+       1.161808340 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2733) (512y:   13) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -174,26 +174,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028818e+00
-Avg ME (F77/C++)    = 2.0288181684445590
-Relative difference = 8.302595855806234e-08
+Avg ME (F77/C++)    = 2.0288179996423423
+Relative difference = 1.7628858734720142e-10
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_f_inl0_hrd0/check.exe -p 2048 256 2 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[16] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.332937e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 8.137553e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 8.137553e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.999480e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.774350e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.774350e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414231e-03 )  GeV^0
-TOTAL       :     1.516868 sec
-     3,052,773,549      cycles                           #    2.006 GHz                    
-     5,834,286,886      instructions                     #    1.91  insn per cycle         
-       1.522319115 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2363) (512y:   24) (512z: 1889)
+TOTAL       :     1.586909 sec
+     2,978,718,352      cycles                           #    1.871 GHz                    
+     5,762,941,941      instructions                     #    1.93  insn per cycle         
+       1.593095591 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2088) (512y:   20) (512z: 1914)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -201,8 +201,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028818e+00
-Avg ME (F77/C++)    = 2.0288183349184692
-Relative difference = 1.6508058850146622e-07
+Avg ME (F77/C++)    = 2.0288183195516467
+Relative difference = 1.5750631496822894e-07
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_f_inl0_hrd0_rmbhst.txt b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_f_inl0_hrd0_rmbhst.txt
index bfc5cc0709..5939268227 100644
--- a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_f_inl0_hrd0_rmbhst.txt
+++ b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_f_inl0_hrd0_rmbhst.txt
@@ -41,24 +41,24 @@ CUDACPP_BUILDDIR='build.512z_f_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
 
-DATE: 2023-11-24_15:18:07
+DATE: 2024-01-30_05:54:04
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd0/gcheck.exe -p 2048 256 2 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 8.776083e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.627754e+08                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.948386e+08                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 8.468280e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.632924e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.955888e+08                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086805e+00 +- 3.414078e-03 )  GeV^0
-TOTAL       :     0.618119 sec
-     2,443,289,573      cycles                           #    2.920 GHz                    
-     3,806,362,954      instructions                     #    1.56  insn per cycle         
-       0.895384570 seconds time elapsed
+TOTAL       :     0.625640 sec
+     2,402,533,839      cycles                           #    2.841 GHz                    
+     3,758,306,095      instructions                     #    1.56  insn per cycle         
+       0.905223049 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd0/gcheck.exe -p 2048 256 1 --rmbhst
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
@@ -74,20 +74,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd0/check.exe -p 2048 256 2 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.285771e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.360708e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.360708e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.221197e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.294941e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.294941e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086780e+00 +- 3.413794e-03 )  GeV^0
-TOTAL       :     4.672900 sec
-    14,007,447,781      cycles                           #    2.995 GHz                    
-    38,341,188,487      instructions                     #    2.74  insn per cycle         
-       4.678325774 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  587) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     4.809027 sec
+    13,889,421,482      cycles                           #    2.885 GHz                    
+    37,078,742,557      instructions                     #    2.67  insn per cycle         
+       4.815296717 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  578) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -95,26 +95,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028820e+00
-Avg ME (F77/C++)    = 2.0288199022179469
-Relative difference = 4.819651478256564e-08
+Avg ME (F77/C++)    = 2.0288197983754799
+Relative difference = 9.938019153537065e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_f_inl0_hrd0/check.exe -p 2048 256 2 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 5.077746e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 5.495127e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 5.495127e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.146065e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 5.592205e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.592205e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086779e+00 +- 3.413793e-03 )  GeV^0
-TOTAL       :     2.152723 sec
-     6,468,101,195      cycles                           #    2.999 GHz                    
-    15,815,386,942      instructions                     #    2.45  insn per cycle         
-       2.158120643 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 2690) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     2.125330 sec
+     6,161,438,553      cycles                           #    2.892 GHz                    
+    15,211,397,983      instructions                     #    2.47  insn per cycle         
+       2.131726868 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 2459) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -122,26 +122,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028819e+00
-Avg ME (F77/C++)    = 2.0288193548331037
-Relative difference = 1.748963824709674e-07
+Avg ME (F77/C++)    = 2.0288191968575120
+Relative difference = 9.703059369476286e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_f_inl0_hrd0/check.exe -p 2048 256 2 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 9.236653e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.062344e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.062344e+06                 )  sec^-1
-MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414231e-03 )  GeV^0
-TOTAL       :     1.220471 sec
-     3,470,157,914      cycles                           #    2.832 GHz                    
-     7,593,755,062      instructions                     #    2.19  insn per cycle         
-       1.225848326 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3049) (512y:    0) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 8.991330e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.034099e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.034099e+06                 )  sec^-1
+MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414230e-03 )  GeV^0
+TOTAL       :     1.255088 sec
+     3,440,029,043      cycles                           #    2.730 GHz                    
+     7,714,775,848      instructions                     #    2.24  insn per cycle         
+       1.261283713 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3071) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -149,26 +149,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028818e+00
-Avg ME (F77/C++)    = 2.0288181684445590
-Relative difference = 8.302595855806234e-08
+Avg ME (F77/C++)    = 2.0288179996423423
+Relative difference = 1.7628858734720142e-10
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_f_inl0_hrd0/check.exe -p 2048 256 2 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 9.915047e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.151420e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.151420e+06                 )  sec^-1
-MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414231e-03 )  GeV^0
-TOTAL       :     1.142912 sec
-     3,256,437,189      cycles                           #    2.838 GHz                    
-     7,202,634,682      instructions                     #    2.21  insn per cycle         
-       1.148660140 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2849) (512y:   23) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 9.843583e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.149362e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.149362e+06                 )  sec^-1
+MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414230e-03 )  GeV^0
+TOTAL       :     1.153935 sec
+     3,172,826,861      cycles                           #    2.738 GHz                    
+     7,109,210,779      instructions                     #    2.24  insn per cycle         
+       1.160268530 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2733) (512y:   13) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -176,26 +176,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028818e+00
-Avg ME (F77/C++)    = 2.0288181684445590
-Relative difference = 8.302595855806234e-08
+Avg ME (F77/C++)    = 2.0288179996423423
+Relative difference = 1.7628858734720142e-10
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_f_inl0_hrd0/check.exe -p 2048 256 2 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[16] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.280239e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 8.088898e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 8.088898e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 7.077925e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.872855e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.872855e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414231e-03 )  GeV^0
-TOTAL       :     1.526066 sec
-     3,058,452,115      cycles                           #    1.998 GHz                    
-     5,834,337,483      instructions                     #    1.91  insn per cycle         
-       1.531519039 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2363) (512y:   24) (512z: 1889)
+TOTAL       :     1.570762 sec
+     2,979,903,068      cycles                           #    1.891 GHz                    
+     5,762,829,882      instructions                     #    1.93  insn per cycle         
+       1.577195857 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2088) (512y:   20) (512z: 1914)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -203,8 +203,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028818e+00
-Avg ME (F77/C++)    = 2.0288183349184692
-Relative difference = 1.6508058850146622e-07
+Avg ME (F77/C++)    = 2.0288183195516467
+Relative difference = 1.5750631496822894e-07
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_f_inl0_hrd1.txt b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_f_inl0_hrd1.txt
index 5c8f13a099..c96a0bb3db 100644
--- a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_f_inl0_hrd1.txt
+++ b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_f_inl0_hrd1.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_f_inl0_hrd1'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
 
-DATE: 2023-11-24_14:32:21
+DATE: 2024-01-30_04:56:32
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd1/gcheck.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 9.367052e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.685993e+08                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.017698e+08                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 8.421312e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.704045e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.041754e+08                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086718e+00 +- 3.413389e-03 )  GeV^0
-TOTAL       :     0.471621 sec
-     2,109,591,981      cycles                           #    3.033 GHz                    
-     3,025,510,144      instructions                     #    1.43  insn per cycle         
-       0.774479825 seconds time elapsed
+TOTAL       :     0.486618 sec
+     2,018,521,842      cycles                           #    2.827 GHz                    
+     2,837,894,141      instructions                     #    1.41  insn per cycle         
+       0.795623791 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd1/gcheck.exe -p 2048 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 127
@@ -72,20 +72,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd1/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.266296e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.337160e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.337160e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.245629e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.320181e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.320181e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086780e+00 +- 3.413794e-03 )  GeV^0
-TOTAL       :     4.710609 sec
-    14,404,518,200      cycles                           #    3.055 GHz                    
-    39,833,821,591      instructions                     #    2.77  insn per cycle         
-       4.718963135 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  570) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     4.758439 sec
+    13,805,800,630      cycles                           #    2.898 GHz                    
+    37,480,161,839      instructions                     #    2.71  insn per cycle         
+       4.770650257 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  503) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -93,26 +93,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl0_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028820e+00
-Avg ME (F77/C++)    = 2.0288199028000236
-Relative difference = 4.790961076489297e-08
+Avg ME (F77/C++)    = 2.0288197983754799
+Relative difference = 9.938019153537065e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_f_inl0_hrd1/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 6.070167e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 6.657007e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 6.657007e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.821274e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.398672e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.398672e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086779e+00 +- 3.413793e-03 )  GeV^0
-TOTAL       :     1.811511 sec
-     5,577,453,918      cycles                           #    3.070 GHz                    
-    15,285,624,341      instructions                     #    2.74  insn per cycle         
-       1.824268128 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 2474) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     1.889404 sec
+     5,475,292,589      cycles                           #    2.889 GHz                    
+    15,244,893,114      instructions                     #    2.78  insn per cycle         
+       1.908184587 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 2330) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_f_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -120,26 +120,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_f_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_f_inl0_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028819e+00
-Avg ME (F77/C++)    = 2.0288193548331037
-Relative difference = 1.748963824709674e-07
+Avg ME (F77/C++)    = 2.0288191968575120
+Relative difference = 9.703059369476286e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_f_inl0_hrd1/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 6.861603e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.557183e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.557183e+05                 )  sec^-1
-MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414231e-03 )  GeV^0
-TOTAL       :     1.611040 sec
-     4,742,441,153      cycles                           #    2.934 GHz                    
-     9,735,233,054      instructions                     #    2.05  insn per cycle         
-       1.623131424 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3708) (512y:    0) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 6.385813e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.037637e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.037637e+05                 )  sec^-1
+MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414230e-03 )  GeV^0
+TOTAL       :     1.731302 sec
+     4,719,001,422      cycles                           #    2.717 GHz                    
+     9,850,811,081      instructions                     #    2.09  insn per cycle         
+       1.750777348 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3750) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_f_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -147,26 +147,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_f_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_f_inl0_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028818e+00
-Avg ME (F77/C++)    = 2.0288182108197361
-Relative difference = 1.0391259163456515e-07
+Avg ME (F77/C++)    = 2.0288180243223906
+Relative difference = 1.1988453753912676e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_f_inl0_hrd1/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 6.988259e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.740532e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.740532e+05                 )  sec^-1
-MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414231e-03 )  GeV^0
-TOTAL       :     1.584956 sec
-     4,623,129,145      cycles                           #    2.906 GHz                    
-     9,326,921,854      instructions                     #    2.02  insn per cycle         
-       1.599836475 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3496) (512y:    0) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 6.683577e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.409489e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.409489e+05                 )  sec^-1
+MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414230e-03 )  GeV^0
+TOTAL       :     1.659075 sec
+     4,492,699,411      cycles                           #    2.699 GHz                    
+     9,202,452,349      instructions                     #    2.05  insn per cycle         
+       1.671352513 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3384) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_f_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -174,26 +174,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_f_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_f_inl0_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028818e+00
-Avg ME (F77/C++)    = 2.0288182108197361
-Relative difference = 1.0391259163456515e-07
+Avg ME (F77/C++)    = 2.0288180243223906
+Relative difference = 1.1988453753912676e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_f_inl0_hrd1/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[16] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 6.277364e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 6.853368e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 6.853368e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.938211e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.486110e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.486110e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414231e-03 )  GeV^0
-TOTAL       :     1.753984 sec
-     3,644,734,123      cycles                           #    2.072 GHz                    
-     7,034,553,715      instructions                     #    1.93  insn per cycle         
-       1.766522849 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2605) (512y:   12) (512z: 2221)
+TOTAL       :     1.854854 sec
+     3,463,720,216      cycles                           #    1.861 GHz                    
+     6,875,040,962      instructions                     #    1.98  insn per cycle         
+       1.876340349 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2257) (512y:    8) (512z: 2261)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_f_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -201,8 +201,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_f_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_f_inl0_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028818e+00
-Avg ME (F77/C++)    = 2.0288183459779248
-Relative difference = 1.7053177021099307e-07
+Avg ME (F77/C++)    = 2.0288183217635378
+Relative difference = 1.5859655131013432e-07
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_f_inl1_hrd0.txt b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_f_inl1_hrd0.txt
index 665123002a..993f4107d6 100644
--- a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_f_inl1_hrd0.txt
+++ b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_f_inl1_hrd0.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_f_inl1_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
 
-DATE: 2023-11-24_15:01:29
+DATE: 2024-01-30_05:36:48
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl1_hrd0/gcheck.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 9.406259e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.654901e+08                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.974381e+08                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 9.377362e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.649325e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.974675e+08                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086718e+00 +- 3.413389e-03 )  GeV^0
-TOTAL       :     0.478015 sec
-     2,052,116,202      cycles                           #    2.923 GHz                    
-     2,932,366,143      instructions                     #    1.43  insn per cycle         
-       0.759588643 seconds time elapsed
+TOTAL       :     0.484150 sec
+     2,005,186,574      cycles                           #    2.831 GHz                    
+     2,872,226,914      instructions                     #    1.43  insn per cycle         
+       0.768013554 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl1_hrd0/gcheck.exe -p 2048 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 128
@@ -72,20 +72,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl1_hrd0/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.529261e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.621856e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.621856e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.479081e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.570421e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.570421e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086780e+00 +- 3.413794e-03 )  GeV^0
-TOTAL       :     4.232103 sec
-    12,595,483,741      cycles                           #    2.973 GHz                    
-    34,372,550,033      instructions                     #    2.73  insn per cycle         
-       4.237616805 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  696) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     4.318821 sec
+    12,411,469,267      cycles                           #    2.871 GHz                    
+    34,216,954,204      instructions                     #    2.76  insn per cycle         
+       4.325006925 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  768) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl1_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -93,26 +93,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl1_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl1_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028820e+00
-Avg ME (F77/C++)    = 2.0288199094356969
-Relative difference = 4.463890496342449e-08
+Avg ME (F77/C++)    = 2.0288199088536203
+Relative difference = 4.4925808981097166e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_f_inl1_hrd0/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 5.409282e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 5.897832e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 5.897832e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.935196e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.540988e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.540988e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086779e+00 +- 3.413793e-03 )  GeV^0
-TOTAL       :     2.025230 sec
-     6,107,661,405      cycles                           #    3.009 GHz                    
-    14,860,133,021      instructions                     #    2.43  insn per cycle         
-       2.031040801 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 3009) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     1.856505 sec
+     5,363,525,325      cycles                           #    2.881 GHz                    
+    14,587,825,944      instructions                     #    2.72  insn per cycle         
+       1.863141926 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 2947) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_f_inl1_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -120,26 +120,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_f_inl1_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_f_inl1_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028819e+00
-Avg ME (F77/C++)    = 2.0288193803280592
-Relative difference = 1.8746278463897685e-07
+Avg ME (F77/C++)    = 2.0288192580919713
+Relative difference = 1.2721291123071246e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_f_inl1_hrd0/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.413017e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 8.265518e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 8.265518e+05                 )  sec^-1
-MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414231e-03 )  GeV^0
-TOTAL       :     1.499210 sec
-     4,274,397,689      cycles                           #    2.842 GHz                    
-     9,028,681,209      instructions                     #    2.11  insn per cycle         
-       1.505009031 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 4443) (512y:    0) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 7.475828e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 8.385170e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.385170e+05                 )  sec^-1
+MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414230e-03 )  GeV^0
+TOTAL       :     1.492139 sec
+     4,058,079,431      cycles                           #    2.710 GHz                    
+     9,088,895,483      instructions                     #    2.24  insn per cycle         
+       1.498802038 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 4501) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_f_inl1_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -147,26 +147,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_f_inl1_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_f_inl1_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028818e+00
-Avg ME (F77/C++)    = 2.0288181999931112
-Relative difference = 9.857617164523888e-08
+Avg ME (F77/C++)    = 2.0288180499337614
+Relative difference = 2.4612242975974814e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_f_inl1_hrd0/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.545085e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 8.419779e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 8.419779e+05                 )  sec^-1
-MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414231e-03 )  GeV^0
-TOTAL       :     1.475217 sec
-     4,206,901,352      cycles                           #    2.842 GHz                    
-     8,664,206,183      instructions                     #    2.06  insn per cycle         
-       1.480947152 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 4243) (512y:    0) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 8.052179e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 9.125609e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.125609e+05                 )  sec^-1
+MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414230e-03 )  GeV^0
+TOTAL       :     1.390912 sec
+     3,795,132,868      cycles                           #    2.718 GHz                    
+     8,440,638,214      instructions                     #    2.22  insn per cycle         
+       1.397579629 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 4043) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_f_inl1_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -174,26 +174,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_f_inl1_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_f_inl1_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028818e+00
-Avg ME (F77/C++)    = 2.0288181999931112
-Relative difference = 9.857617164523888e-08
+Avg ME (F77/C++)    = 2.0288180499337614
+Relative difference = 2.4612242975974814e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_f_inl1_hrd0/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[16] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 5.444457e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 5.903029e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 5.903029e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.426211e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 5.889827e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.889827e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414231e-03 )  GeV^0
-TOTAL       :     2.015920 sec
-     3,844,199,069      cycles                           #    1.909 GHz                    
-     7,810,135,811      instructions                     #    2.03  insn per cycle         
-       2.021868576 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 4419) (512y:    0) (512z: 2556)
+TOTAL       :     2.021883 sec
+     3,727,709,927      cycles                           #    1.839 GHz                    
+     7,572,021,248      instructions                     #    2.03  insn per cycle         
+       2.028341317 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3646) (512y:    1) (512z: 2853)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_f_inl1_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -201,8 +201,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_f_inl1_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_f_inl1_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028818e+00
-Avg ME (F77/C++)    = 2.0288183246739209
-Relative difference = 1.6003107281264138e-07
+Avg ME (F77/C++)    = 2.0288183350348845
+Relative difference = 1.6513796936156652e-07
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_f_inl1_hrd1.txt b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_f_inl1_hrd1.txt
index fa97bf17a1..2891f046ff 100644
--- a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_f_inl1_hrd1.txt
+++ b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_f_inl1_hrd1.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_f_inl1_hrd1'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
 
-DATE: 2023-11-24_15:01:53
+DATE: 2024-01-30_05:37:12
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl1_hrd1/gcheck.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 9.425807e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.669791e+08                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.006545e+08                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 9.485748e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.689974e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.027356e+08                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086718e+00 +- 3.413389e-03 )  GeV^0
-TOTAL       :     0.480700 sec
-     2,057,638,891      cycles                           #    2.914 GHz                    
-     2,939,721,478      instructions                     #    1.43  insn per cycle         
-       0.763589412 seconds time elapsed
+TOTAL       :     0.482102 sec
+     1,996,662,355      cycles                           #    2.812 GHz                    
+     2,850,200,230      instructions                     #    1.43  insn per cycle         
+       0.768087139 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl1_hrd1/gcheck.exe -p 2048 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 127
@@ -72,20 +72,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl1_hrd1/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.734181e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.846132e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.846132e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.596095e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.696763e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.696763e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086780e+00 +- 3.413794e-03 )  GeV^0
-TOTAL       :     3.923431 sec
-    11,754,442,627      cycles                           #    2.992 GHz                    
-    35,109,278,353      instructions                     #    2.99  insn per cycle         
-       3.929236394 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  470) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     4.127183 sec
+    11,946,394,247      cycles                           #    2.891 GHz                    
+    35,407,075,530      instructions                     #    2.96  insn per cycle         
+       4.133301161 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  469) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl1_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -93,26 +93,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl1_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_f_inl1_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028820e+00
-Avg ME (F77/C++)    = 2.0288199094356969
-Relative difference = 4.463890496342449e-08
+Avg ME (F77/C++)    = 2.0288199088536203
+Relative difference = 4.4925808981097166e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_f_inl1_hrd1/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 5.529892e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 6.033657e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 6.033657e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.250434e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.927787e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.927787e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086779e+00 +- 3.413793e-03 )  GeV^0
-TOTAL       :     1.982127 sec
-     5,950,643,888      cycles                           #    2.995 GHz                    
-    14,469,850,336      instructions                     #    2.43  insn per cycle         
-       1.987737448 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 2572) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     1.766919 sec
+     5,069,845,731      cycles                           #    2.861 GHz                    
+    14,044,971,447      instructions                     #    2.77  insn per cycle         
+       1.773365949 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 2487) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_f_inl1_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -120,26 +120,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_f_inl1_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_f_inl1_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028819e+00
-Avg ME (F77/C++)    = 2.0288193583255634
-Relative difference = 1.7661780742548925e-07
+Avg ME (F77/C++)    = 2.0288192554144189
+Relative difference = 1.2589315209891237e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_f_inl1_hrd1/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.598308e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 8.516031e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 8.516031e+05                 )  sec^-1
-MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414231e-03 )  GeV^0
-TOTAL       :     1.464407 sec
-     4,161,674,169      cycles                           #    2.832 GHz                    
-     8,874,846,514      instructions                     #    2.13  insn per cycle         
-       1.469970354 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3574) (512y:    0) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 7.559784e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 8.492213e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.492213e+05                 )  sec^-1
+MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414230e-03 )  GeV^0
+TOTAL       :     1.476405 sec
+     3,988,953,115      cycles                           #    2.692 GHz                    
+     8,629,569,798      instructions                     #    2.16  insn per cycle         
+       1.482936821 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3422) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_f_inl1_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -147,26 +147,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_f_inl1_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_f_inl1_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028818e+00
-Avg ME (F77/C++)    = 2.0288182107033208
-Relative difference = 1.0385521077446488e-07
+Avg ME (F77/C++)    = 2.0288180815987289
+Relative difference = 4.021983692325164e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_f_inl1_hrd1/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.657230e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 8.559591e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 8.559591e+05                 )  sec^-1
-MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414231e-03 )  GeV^0
-TOTAL       :     1.454835 sec
-     4,142,234,613      cycles                           #    2.837 GHz                    
-     8,411,473,385      instructions                     #    2.03  insn per cycle         
-       1.460527785 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3319) (512y:    0) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 8.210818e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 9.331985e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.331985e+05                 )  sec^-1
+MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414230e-03 )  GeV^0
+TOTAL       :     1.366166 sec
+     3,694,176,022      cycles                           #    2.694 GHz                    
+     8,100,845,822      instructions                     #    2.19  insn per cycle         
+       1.372646371 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3105) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_f_inl1_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -174,26 +174,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_f_inl1_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_f_inl1_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028818e+00
-Avg ME (F77/C++)    = 2.0288182107033208
-Relative difference = 1.0385521077446488e-07
+Avg ME (F77/C++)    = 2.0288180815987289
+Relative difference = 4.021983692325164e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_f_inl1_hrd1/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[16] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 5.747619e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 6.236667e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 6.236667e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.670710e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.170464e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.170464e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086810e+00 +- 3.414231e-03 )  GeV^0
-TOTAL       :     1.909420 sec
-     3,779,842,439      cycles                           #    1.975 GHz                    
-     7,700,519,284      instructions                     #    2.04  insn per cycle         
-       1.914942288 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3435) (512y:    0) (512z: 2108)
+TOTAL       :     1.938240 sec
+     3,580,879,514      cycles                           #    1.843 GHz                    
+     7,373,942,234      instructions                     #    2.06  insn per cycle         
+       1.944698982 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2803) (512y:    1) (512z: 2230)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_f_inl1_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -201,8 +201,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_f_inl1_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_f_inl1_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028818e+00
-Avg ME (F77/C++)    = 2.0288183204829693
-Relative difference = 1.5796536184903122e-07
+Avg ME (F77/C++)    = 2.0288183569209650
+Relative difference = 1.7592557106041962e-07
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_m_inl0_hrd0.txt b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_m_inl0_hrd0.txt
index f4ab44e796..26cb412a69 100644
--- a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_m_inl0_hrd0.txt
+++ b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_m_inl0_hrd0.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_m_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
 
-DATE: 2023-11-24_14:32:45
+DATE: 2024-01-30_04:56:57
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_m_inl0_hrd0/gcheck.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 4.583584e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.155538e+08                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.269480e+08                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.567190e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.153367e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.271156e+08                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     0.521409 sec
-     2,266,982,107      cycles                           #    3.000 GHz                    
-     3,245,259,859      instructions                     #    1.43  insn per cycle         
-       0.825828714 seconds time elapsed
+TOTAL       :     0.531785 sec
+     2,166,596,506      cycles                           #    2.818 GHz                    
+     3,096,992,570      instructions                     #    1.43  insn per cycle         
+       0.839064322 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_m_inl0_hrd0/gcheck.exe -p 2048 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 214
@@ -72,20 +72,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_m_inl0_hrd0/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.183966e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.247909e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.247909e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.035137e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.096372e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.096372e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     4.906531 sec
-    15,256,207,680      cycles                           #    3.106 GHz                    
-    38,576,110,907      instructions                     #    2.53  insn per cycle         
-       4.915147304 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  672) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     5.263186 sec
+    15,248,441,904      cycles                           #    2.894 GHz                    
+    39,293,765,746      instructions                     #    2.58  insn per cycle         
+       5.273287972 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  740) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_m_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -99,20 +99,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_m_inl0_hrd0/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.714814e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.918909e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.918909e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.565129e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.766484e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.766484e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     2.927400 sec
-     8,963,531,758      cycles                           #    3.056 GHz                    
-    24,224,066,775      instructions                     #    2.70  insn per cycle         
-       2.945563789 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 2188) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     3.050997 sec
+     8,847,131,595      cycles                           #    2.894 GHz                    
+    24,093,216,326      instructions                     #    2.72  insn per cycle         
+       3.069927720 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 2102) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_m_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -126,20 +126,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_m_inl0_hrd0/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 6.052463e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 6.591282e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 6.591282e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.446912e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 5.914435e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.914435e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     1.836853 sec
-     5,386,059,012      cycles                           #    2.923 GHz                    
-    11,276,681,340      instructions                     #    2.09  insn per cycle         
-       1.853860147 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2480) (512y:    0) (512z:    0)
+TOTAL       :     2.035886 sec
+     5,501,574,982      cycles                           #    2.694 GHz                    
+    11,449,152,902      instructions                     #    2.08  insn per cycle         
+       2.052044507 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2467) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_m_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -147,26 +147,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_m_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_m_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028807e+00
-Avg ME (F77/C++)    = 2.0288064057068964
-Relative difference = 2.9292737240031234e-07
+Avg ME (F77/C++)    = 2.0288063930599014
+Relative difference = 2.9916108265801754e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_m_inl0_hrd0/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 6.743199e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.418159e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.418159e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.398707e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.055840e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.055840e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     1.658034 sec
-     4,871,841,569      cycles                           #    2.928 GHz                    
-    10,526,597,351      instructions                     #    2.16  insn per cycle         
-       1.674830124 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2167) (512y:  148) (512z:    0)
+TOTAL       :     1.750074 sec
+     4,773,598,492      cycles                           #    2.718 GHz                    
+    10,317,257,525      instructions                     #    2.16  insn per cycle         
+       1.763056572 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2076) (512y:  133) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_m_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -174,26 +174,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_m_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_m_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028807e+00
-Avg ME (F77/C++)    = 2.0288064057068964
-Relative difference = 2.9292737240031234e-07
+Avg ME (F77/C++)    = 2.0288063930599014
+Relative difference = 2.9916108265801754e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_m_inl0_hrd0/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 4.142731e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.384818e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.384818e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.115786e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.377584e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.377584e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     2.635826 sec
-     5,341,588,845      cycles                           #    2.022 GHz                    
-     7,603,619,006      instructions                     #    1.42  insn per cycle         
-       2.650860981 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1633) (512y:  126) (512z: 1608)
+TOTAL       :     2.659793 sec
+     4,851,599,101      cycles                           #    1.820 GHz                    
+     7,367,812,046      instructions                     #    1.52  insn per cycle         
+       2.678537528 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1366) (512y:   69) (512z: 1611)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_m_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -201,8 +201,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_m_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_m_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028807e+00
-Avg ME (F77/C++)    = 2.0288064057068964
-Relative difference = 2.9292737240031234e-07
+Avg ME (F77/C++)    = 2.0288063930599014
+Relative difference = 2.9916108265801754e-07
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_m_inl0_hrd1.txt b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_m_inl0_hrd1.txt
index 261980075b..3aadf8f9be 100644
--- a/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_m_inl0_hrd1.txt
+++ b/epochX/cudacpp/tput/logs_ggtt_mad/log_ggtt_mad_m_inl0_hrd1.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_m_inl0_hrd1'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx'
 
-DATE: 2023-11-24_14:33:12
+DATE: 2024-01-30_04:57:25
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_m_inl0_hrd1/gcheck.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 4.569187e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.154655e+08                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.270741e+08                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.571537e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.158030e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.273800e+08                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     0.519942 sec
-     2,236,595,831      cycles                           #    2.972 GHz                    
-     3,152,119,637      instructions                     #    1.41  insn per cycle         
-       0.821038019 seconds time elapsed
+TOTAL       :     0.527677 sec
+     2,187,527,722      cycles                           #    2.838 GHz                    
+     3,113,906,107      instructions                     #    1.42  insn per cycle         
+       0.843196902 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_m_inl0_hrd1/gcheck.exe -p 2048 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 208
@@ -72,20 +72,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_m_inl0_hrd1/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.146577e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.209722e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.209722e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.053597e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.114429e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.114429e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     4.990433 sec
-    15,248,789,401      cycles                           #    3.053 GHz                    
-    40,369,458,696      instructions                     #    2.65  insn per cycle         
-       4.998323590 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  669) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     5.216554 sec
+    15,076,935,035      cycles                           #    2.887 GHz                    
+    40,115,062,840      instructions                     #    2.66  insn per cycle         
+       5.225437216 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  630) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.none_m_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -99,20 +99,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_m_inl0_hrd1/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.998701e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.233932e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.233932e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.498695e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.695294e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.695294e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     2.726132 sec
-     8,491,093,801      cycles                           #    3.108 GHz                    
-    23,253,788,222      instructions                     #    2.74  insn per cycle         
-       2.739959213 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 2091) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     3.107625 sec
+     8,698,982,275      cycles                           #    2.794 GHz                    
+    23,534,504,437      instructions                     #    2.71  insn per cycle         
+       3.124975720 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 1993) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.sse4_m_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -126,20 +126,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_m_inl0_hrd1/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 5.127833e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 5.510579e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 5.510579e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.826638e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 5.191418e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.191418e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     2.148617 sec
-     6,247,078,338      cycles                           #    2.900 GHz                    
-    12,963,549,380      instructions                     #    2.08  insn per cycle         
-       2.164643261 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2669) (512y:    0) (512z:    0)
+TOTAL       :     2.282934 sec
+     6,198,059,216      cycles                           #    2.708 GHz                    
+    13,103,377,766      instructions                     #    2.11  insn per cycle         
+       2.300648997 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2711) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_m_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -147,26 +147,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_m_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.avx2_m_inl0_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028807e+00
-Avg ME (F77/C++)    = 2.0288064057068964
-Relative difference = 2.9292737240031234e-07
+Avg ME (F77/C++)    = 2.0288063930599014
+Relative difference = 2.9916108265801754e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_m_inl0_hrd1/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 5.424850e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 5.870009e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 5.870009e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.224417e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 5.653642e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.653642e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     2.038674 sec
-     5,929,446,799      cycles                           #    2.902 GHz                    
-    12,241,198,193      instructions                     #    2.06  insn per cycle         
-       2.052944380 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2209) (512y:  296) (512z:    0)
+TOTAL       :     2.117622 sec
+     5,754,647,700      cycles                           #    2.709 GHz                    
+    12,210,180,073      instructions                     #    2.12  insn per cycle         
+       2.133681313 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2201) (512y:  282) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_m_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -174,26 +174,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_m_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512y_m_inl0_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028807e+00
-Avg ME (F77/C++)    = 2.0288064057068964
-Relative difference = 2.9292737240031234e-07
+Avg ME (F77/C++)    = 2.0288063930599014
+Relative difference = 2.9916108265801754e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_m_inl0_hrd1/check.exe -p 2048 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.928483e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.143845e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.143845e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.752218e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.971190e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.971190e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.086689e+00 +- 3.413217e-03 )  GeV^0
-TOTAL       :     2.772333 sec
-     5,602,682,561      cycles                           #    2.017 GHz                    
-     8,743,458,912      instructions                     #    1.56  insn per cycle         
-       2.786330437 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1490) (512y:  183) (512z: 1909)
+TOTAL       :     2.905718 sec
+     5,261,261,771      cycles                           #    1.807 GHz                    
+     8,449,535,603      instructions                     #    1.61  insn per cycle         
+       2.918034623 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1324) (512y:   84) (512z: 1919)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_m_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -201,8 +201,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesse
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_m_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/build.512z_m_inl0_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 2.028807e+00
-Avg ME (F77/C++)    = 2.0288064057068964
-Relative difference = 2.9292737240031234e-07
+Avg ME (F77/C++)    = 2.0288063930599014
+Relative difference = 2.9916108265801754e-07
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_ggttg_mad/log_ggttg_mad_d_inl0_hrd0.txt b/epochX/cudacpp/tput/logs_ggttg_mad/log_ggttg_mad_d_inl0_hrd0.txt
index 157dda07e9..93e04f110e 100644
--- a/epochX/cudacpp/tput/logs_ggttg_mad/log_ggttg_mad_d_inl0_hrd0.txt
+++ b/epochX/cudacpp/tput/logs_ggttg_mad/log_ggttg_mad_d_inl0_hrd0.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg'
 
-DATE: 2023-11-24_14:33:40
+DATE: 2024-01-30_04:57:55
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_d_inl0_hrd0/gcheck.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 8.750385e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.049934e+07                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.064279e+07                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 8.751466e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.044991e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.059567e+07                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008920e+02 +- 5.001681e+01 )  GeV^-2
-TOTAL       :     0.462904 sec
-     2,042,714,082      cycles                           #    3.011 GHz                    
-     2,902,296,650      instructions                     #    1.42  insn per cycle         
-       0.758295254 seconds time elapsed
+TOTAL       :     0.471853 sec
+     1,938,631,197      cycles                           #    2.818 GHz                    
+     2,775,429,754      instructions                     #    1.43  insn per cycle         
+       0.768838740 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_d_inl0_hrd0/gcheck.exe -p 64 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 255
@@ -65,17 +65,17 @@ WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 .........................................................................
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_d_inl0_hrd0/gcheck.exe -p 2048 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 1.085304e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.323684e+07                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.337692e+07                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.083310e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.323559e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.337755e+07                 )  sec^-1
 MeanMatrixElemValue         = ( 6.734461e+02 +- 4.775415e+02 )  GeV^-2
-TOTAL       :     0.603567 sec
-     2,537,612,301      cycles                           #    3.015 GHz                    
-     3,865,051,021      instructions                     #    1.52  insn per cycle         
-       0.900037358 seconds time elapsed
+TOTAL       :     0.612248 sec
+     2,402,912,694      cycles                           #    2.815 GHz                    
+     3,669,599,520      instructions                     #    1.53  insn per cycle         
+       0.914185147 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_d_inl0_hrd0/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_d_inl0_hrd0/fgcheck.exe 2 64 2
@@ -86,20 +86,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_d_inl0_hrd0/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.562300e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.574662e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.574662e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.436781e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.449292e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.449292e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008920e+02 +- 5.001681e+01 )  GeV^-2
-TOTAL       :     6.416503 sec
-    19,738,905,705      cycles                           #    3.075 GHz                    
-    59,604,379,345      instructions                     #    3.02  insn per cycle         
-       6.423169029 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 1466) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     6.748293 sec
+    19,527,368,133      cycles                           #    2.892 GHz                    
+    57,921,410,950      instructions                     #    2.97  insn per cycle         
+       6.756473501 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 1134) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -107,26 +107,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.413122e+00
-Avg ME (F77/C++)    = 1.4131213684432433
-Relative difference = 4.46923023397472e-07
+Avg ME (F77/C++)    = 1.4131213684432431
+Relative difference = 4.4692302355460254e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.sse4_d_inl0_hrd0/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 4.963631e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 5.008800e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 5.008800e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.689715e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.736371e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.736371e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008920e+02 +- 5.001681e+01 )  GeV^-2
-TOTAL       :     3.322549 sec
-    10,350,945,705      cycles                           #    3.112 GHz                    
-    30,674,390,605      instructions                     #    2.96  insn per cycle         
-       3.334406966 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 5153) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     3.517761 sec
+    10,204,769,485      cycles                           #    2.897 GHz                    
+    29,944,325,485      instructions                     #    2.93  insn per cycle         
+       3.533017528 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 4742) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.sse4_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -140,20 +140,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.avx2_d_inl0_hrd0/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 9.645905e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 9.824459e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 9.824459e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 9.110539e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 9.290286e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.290286e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008920e+02 +- 5.001681e+01 )  GeV^-2
-TOTAL       :     1.720961 sec
-     4,880,621,681      cycles                           #    2.829 GHz                    
-    11,019,918,413      instructions                     #    2.26  insn per cycle         
-       1.732752875 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 4467) (512y:    0) (512z:    0)
+TOTAL       :     1.822880 sec
+     4,929,256,319      cycles                           #    2.697 GHz                    
+    11,212,094,634      instructions                     #    2.27  insn per cycle         
+       1.842452367 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 4396) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.avx2_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -167,20 +167,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512y_d_inl0_hrd0/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.100615e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.123065e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.123065e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.045459e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.068242e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.068242e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008920e+02 +- 5.001681e+01 )  GeV^-2
-TOTAL       :     1.510594 sec
-     4,368,423,747      cycles                           #    2.884 GHz                    
-    10,296,629,918      instructions                     #    2.36  insn per cycle         
-       1.523579096 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 4137) (512y:   91) (512z:    0)
+TOTAL       :     1.591153 sec
+     4,310,771,194      cycles                           #    2.701 GHz                    
+    10,188,135,001      instructions                     #    2.36  insn per cycle         
+       1.604477930 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3895) (512y:   81) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512y_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -194,20 +194,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512z_d_inl0_hrd0/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.837255e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.949435e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.949435e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 7.350984e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.465337e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.465337e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008920e+02 +- 5.001681e+01 )  GeV^-2
-TOTAL       :     2.114472 sec
-     4,096,385,157      cycles                           #    1.934 GHz                    
-     5,842,611,877      instructions                     #    1.43  insn per cycle         
-       2.127694185 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1540) (512y:   95) (512z: 3466)
+TOTAL       :     2.255127 sec
+     3,913,955,092      cycles                           #    1.732 GHz                    
+     5,709,470,043      instructions                     #    1.46  insn per cycle         
+       2.269083887 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1258) (512y:   74) (512z: 3396)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512z_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
diff --git a/epochX/cudacpp/tput/logs_ggttg_mad/log_ggttg_mad_d_inl0_hrd0_bridge.txt b/epochX/cudacpp/tput/logs_ggttg_mad/log_ggttg_mad_d_inl0_hrd0_bridge.txt
index 62c84d1195..ec4707eb36 100644
--- a/epochX/cudacpp/tput/logs_ggttg_mad/log_ggttg_mad_d_inl0_hrd0_bridge.txt
+++ b/epochX/cudacpp/tput/logs_ggttg_mad/log_ggttg_mad_d_inl0_hrd0_bridge.txt
@@ -41,7 +41,7 @@ CUDACPP_BUILDDIR='build.512z_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg'
 
-DATE: 2023-11-24_15:11:49
+DATE: 2024-01-30_05:47:33
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
@@ -51,17 +51,17 @@ WARNING! Bridge selected: cannot use RamboDevice, will use RamboHost
 WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
 WARNING! Instantiate device Bridge (nevt=16384, gpublocks=64, gputhreads=256, gpublocks*gputhreads=16384)
 WARNING! Set grid in Bridge (nevt=16384, gpublocks=64, gputhreads=256, gpublocks*gputhreads=16384)
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 2.639935e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 6.816863e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 6.816863e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.528893e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.736864e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.736864e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008920e+02 +- 5.001681e+01 )  GeV^-2
-TOTAL       :     0.492213 sec
-     2,054,207,791      cycles                           #    2.926 GHz                    
-     3,126,128,113      instructions                     #    1.52  insn per cycle         
-       0.760942013 seconds time elapsed
+TOTAL       :     0.499721 sec
+     2,019,319,467      cycles                           #    2.834 GHz                    
+     3,049,308,251      instructions                     #    1.51  insn per cycle         
+       0.770515897 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_d_inl0_hrd0/gcheck.exe -p 64 256 1 --bridge
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Bridge selected: cannot use RamboDevice, will use RamboHost
@@ -77,17 +77,17 @@ WARNING! Bridge selected: cannot use RamboDevice, will use RamboHost
 WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
 WARNING! Instantiate device Bridge (nevt=524288, gpublocks=2048, gputhreads=256, gpublocks*gputhreads=524288)
 WARNING! Set grid in Bridge (nevt=524288, gpublocks=2048, gputhreads=256, gpublocks*gputhreads=524288)
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 2.653176e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 9.472944e+06                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 9.472944e+06                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.631733e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 9.469522e+06                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.469522e+06                 )  sec^-1
 MeanMatrixElemValue         = ( 6.734461e+02 +- 4.775415e+02 )  GeV^-2
-TOTAL       :     0.830881 sec
-     3,123,864,419      cycles                           #    2.883 GHz                    
-     4,825,503,220      instructions                     #    1.54  insn per cycle         
-       1.142577785 seconds time elapsed
+TOTAL       :     0.838079 sec
+     3,105,645,423      cycles                           #    2.841 GHz                    
+     4,885,001,867      instructions                     #    1.57  insn per cycle         
+       1.151319170 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_d_inl0_hrd0/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_d_inl0_hrd0/fgcheck.exe 2 64 2
@@ -99,20 +99,20 @@ OK (relative difference <= 5E-3)
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_d_inl0_hrd0/check.exe -p 64 256 10 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=16384)
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.499287e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.511764e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.511764e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.430928e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.443345e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.443345e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008920e+02 +- 5.001681e+01 )  GeV^-2
-TOTAL       :     6.585609 sec
-    19,745,466,264      cycles                           #    2.998 GHz                    
-    59,612,197,800      instructions                     #    3.02  insn per cycle         
-       6.590135647 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 1466) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     6.772792 sec
+    19,550,332,735      cycles                           #    2.885 GHz                    
+    57,928,238,854      instructions                     #    2.96  insn per cycle         
+       6.778111068 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 1134) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -120,27 +120,27 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.413122e+00
-Avg ME (F77/C++)    = 1.4131213684432433
-Relative difference = 4.46923023397472e-07
+Avg ME (F77/C++)    = 1.4131213684432431
+Relative difference = 4.4692302355460254e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.sse4_d_inl0_hrd0/check.exe -p 64 256 10 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=16384)
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/sse4+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 4.693234e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.737492e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.737492e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.642090e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.688492e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.688492e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008920e+02 +- 5.001681e+01 )  GeV^-2
-TOTAL       :     3.521832 sec
-    10,407,813,275      cycles                           #    2.952 GHz                    
-    30,725,415,932      instructions                     #    2.95  insn per cycle         
-       3.526742122 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 5153) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     3.562259 sec
+    10,259,962,003      cycles                           #    2.883 GHz                    
+    29,997,071,393      instructions                     #    2.92  insn per cycle         
+       3.567805037 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 4742) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.sse4_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -155,20 +155,20 @@ OK (relative difference <= 5E-3)
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.avx2_d_inl0_hrd0/check.exe -p 64 256 10 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=16384)
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/avx2+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 9.553194e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 9.735067e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 9.735067e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 9.060333e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 9.240360e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.240360e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008920e+02 +- 5.001681e+01 )  GeV^-2
-TOTAL       :     1.745455 sec
-     4,924,157,638      cycles                           #    2.815 GHz                    
-    11,068,084,114      instructions                     #    2.25  insn per cycle         
-       1.749966089 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 4467) (512y:    0) (512z:    0)
+TOTAL       :     1.842606 sec
+     4,975,429,359      cycles                           #    2.695 GHz                    
+    11,262,132,806      instructions                     #    2.26  insn per cycle         
+       1.848498494 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 4396) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.avx2_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -183,20 +183,20 @@ OK (relative difference <= 5E-3)
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512y_d_inl0_hrd0/check.exe -p 64 256 10 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=16384)
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/512y+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.070624e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.092463e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.092463e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.041344e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.064837e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.064837e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008920e+02 +- 5.001681e+01 )  GeV^-2
-TOTAL       :     1.559525 sec
-     4,401,889,175      cycles                           #    2.815 GHz                    
-    10,345,025,900      instructions                     #    2.35  insn per cycle         
-       1.564094276 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 4137) (512y:   91) (512z:    0)
+TOTAL       :     1.605579 sec
+     4,356,497,896      cycles                           #    2.706 GHz                    
+    10,236,092,665      instructions                     #    2.35  insn per cycle         
+       1.611218031 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3895) (512y:   81) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512y_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -211,20 +211,20 @@ OK (relative difference <= 5E-3)
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512z_d_inl0_hrd0/check.exe -p 64 256 10 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=16384)
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/512z+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.394172e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.507838e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.507838e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 7.341333e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.457820e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.457820e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008920e+02 +- 5.001681e+01 )  GeV^-2
-TOTAL       :     2.248960 sec
-     4,150,304,517      cycles                           #    1.842 GHz                    
-     5,879,795,808      instructions                     #    1.42  insn per cycle         
-       2.253443263 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1540) (512y:   95) (512z: 3466)
+TOTAL       :     2.270029 sec
+     3,960,771,261      cycles                           #    1.743 GHz                    
+     5,748,864,563      instructions                     #    1.45  insn per cycle         
+       2.275659808 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1258) (512y:   74) (512z: 3396)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512z_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
diff --git a/epochX/cudacpp/tput/logs_ggttg_mad/log_ggttg_mad_d_inl0_hrd1.txt b/epochX/cudacpp/tput/logs_ggttg_mad/log_ggttg_mad_d_inl0_hrd1.txt
index 739a090c03..e0fcb209a0 100644
--- a/epochX/cudacpp/tput/logs_ggttg_mad/log_ggttg_mad_d_inl0_hrd1.txt
+++ b/epochX/cudacpp/tput/logs_ggttg_mad/log_ggttg_mad_d_inl0_hrd1.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_d_inl0_hrd1'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg'
 
-DATE: 2023-11-24_14:34:09
+DATE: 2024-01-30_04:58:25
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_d_inl0_hrd1/gcheck.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 8.713263e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.038994e+07                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.053292e+07                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 8.715814e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.042075e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.056833e+07                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008920e+02 +- 5.001681e+01 )  GeV^-2
-TOTAL       :     0.459803 sec
-     2,029,299,480      cycles                           #    3.004 GHz                    
-     2,918,522,159      instructions                     #    1.44  insn per cycle         
-       0.740886124 seconds time elapsed
+TOTAL       :     0.470883 sec
+     1,939,912,503      cycles                           #    2.822 GHz                    
+     2,790,884,564      instructions                     #    1.44  insn per cycle         
+       0.765236939 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_d_inl0_hrd1/gcheck.exe -p 64 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 255
@@ -65,17 +65,17 @@ WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 .........................................................................
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_d_inl0_hrd1/gcheck.exe -p 2048 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 1.076194e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.311479e+07                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.325166e+07                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.074401e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.309128e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.323134e+07                 )  sec^-1
 MeanMatrixElemValue         = ( 6.734461e+02 +- 4.775415e+02 )  GeV^-2
-TOTAL       :     0.596867 sec
-     2,506,078,107      cycles                           #    3.006 GHz                    
-     3,672,565,526      instructions                     #    1.47  insn per cycle         
-       0.892288871 seconds time elapsed
+TOTAL       :     0.606508 sec
+     2,399,848,951      cycles                           #    2.837 GHz                    
+     3,558,977,452      instructions                     #    1.48  insn per cycle         
+       0.907497861 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_d_inl0_hrd1/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_d_inl0_hrd1/fgcheck.exe 2 64 2
@@ -86,20 +86,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_d_inl0_hrd1/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.614658e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.627547e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.627547e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.442527e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.455052e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.455052e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008920e+02 +- 5.001681e+01 )  GeV^-2
-TOTAL       :     6.288241 sec
-    19,506,895,795      cycles                           #    3.101 GHz                    
-    58,795,912,300      instructions                     #    3.01  insn per cycle         
-       6.294901652 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 1313) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     6.732470 sec
+    19,518,863,765      cycles                           #    2.898 GHz                    
+    57,747,544,085      instructions                     #    2.96  insn per cycle         
+       6.739693684 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 1087) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_d_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -107,26 +107,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_d_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_d_inl0_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.413122e+00
-Avg ME (F77/C++)    = 1.4131213684432433
-Relative difference = 4.46923023397472e-07
+Avg ME (F77/C++)    = 1.4131213684432431
+Relative difference = 4.4692302355460254e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.sse4_d_inl0_hrd1/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 4.998673e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 5.044453e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 5.044453e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.661123e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.707073e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.707073e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008920e+02 +- 5.001681e+01 )  GeV^-2
-TOTAL       :     3.299411 sec
-    10,248,689,037      cycles                           #    3.103 GHz                    
-    30,346,682,819      instructions                     #    2.96  insn per cycle         
-       3.310933910 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 4970) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     3.538386 sec
+    10,268,038,737      cycles                           #    2.898 GHz                    
+    30,334,584,369      instructions                     #    2.95  insn per cycle         
+       3.554140482 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 4806) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.sse4_d_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -140,20 +140,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.avx2_d_inl0_hrd1/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 9.591359e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 9.761947e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 9.761947e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 8.842618e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 9.012045e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.012045e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008920e+02 +- 5.001681e+01 )  GeV^-2
-TOTAL       :     1.730287 sec
-     5,042,983,525      cycles                           #    2.908 GHz                    
-    11,483,783,187      instructions                     #    2.28  insn per cycle         
-       1.743862515 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 4591) (512y:    0) (512z:    0)
+TOTAL       :     1.876874 sec
+     5,068,616,518      cycles                           #    2.693 GHz                    
+    11,664,707,542      instructions                     #    2.30  insn per cycle         
+       1.896780245 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 4489) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.avx2_d_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -167,20 +167,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512y_d_inl0_hrd1/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.041838e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.061977e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.061977e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 9.766097e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 9.969139e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.969139e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008920e+02 +- 5.001681e+01 )  GeV^-2
-TOTAL       :     1.594766 sec
-     4,637,285,824      cycles                           #    2.900 GHz                    
-    10,841,968,745      instructions                     #    2.34  insn per cycle         
-       1.604083789 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 4183) (512y:  244) (512z:    0)
+TOTAL       :     1.701579 sec
+     4,623,474,911      cycles                           #    2.710 GHz                    
+    10,806,178,257      instructions                     #    2.34  insn per cycle         
+       1.712732749 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3988) (512y:  237) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512y_d_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -194,20 +194,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512z_d_inl0_hrd1/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.531164e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.641315e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.641315e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 7.261988e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.377447e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.377447e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008920e+02 +- 5.001681e+01 )  GeV^-2
-TOTAL       :     2.200047 sec
-     4,116,597,786      cycles                           #    1.868 GHz                    
-     6,106,799,640      instructions                     #    1.48  insn per cycle         
-       2.213612018 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1457) (512y:  139) (512z: 3568)
+TOTAL       :     2.282726 sec
+     3,962,643,032      cycles                           #    1.733 GHz                    
+     5,999,265,657      instructions                     #    1.51  insn per cycle         
+       2.297742409 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1241) (512y:   81) (512z: 3500)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512z_d_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
diff --git a/epochX/cudacpp/tput/logs_ggttg_mad/log_ggttg_mad_f_inl0_hrd0.txt b/epochX/cudacpp/tput/logs_ggttg_mad/log_ggttg_mad_f_inl0_hrd0.txt
index ee258b9eb8..809c0d4a45 100644
--- a/epochX/cudacpp/tput/logs_ggttg_mad/log_ggttg_mad_f_inl0_hrd0.txt
+++ b/epochX/cudacpp/tput/logs_ggttg_mad/log_ggttg_mad_f_inl0_hrd0.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_f_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg'
 
-DATE: 2023-11-24_14:34:38
+DATE: 2024-01-30_04:58:56
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_f_inl0_hrd0/gcheck.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 2.486813e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.354265e+07                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.453893e+07                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.450759e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.307242e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.403943e+07                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008472e+02 +- 5.002447e+01 )  GeV^-2
-TOTAL       :     0.444260 sec
-     1,976,391,201      cycles                           #    2.992 GHz                    
-     2,802,325,698      instructions                     #    1.42  insn per cycle         
-       0.735914466 seconds time elapsed
+TOTAL       :     0.453655 sec
+     1,885,130,441      cycles                           #    2.809 GHz                    
+     2,653,723,410      instructions                     #    1.41  insn per cycle         
+       0.747134110 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_f_inl0_hrd0/gcheck.exe -p 64 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 254
@@ -65,17 +65,17 @@ WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 .........................................................................
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_f_inl0_hrd0/gcheck.exe -p 2048 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 3.217656e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.394394e+07                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.476515e+07                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.211065e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.390139e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.474767e+07                 )  sec^-1
 MeanMatrixElemValue         = ( 6.630099e+02 +- 4.770719e+02 )  GeV^-2
-TOTAL       :     0.493371 sec
-     2,193,458,897      cycles                           #    3.016 GHz                    
-     3,142,646,768      instructions                     #    1.43  insn per cycle         
-       0.786643658 seconds time elapsed
+TOTAL       :     0.497953 sec
+     2,053,184,300      cycles                           #    2.823 GHz                    
+     2,862,941,904      instructions                     #    1.39  insn per cycle         
+       0.785494017 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_f_inl0_hrd0/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_f_inl0_hrd0/fgcheck.exe 2 64 2
@@ -86,74 +86,74 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_f_inl0_hrd0/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.645635e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.659236e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.659236e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.619709e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.634289e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.634289e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 1.009236e+02 +- 5.002643e+01 )  GeV^-2
-TOTAL       :     6.214195 sec
-    19,058,711,796      cycles                           #    3.065 GHz                    
-    58,958,354,991      instructions                     #    3.09  insn per cycle         
-       6.220591504 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 1034) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     6.276631 sec
+    18,176,411,104      cycles                           #    2.894 GHz                    
+    55,238,700,170      instructions                     #    3.04  insn per cycle         
+       6.284146623 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 1229) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_f_inl0_hrd0/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 1.412986e+00
-Avg ME (F77/C++)    = 1.4129858051842916
-Relative difference = 1.3787518662898538e-07
+Avg ME (C++/C++)    = 1.412998e+00
+Avg ME (F77/C++)    = 1.4129977771372637
+Relative difference = 1.5772332039074602e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.sse4_f_inl0_hrd0/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 8.795314e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 8.946425e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 8.946425e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 8.447433e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 8.602543e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.602543e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 1.009236e+02 +- 5.002643e+01 )  GeV^-2
-TOTAL       :     1.882613 sec
-     5,847,309,673      cycles                           #    3.099 GHz                    
-    16,694,109,682      instructions                     #    2.86  insn per cycle         
-       1.898696341 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 5765) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     1.961648 sec
+     5,691,843,956      cycles                           #    2.895 GHz                    
+    16,128,541,176      instructions                     #    2.83  insn per cycle         
+       1.980848485 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 5205) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.sse4_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.sse4_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.sse4_f_inl0_hrd0/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 1.412987e+00
-Avg ME (F77/C++)    = 1.4129865669244737
-Relative difference = 3.06496469061158e-07
+Avg ME (C++/C++)    = 1.412986e+00
+Avg ME (F77/C++)    = 1.4129864902818952
+Relative difference = 3.469828399449743e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.avx2_f_inl0_hrd0/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.901790e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.970007e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.970007e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.757867e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.823085e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.823085e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008855e+02 +- 5.002467e+01 )  GeV^-2
-TOTAL       :     0.881272 sec
-     2,581,984,725      cycles                           #    2.916 GHz                    
-     5,980,308,545      instructions                     #    2.32  insn per cycle         
-       0.895784511 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 4917) (512y:    0) (512z:    0)
+TOTAL       :     0.954501 sec
+     2,591,810,421      cycles                           #    2.702 GHz                    
+     6,085,915,267      instructions                     #    2.35  insn per cycle         
+       0.966912682 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 4878) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.avx2_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -161,26 +161,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.avx2_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.avx2_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.413316e+00
-Avg ME (F77/C++)    = 1.4133161655815059
-Relative difference = 1.1715816267550621e-07
+Avg ME (F77/C++)    = 1.4133158486847037
+Relative difference = 1.0706402269051248e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512y_f_inl0_hrd0/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.105287e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.189516e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.189516e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.986474e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.069956e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.069956e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008855e+02 +- 5.002467e+01 )  GeV^-2
-TOTAL       :     0.797874 sec
-     2,345,612,557      cycles                           #    2.925 GHz                    
-     5,602,963,634      instructions                     #    2.39  insn per cycle         
-       0.810229657 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 4645) (512y:   36) (512z:    0)
+TOTAL       :     0.846832 sec
+     2,295,114,840      cycles                           #    2.696 GHz                    
+     5,552,751,365      instructions                     #    2.42  insn per cycle         
+       0.861502194 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 4415) (512y:   30) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512y_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -188,26 +188,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512y_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512y_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.413316e+00
-Avg ME (F77/C++)    = 1.4133161655815059
-Relative difference = 1.1715816267550621e-07
+Avg ME (F77/C++)    = 1.4133158486847037
+Relative difference = 1.0706402269051248e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512z_f_inl0_hrd0/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[16] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.549899e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.597374e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.597374e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.460942e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.506292e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.506292e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008856e+02 +- 5.002468e+01 )  GeV^-2
-TOTAL       :     1.080161 sec
-     2,054,065,161      cycles                           #    1.894 GHz                    
-     3,334,275,922      instructions                     #    1.62  insn per cycle         
-       1.095623221 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2131) (512y:   39) (512z: 3668)
+TOTAL       :     1.145570 sec
+     2,022,184,795      cycles                           #    1.758 GHz                    
+     3,286,748,929      instructions                     #    1.63  insn per cycle         
+       1.163999883 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1905) (512y:   28) (512z: 3597)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512z_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -215,8 +215,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512z_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512z_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.413316e+00
-Avg ME (F77/C++)    = 1.4133164033579249
-Relative difference = 2.85398258307829e-07
+Avg ME (F77/C++)    = 1.4133164031689205
+Relative difference = 2.852645271622733e-07
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_ggttg_mad/log_ggttg_mad_f_inl0_hrd0_bridge.txt b/epochX/cudacpp/tput/logs_ggttg_mad/log_ggttg_mad_f_inl0_hrd0_bridge.txt
index d8dc2c3678..8f1e29c773 100644
--- a/epochX/cudacpp/tput/logs_ggttg_mad/log_ggttg_mad_f_inl0_hrd0_bridge.txt
+++ b/epochX/cudacpp/tput/logs_ggttg_mad/log_ggttg_mad_f_inl0_hrd0_bridge.txt
@@ -41,7 +41,7 @@ CUDACPP_BUILDDIR='build.512z_f_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg'
 
-DATE: 2023-11-24_15:12:19
+DATE: 2024-01-30_05:48:04
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
@@ -51,17 +51,17 @@ WARNING! Bridge selected: cannot use RamboDevice, will use RamboHost
 WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
 WARNING! Instantiate device Bridge (nevt=16384, gpublocks=64, gputhreads=256, gpublocks*gputhreads=16384)
 WARNING! Set grid in Bridge (nevt=16384, gpublocks=64, gputhreads=256, gpublocks*gputhreads=16384)
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 5.001518e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.137858e+07                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.137858e+07                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.794241e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.099961e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.099961e+07                 )  sec^-1
 MeanMatrixElemValue         = ( 1.009071e+02 +- 5.002295e+01 )  GeV^-2
-TOTAL       :     0.455815 sec
-     1,948,301,357      cycles                           #    2.927 GHz                    
-     2,907,177,849      instructions                     #    1.49  insn per cycle         
-       0.723196787 seconds time elapsed
+TOTAL       :     0.464854 sec
+     1,913,128,801      cycles                           #    2.831 GHz                    
+     2,814,269,280      instructions                     #    1.47  insn per cycle         
+       0.735191571 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_f_inl0_hrd0/gcheck.exe -p 64 256 1 --bridge
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Bridge selected: cannot use RamboDevice, will use RamboHost
@@ -77,17 +77,17 @@ WARNING! Bridge selected: cannot use RamboDevice, will use RamboHost
 WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
 WARNING! Instantiate device Bridge (nevt=524288, gpublocks=2048, gputhreads=256, gpublocks*gputhreads=524288)
 WARNING! Set grid in Bridge (nevt=524288, gpublocks=2048, gputhreads=256, gpublocks*gputhreads=524288)
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 4.698909e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.564625e+07                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.564625e+07                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.563056e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.567773e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.567773e+07                 )  sec^-1
 MeanMatrixElemValue         = ( 6.737500e+02 +- 4.776370e+02 )  GeV^-2
-TOTAL       :     0.639136 sec
-     2,545,540,064      cycles                           #    2.932 GHz                    
-     3,908,776,443      instructions                     #    1.54  insn per cycle         
-       0.927730276 seconds time elapsed
+TOTAL       :     0.649992 sec
+     2,514,728,119      cycles                           #    2.840 GHz                    
+     3,857,856,675      instructions                     #    1.53  insn per cycle         
+       0.945286461 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_f_inl0_hrd0/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_f_inl0_hrd0/fgcheck.exe 2 64 2
@@ -99,76 +99,76 @@ OK (relative difference <= 5E-3)
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_f_inl0_hrd0/check.exe -p 64 256 10 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=16384)
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.559629e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.572951e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.572951e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.612332e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.626818e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.626818e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 1.009236e+02 +- 5.002643e+01 )  GeV^-2
-TOTAL       :     6.425977 sec
-    19,091,912,214      cycles                           #    2.972 GHz                    
-    58,967,324,889      instructions                     #    3.09  insn per cycle         
-       6.430306421 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 1034) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     6.299552 sec
+    18,207,767,275      cycles                           #    2.889 GHz                    
+    55,242,943,760      instructions                     #    3.03  insn per cycle         
+       6.304483382 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 1229) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_f_inl0_hrd0/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 1.412986e+00
-Avg ME (F77/C++)    = 1.4129858051842916
-Relative difference = 1.3787518662898538e-07
+Avg ME (C++/C++)    = 1.412998e+00
+Avg ME (F77/C++)    = 1.4129977771372637
+Relative difference = 1.5772332039074602e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.sse4_f_inl0_hrd0/check.exe -p 64 256 10 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=16384)
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/sse4+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 8.438378e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 8.588398e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 8.588398e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 8.365917e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 8.522444e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.522444e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 1.009236e+02 +- 5.002643e+01 )  GeV^-2
-TOTAL       :     1.967455 sec
-     5,879,551,760      cycles                           #    2.983 GHz                    
-    16,741,960,213      instructions                     #    2.85  insn per cycle         
-       1.971863675 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 5765) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     1.986287 sec
+     5,717,011,577      cycles                           #    2.873 GHz                    
+    16,175,954,346      instructions                     #    2.83  insn per cycle         
+       1.991587162 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 5205) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.sse4_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.sse4_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.sse4_f_inl0_hrd0/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 1.412987e+00
-Avg ME (F77/C++)    = 1.4129865669244737
-Relative difference = 3.06496469061158e-07
+Avg ME (C++/C++)    = 1.412986e+00
+Avg ME (F77/C++)    = 1.4129864902818952
+Relative difference = 3.469828399449743e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.avx2_f_inl0_hrd0/check.exe -p 64 256 10 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=16384)
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/avx2+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.817320e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.883646e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.883646e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.741687e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.807547e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.807547e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008855e+02 +- 5.002467e+01 )  GeV^-2
-TOTAL       :     0.926249 sec
-     2,607,056,313      cycles                           #    2.804 GHz                    
-     6,016,570,943      instructions                     #    2.31  insn per cycle         
-       0.930566113 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 4917) (512y:    0) (512z:    0)
+TOTAL       :     0.968315 sec
+     2,618,792,433      cycles                           #    2.693 GHz                    
+     6,122,206,815      instructions                     #    2.34  insn per cycle         
+       0.973667021 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 4878) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.avx2_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -176,27 +176,27 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.avx2_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.avx2_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.413316e+00
-Avg ME (F77/C++)    = 1.4133161655815059
-Relative difference = 1.1715816267550621e-07
+Avg ME (F77/C++)    = 1.4133158486847037
+Relative difference = 1.0706402269051248e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512y_f_inl0_hrd0/check.exe -p 64 256 10 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=16384)
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/512y+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.015882e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.096022e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.096022e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.976057e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.060749e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.060749e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008855e+02 +- 5.002467e+01 )  GeV^-2
-TOTAL       :     0.837462 sec
-     2,364,091,843      cycles                           #    2.810 GHz                    
-     5,638,984,109      instructions                     #    2.39  insn per cycle         
-       0.841852545 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 4645) (512y:   36) (512z:    0)
+TOTAL       :     0.855993 sec
+     2,321,654,642      cycles                           #    2.699 GHz                    
+     5,589,002,861      instructions                     #    2.41  insn per cycle         
+       0.861171520 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 4415) (512y:   30) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512y_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -204,27 +204,27 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512y_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512y_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.413316e+00
-Avg ME (F77/C++)    = 1.4133161655815059
-Relative difference = 1.1715816267550621e-07
+Avg ME (F77/C++)    = 1.4133158486847037
+Relative difference = 1.0706402269051248e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512z_f_inl0_hrd0/check.exe -p 64 256 10 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=16384)
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/512z+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[16] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.519186e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.565700e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.565700e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.455132e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.500155e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.500155e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008856e+02 +- 5.002468e+01 )  GeV^-2
-TOTAL       :     1.105946 sec
-     2,075,898,441      cycles                           #    1.871 GHz                    
-     3,374,696,524      instructions                     #    1.63  insn per cycle         
-       1.110329864 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2131) (512y:   39) (512z: 3668)
+TOTAL       :     1.154878 sec
+     2,044,999,339      cycles                           #    1.765 GHz                    
+     3,327,504,110      instructions                     #    1.63  insn per cycle         
+       1.160035358 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1905) (512y:   28) (512z: 3597)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512z_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -232,8 +232,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512z_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512z_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.413316e+00
-Avg ME (F77/C++)    = 1.4133164033579249
-Relative difference = 2.85398258307829e-07
+Avg ME (F77/C++)    = 1.4133164031689205
+Relative difference = 2.852645271622733e-07
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_ggttg_mad/log_ggttg_mad_f_inl0_hrd1.txt b/epochX/cudacpp/tput/logs_ggttg_mad/log_ggttg_mad_f_inl0_hrd1.txt
index 69970d8c55..71f99cc0f9 100644
--- a/epochX/cudacpp/tput/logs_ggttg_mad/log_ggttg_mad_f_inl0_hrd1.txt
+++ b/epochX/cudacpp/tput/logs_ggttg_mad/log_ggttg_mad_f_inl0_hrd1.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_f_inl0_hrd1'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg'
 
-DATE: 2023-11-24_14:35:02
+DATE: 2024-01-30_04:59:21
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_f_inl0_hrd1/gcheck.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 2.373914e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.228922e+07                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.325879e+07                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.454028e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.326749e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.426065e+07                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008472e+02 +- 5.002447e+01 )  GeV^-2
-TOTAL       :     0.443366 sec
-     2,003,938,604      cycles                           #    2.996 GHz                    
-     2,792,184,039      instructions                     #    1.39  insn per cycle         
-       0.740353888 seconds time elapsed
+TOTAL       :     0.451891 sec
+     1,884,922,304      cycles                           #    2.826 GHz                    
+     2,675,153,942      instructions                     #    1.42  insn per cycle         
+       0.742708600 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_f_inl0_hrd1/gcheck.exe -p 64 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 248
@@ -65,17 +65,17 @@ WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 .........................................................................
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_f_inl0_hrd1/gcheck.exe -p 2048 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 3.237434e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.431964e+07                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.514879e+07                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.211971e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.383449e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.465566e+07                 )  sec^-1
 MeanMatrixElemValue         = ( 6.630099e+02 +- 4.770719e+02 )  GeV^-2
-TOTAL       :     0.491294 sec
-     2,181,970,196      cycles                           #    3.018 GHz                    
-     3,127,779,956      instructions                     #    1.43  insn per cycle         
-       0.781703991 seconds time elapsed
+TOTAL       :     0.498447 sec
+     2,066,378,383      cycles                           #    2.841 GHz                    
+     2,912,189,828      instructions                     #    1.41  insn per cycle         
+       0.785425719 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_f_inl0_hrd1/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_f_inl0_hrd1/fgcheck.exe 2 64 2
@@ -86,74 +86,74 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_f_inl0_hrd1/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.673738e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.687597e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.687597e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.621420e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.635929e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.635929e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 1.009236e+02 +- 5.002643e+01 )  GeV^-2
-TOTAL       :     6.148417 sec
-    18,978,547,749      cycles                           #    3.085 GHz                    
-    58,701,989,638      instructions                     #    3.09  insn per cycle         
-       6.154751887 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 1029) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     6.272536 sec
+    18,133,908,438      cycles                           #    2.889 GHz                    
+    54,991,536,969      instructions                     #    3.03  insn per cycle         
+       6.280002049 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 1171) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_f_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_f_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_f_inl0_hrd1/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 1.412986e+00
-Avg ME (F77/C++)    = 1.4129858051842916
-Relative difference = 1.3787518662898538e-07
+Avg ME (C++/C++)    = 1.412998e+00
+Avg ME (F77/C++)    = 1.4129977771372637
+Relative difference = 1.5772332039074602e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.sse4_f_inl0_hrd1/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 9.234838e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 9.399257e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 9.399257e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 8.675526e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 8.845155e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.845155e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 1.009236e+02 +- 5.002643e+01 )  GeV^-2
-TOTAL       :     1.793566 sec
-     5,583,260,034      cycles                           #    3.106 GHz                    
-    16,511,055,213      instructions                     #    2.96  insn per cycle         
-       1.805973554 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 5551) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     1.910252 sec
+     5,541,476,355      cycles                           #    2.894 GHz                    
+    16,222,950,904      instructions                     #    2.93  insn per cycle         
+       1.926546393 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 5136) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.sse4_f_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.sse4_f_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.sse4_f_inl0_hrd1/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 1.412987e+00
-Avg ME (F77/C++)    = 1.4129865669244737
-Relative difference = 3.06496469061158e-07
+Avg ME (C++/C++)    = 1.412986e+00
+Avg ME (F77/C++)    = 1.4129863487235070
+Relative difference = 2.4679898241023883e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.avx2_f_inl0_hrd1/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.640420e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.690986e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.690986e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.524928e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.573974e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.573974e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008855e+02 +- 5.002467e+01 )  GeV^-2
-TOTAL       :     1.018575 sec
-     2,975,461,813      cycles                           #    2.909 GHz                    
-     6,634,551,571      instructions                     #    2.23  insn per cycle         
-       1.032313323 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 5568) (512y:    0) (512z:    0)
+TOTAL       :     1.096795 sec
+     2,981,881,341      cycles                           #    2.708 GHz                    
+     6,708,240,605      instructions                     #    2.25  insn per cycle         
+       1.109848469 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 5430) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.avx2_f_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -161,26 +161,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.avx2_f_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.avx2_f_inl0_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.413316e+00
-Avg ME (F77/C++)    = 1.4133161655815059
-Relative difference = 1.1715816267550621e-07
+Avg ME (F77/C++)    = 1.4133158486847037
+Relative difference = 1.0706402269051248e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512y_f_inl0_hrd1/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.779474e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.839054e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.839054e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.679205e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.738776e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.738776e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008855e+02 +- 5.002467e+01 )  GeV^-2
-TOTAL       :     0.940459 sec
-     2,751,956,582      cycles                           #    2.914 GHz                    
-     6,255,845,013      instructions                     #    2.27  insn per cycle         
-       0.955633764 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 5279) (512y:   25) (512z:    0)
+TOTAL       :     0.997879 sec
+     2,711,169,290      cycles                           #    2.704 GHz                    
+     6,222,713,478      instructions                     #    2.30  insn per cycle         
+       1.012945753 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 5056) (512y:   24) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512y_f_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -188,26 +188,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512y_f_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512y_f_inl0_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.413316e+00
-Avg ME (F77/C++)    = 1.4133161655815059
-Relative difference = 1.1715816267550621e-07
+Avg ME (F77/C++)    = 1.4133158486847037
+Relative difference = 1.0706402269051248e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512z_f_inl0_hrd1/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[16] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.472370e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.513955e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.513955e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.374736e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.414577e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.414577e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008856e+02 +- 5.002468e+01 )  GeV^-2
-TOTAL       :     1.135039 sec
-     2,225,305,054      cycles                           #    1.953 GHz                    
-     3,698,319,936      instructions                     #    1.66  insn per cycle         
-       1.145678367 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2378) (512y:   29) (512z: 3963)
+TOTAL       :     1.216016 sec
+     2,159,440,418      cycles                           #    1.769 GHz                    
+     3,642,249,109      instructions                     #    1.69  insn per cycle         
+       1.228978695 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2070) (512y:   21) (512z: 3922)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512z_f_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -215,8 +215,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512z_f_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512z_f_inl0_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.413316e+00
-Avg ME (F77/C++)    = 1.4133164033579249
-Relative difference = 2.85398258307829e-07
+Avg ME (F77/C++)    = 1.4133164031689205
+Relative difference = 2.852645271622733e-07
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_ggttg_mad/log_ggttg_mad_m_inl0_hrd0.txt b/epochX/cudacpp/tput/logs_ggttg_mad/log_ggttg_mad_m_inl0_hrd0.txt
index ecdaa6bbe5..c3bf1d184f 100644
--- a/epochX/cudacpp/tput/logs_ggttg_mad/log_ggttg_mad_m_inl0_hrd0.txt
+++ b/epochX/cudacpp/tput/logs_ggttg_mad/log_ggttg_mad_m_inl0_hrd0.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_m_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg'
 
-DATE: 2023-11-24_14:35:27
+DATE: 2024-01-30_04:59:47
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_m_inl0_hrd0/gcheck.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 8.701296e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.042819e+07                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.057341e+07                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 8.711100e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.041363e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.056144e+07                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008920e+02 +- 5.001681e+01 )  GeV^-2
-TOTAL       :     0.460305 sec
-     2,033,483,355      cycles                           #    3.015 GHz                    
-     2,914,097,696      instructions                     #    1.43  insn per cycle         
-       0.748064785 seconds time elapsed
+TOTAL       :     0.470768 sec
+     1,937,905,575      cycles                           #    2.825 GHz                    
+     2,769,085,725      instructions                     #    1.43  insn per cycle         
+       0.764309757 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_m_inl0_hrd0/gcheck.exe -p 64 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 255
@@ -65,17 +65,17 @@ WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 .........................................................................
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_m_inl0_hrd0/gcheck.exe -p 2048 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 1.081681e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.313999e+07                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.327824e+07                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.077034e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.312199e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.326400e+07                 )  sec^-1
 MeanMatrixElemValue         = ( 6.734461e+02 +- 4.775415e+02 )  GeV^-2
-TOTAL       :     0.605547 sec
-     2,520,007,236      cycles                           #    2.997 GHz                    
-     3,773,324,991      instructions                     #    1.50  insn per cycle         
-       0.902409459 seconds time elapsed
+TOTAL       :     0.614129 sec
+     2,415,403,751      cycles                           #    2.830 GHz                    
+     3,662,132,699      instructions                     #    1.52  insn per cycle         
+       0.915037914 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_m_inl0_hrd0/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_m_inl0_hrd0/fgcheck.exe 2 64 2
@@ -86,20 +86,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_m_inl0_hrd0/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.516282e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.528433e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.528433e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.370924e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.382846e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.382846e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008920e+02 +- 5.001681e+01 )  GeV^-2
-TOTAL       :     6.534539 sec
-    20,013,949,237      cycles                           #    3.061 GHz                    
-    60,533,001,946      instructions                     #    3.02  insn per cycle         
-       6.541112859 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 1399) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     6.936117 sec
+    19,978,394,912      cycles                           #    2.879 GHz                    
+    59,162,561,873      instructions                     #    2.96  insn per cycle         
+       6.944191465 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 1149) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_m_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -113,20 +113,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.sse4_m_inl0_hrd0/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 4.912613e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.958884e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.958884e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.694585e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.741387e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.741387e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008920e+02 +- 5.001681e+01 )  GeV^-2
-TOTAL       :     3.357235 sec
-    10,182,360,341      cycles                           #    3.029 GHz                    
-    30,384,902,169      instructions                     #    2.98  insn per cycle         
-       3.367308670 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 5280) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     3.513998 sec
+    10,104,341,088      cycles                           #    2.872 GHz                    
+    29,763,867,436      instructions                     #    2.95  insn per cycle         
+       3.532062820 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 4873) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.sse4_m_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -140,20 +140,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.avx2_m_inl0_hrd0/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 9.926374e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.010833e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.010833e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 9.157849e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 9.336120e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.336120e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008920e+02 +- 5.001681e+01 )  GeV^-2
-TOTAL       :     1.672181 sec
-     4,864,081,957      cycles                           #    2.901 GHz                    
-    10,979,232,419      instructions                     #    2.26  insn per cycle         
-       1.684140082 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 4624) (512y:    0) (512z:    0)
+TOTAL       :     1.813275 sec
+     4,888,809,789      cycles                           #    2.689 GHz                    
+    11,200,775,616      instructions                     #    2.29  insn per cycle         
+       1.831194346 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 4581) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.avx2_m_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -167,20 +167,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512y_m_inl0_hrd0/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.137450e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.161103e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.161103e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.059295e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.083013e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.083013e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008920e+02 +- 5.001681e+01 )  GeV^-2
-TOTAL       :     1.462186 sec
-     4,272,825,074      cycles                           #    2.914 GHz                    
-    10,248,981,944      instructions                     #    2.40  insn per cycle         
-       1.474897364 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 4280) (512y:   82) (512z:    0)
+TOTAL       :     1.571072 sec
+     4,240,948,322      cycles                           #    2.691 GHz                    
+    10,146,075,765      instructions                     #    2.39  insn per cycle         
+       1.585395140 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 4064) (512y:   73) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512y_m_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -194,20 +194,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512z_m_inl0_hrd0/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.677106e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.786537e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.786537e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 7.157625e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.268151e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.268151e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008920e+02 +- 5.001681e+01 )  GeV^-2
-TOTAL       :     2.158029 sec
-     4,199,815,524      cycles                           #    1.943 GHz                    
-     6,043,516,368      instructions                     #    1.44  insn per cycle         
-       2.169397232 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2066) (512y:  117) (512z: 3540)
+TOTAL       :     2.315387 sec
+     4,011,221,101      cycles                           #    1.729 GHz                    
+     5,838,969,816      instructions                     #    1.46  insn per cycle         
+       2.328222904 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1778) (512y:   97) (512z: 3502)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512z_m_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -215,8 +215,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512z_m_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512z_m_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.413122e+00
-Avg ME (F77/C++)    = 1.4131213786174055
-Relative difference = 4.3972324717191576e-07
+Avg ME (F77/C++)    = 1.4131213600217192
+Relative difference = 4.5288254008796884e-07
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_ggttg_mad/log_ggttg_mad_m_inl0_hrd1.txt b/epochX/cudacpp/tput/logs_ggttg_mad/log_ggttg_mad_m_inl0_hrd1.txt
index f9629437a1..0465a21327 100644
--- a/epochX/cudacpp/tput/logs_ggttg_mad/log_ggttg_mad_m_inl0_hrd1.txt
+++ b/epochX/cudacpp/tput/logs_ggttg_mad/log_ggttg_mad_m_inl0_hrd1.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_m_inl0_hrd1'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg'
 
-DATE: 2023-11-24_14:35:56
+DATE: 2024-01-30_05:00:18
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_m_inl0_hrd1/gcheck.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 8.718207e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.043711e+07                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.058000e+07                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 8.666023e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.032901e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.046936e+07                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008920e+02 +- 5.001681e+01 )  GeV^-2
-TOTAL       :     0.460190 sec
-     2,020,498,162      cycles                           #    3.001 GHz                    
-     2,893,283,708      instructions                     #    1.43  insn per cycle         
-       0.740481359 seconds time elapsed
+TOTAL       :     0.468226 sec
+     1,937,873,508      cycles                           #    2.824 GHz                    
+     2,759,069,461      instructions                     #    1.42  insn per cycle         
+       0.754370762 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_m_inl0_hrd1/gcheck.exe -p 64 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 255
@@ -65,17 +65,17 @@ WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 .........................................................................
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_m_inl0_hrd1/gcheck.exe -p 2048 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 1.069988e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.302141e+07                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.315703e+07                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.070939e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.304690e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.318717e+07                 )  sec^-1
 MeanMatrixElemValue         = ( 6.734461e+02 +- 4.775415e+02 )  GeV^-2
-TOTAL       :     0.603031 sec
-     2,496,488,169      cycles                           #    2.975 GHz                    
-     3,745,992,997      instructions                     #    1.50  insn per cycle         
-       0.898963383 seconds time elapsed
+TOTAL       :     0.607643 sec
+     2,403,195,178      cycles                           #    2.827 GHz                    
+     3,555,740,714      instructions                     #    1.48  insn per cycle         
+       0.909921855 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_m_inl0_hrd1/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_m_inl0_hrd1/fgcheck.exe 2 64 2
@@ -86,20 +86,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_m_inl0_hrd1/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.573744e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.586335e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.586335e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.404971e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.417275e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.417275e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008920e+02 +- 5.001681e+01 )  GeV^-2
-TOTAL       :     6.388339 sec
-    19,817,061,675      cycles                           #    3.101 GHz                    
-    59,934,412,865      instructions                     #    3.02  insn per cycle         
-       6.394993907 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 1276) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     6.840299 sec
+    19,736,673,501      cycles                           #    2.886 GHz                    
+    58,709,690,472      instructions                     #    2.97  insn per cycle         
+       6.847451518 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 1026) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.none_m_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -113,20 +113,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.sse4_m_inl0_hrd1/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 5.068721e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 5.116534e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 5.116534e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.708829e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.755468e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.755468e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008920e+02 +- 5.001681e+01 )  GeV^-2
-TOTAL       :     3.254189 sec
-    10,074,392,918      cycles                           #    3.092 GHz                    
-    30,097,970,506      instructions                     #    2.99  insn per cycle         
-       3.269644857 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 5082) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     3.503073 sec
+    10,118,973,746      cycles                           #    2.885 GHz                    
+    30,158,905,101      instructions                     #    2.98  insn per cycle         
+       3.519090284 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 4944) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.sse4_m_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -140,20 +140,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.avx2_m_inl0_hrd1/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 9.551013e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 9.718637e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 9.718637e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 8.784663e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 8.950747e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.950747e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008920e+02 +- 5.001681e+01 )  GeV^-2
-TOTAL       :     1.737428 sec
-     5,013,997,017      cycles                           #    2.879 GHz                    
-    11,483,477,844      instructions                     #    2.29  insn per cycle         
-       1.752998506 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 4723) (512y:    0) (512z:    0)
+TOTAL       :     1.889159 sec
+     5,039,949,395      cycles                           #    2.661 GHz                    
+    11,663,409,755      instructions                     #    2.31  insn per cycle         
+       1.981495827 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 4685) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.avx2_m_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -167,20 +167,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512y_m_inl0_hrd1/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.052618e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.072863e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.072863e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 9.838137e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.004758e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.004758e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008920e+02 +- 5.001681e+01 )  GeV^-2
-TOTAL       :     1.578372 sec
-     4,590,458,207      cycles                           #    2.900 GHz                    
-    10,810,507,291      instructions                     #    2.35  insn per cycle         
-       1.591783419 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 4285) (512y:  234) (512z:    0)
+TOTAL       :     1.689521 sec
+     4,555,347,979      cycles                           #    2.689 GHz                    
+    10,787,640,248      instructions                     #    2.37  insn per cycle         
+       1.702819632 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 4159) (512y:  233) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512y_m_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -194,20 +194,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512z_m_inl0_hrd1/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.627027e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.731488e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.731488e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 7.077813e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.181685e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.181685e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 1.008920e+02 +- 5.001681e+01 )  GeV^-2
-TOTAL       :     2.171750 sec
-     4,214,575,835      cycles                           #    1.937 GHz                    
-     6,273,425,611      instructions                     #    1.49  insn per cycle         
-       2.183570227 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1961) (512y:  163) (512z: 3617)
+TOTAL       :     2.340568 sec
+     4,064,413,524      cycles                           #    1.733 GHz                    
+     6,073,601,897      instructions                     #    1.49  insn per cycle         
+       2.356439472 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1725) (512y:  104) (512z: 3609)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512z_m_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -215,8 +215,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512z_m_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttg.mad/SubProcesses/P1_gg_ttxg/build.512z_m_inl0_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.413122e+00
-Avg ME (F77/C++)    = 1.4131213786174055
-Relative difference = 4.3972324717191576e-07
+Avg ME (F77/C++)    = 1.4131213600217192
+Relative difference = 4.5288254008796884e-07
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_d_inl0_hrd0.txt b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_d_inl0_hrd0.txt
index 9b25288b8a..53bd28a5bd 100644
--- a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_d_inl0_hrd0.txt
+++ b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_d_inl0_hrd0.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
 
-DATE: 2023-11-24_14:36:25
+DATE: 2024-01-30_05:00:48
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd0/gcheck.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 3.502935e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.531164e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.533426e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.507010e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.536029e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.538733e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     0.519828 sec
-     2,267,325,261      cycles                           #    3.026 GHz                    
-     3,564,664,052      instructions                     #    1.57  insn per cycle         
-       0.820257033 seconds time elapsed
+TOTAL       :     0.531942 sec
+     2,193,462,671      cycles                           #    2.834 GHz                    
+     3,356,973,773      instructions                     #    1.53  insn per cycle         
+       0.849346656 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd0/gcheck.exe -p 64 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 255
@@ -65,17 +65,17 @@ WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 .........................................................................
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd0/gcheck.exe -p 2048 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 4.129197e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.163245e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.164643e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.126743e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.160620e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.162100e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 6.665112e+00 +- 5.002651e+00 )  GeV^-4
-TOTAL       :     3.029980 sec
-    10,118,272,870      cycles                           #    3.068 GHz                    
-    21,084,266,771      instructions                     #    2.08  insn per cycle         
-       3.354068412 seconds time elapsed
+TOTAL       :     3.043092 sec
+     9,489,937,514      cycles                           #    2.875 GHz                    
+    19,463,317,431      instructions                     #    2.05  insn per cycle         
+       3.359518634 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd0/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd0/fgcheck.exe 2 64 2
@@ -86,20 +86,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd0/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.983194e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.984194e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.984194e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.787937e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.788754e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.788754e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     8.279983 sec
-    25,558,009,617      cycles                           #    3.086 GHz                    
-    78,936,663,909      instructions                     #    3.09  insn per cycle         
-       8.286888065 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 4893) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     9.184792 sec
+    26,445,376,310      cycles                           #    2.879 GHz                    
+    81,759,262,253      instructions                     #    3.09  insn per cycle         
+       9.200099621 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 6614) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -113,20 +113,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_d_inl0_hrd0/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.793332e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.796868e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.796868e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.595033e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.598347e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.598347e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     4.332856 sec
-    12,910,305,211      cycles                           #    2.977 GHz                    
-    39,281,355,341      instructions                     #    3.04  insn per cycle         
-       4.346559861 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:13184) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     4.572055 sec
+    12,894,491,420      cycles                           #    2.818 GHz                    
+    39,242,650,330      instructions                     #    3.04  insn per cycle         
+       4.588188651 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:12814) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -140,20 +140,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_d_inl0_hrd0/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 8.457803e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 8.475319e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 8.475319e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 7.988905e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 8.005063e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.005063e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     1.950206 sec
-     5,559,743,200      cycles                           #    2.847 GHz                    
-    13,686,596,932      instructions                     #    2.46  insn per cycle         
-       2.032901077 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:11357) (512y:    0) (512z:    0)
+TOTAL       :     2.062926 sec
+     5,559,157,847      cycles                           #    2.689 GHz                    
+    13,789,744,695      instructions                     #    2.48  insn per cycle         
+       2.079268197 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:11059) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -161,26 +161,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.626675e-04
-Avg ME (F77/C++)    = 6.6266731198157320E-004
-Relative difference = 2.837296634927675e-07
+Avg ME (F77/C++)    = 6.6266731198157309E-004
+Relative difference = 2.837296636563793e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_d_inl0_hrd0/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 9.800481e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 9.822076e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 9.822076e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 9.113130e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 9.134504e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.134504e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     1.683777 sec
-     4,896,997,214      cycles                           #    2.903 GHz                    
-    12,341,617,992      instructions                     #    2.52  insn per cycle         
-       1.697194330 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:10266) (512y:   88) (512z:    0)
+TOTAL       :     1.809806 sec
+     4,899,980,729      cycles                           #    2.701 GHz                    
+    12,319,200,932      instructions                     #    2.51  insn per cycle         
+       1.824526773 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 9762) (512y:   94) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -188,26 +188,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.626675e-04
-Avg ME (F77/C++)    = 6.6266731198157320E-004
-Relative difference = 2.837296634927675e-07
+Avg ME (F77/C++)    = 6.6266731198157309E-004
+Relative difference = 2.837296636563793e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_d_inl0_hrd0/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.669824e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.683748e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.683748e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.926484e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.938620e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.938620e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     2.147707 sec
-     4,120,375,944      cycles                           #    1.915 GHz                    
-     6,335,709,132      instructions                     #    1.54  insn per cycle         
-       2.163638241 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1821) (512y:  102) (512z: 9375)
+TOTAL       :     2.377893 sec
+     4,078,713,187      cycles                           #    1.712 GHz                    
+     6,287,612,851      instructions                     #    1.54  insn per cycle         
+       2.391138362 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1516) (512y:   94) (512z: 9019)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -215,8 +215,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.626675e-04
-Avg ME (F77/C++)    = 6.6266731198157320E-004
-Relative difference = 2.837296634927675e-07
+Avg ME (F77/C++)    = 6.6266731198157309E-004
+Relative difference = 2.837296636563793e-07
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_d_inl0_hrd0_bridge.txt b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_d_inl0_hrd0_bridge.txt
index 3e466c9dbd..ba45d149aa 100644
--- a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_d_inl0_hrd0_bridge.txt
+++ b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_d_inl0_hrd0_bridge.txt
@@ -41,7 +41,7 @@ CUDACPP_BUILDDIR='build.512z_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
 
-DATE: 2023-11-24_15:13:18
+DATE: 2024-01-30_05:49:05
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
@@ -51,17 +51,17 @@ WARNING! Bridge selected: cannot use RamboDevice, will use RamboHost
 WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
 WARNING! Instantiate device Bridge (nevt=16384, gpublocks=64, gputhreads=256, gpublocks*gputhreads=16384)
 WARNING! Set grid in Bridge (nevt=16384, gpublocks=64, gputhreads=256, gpublocks*gputhreads=16384)
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 3.151810e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.491547e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.491547e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.099881e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.447326e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.447326e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     0.513888 sec
-     2,172,731,633      cycles                           #    2.929 GHz                    
-     3,459,636,945      instructions                     #    1.59  insn per cycle         
-       0.802070844 seconds time elapsed
+TOTAL       :     0.520843 sec
+     2,128,361,154      cycles                           #    2.833 GHz                    
+     3,379,769,914      instructions                     #    1.59  insn per cycle         
+       0.811374229 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd0/gcheck.exe -p 64 256 1 --bridge
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Bridge selected: cannot use RamboDevice, will use RamboHost
@@ -77,17 +77,17 @@ WARNING! Bridge selected: cannot use RamboDevice, will use RamboHost
 WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
 WARNING! Instantiate device Bridge (nevt=524288, gpublocks=2048, gputhreads=256, gpublocks*gputhreads=524288)
 WARNING! Set grid in Bridge (nevt=524288, gpublocks=2048, gputhreads=256, gpublocks*gputhreads=524288)
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 3.626462e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.109948e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.109948e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.602295e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.096469e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.096469e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 6.665112e+00 +- 5.002651e+00 )  GeV^-4
-TOTAL       :     3.316369 sec
-    10,702,439,296      cycles                           #    2.980 GHz                    
-    23,311,698,840      instructions                     #    2.18  insn per cycle         
-       3.648833219 seconds time elapsed
+TOTAL       :     3.329754 sec
+    10,358,756,104      cycles                           #    2.872 GHz                    
+    22,944,085,739      instructions                     #    2.21  insn per cycle         
+       3.663397648 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd0/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd0/fgcheck.exe 2 64 2
@@ -99,20 +99,20 @@ OK (relative difference <= 5E-3)
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd0/check.exe -p 64 256 1 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=16384)
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.926732e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.927660e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.927660e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.794771e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.795632e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.795632e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     8.525962 sec
-    25,585,926,017      cycles                           #    3.000 GHz                    
-    78,942,573,288      instructions                     #    3.09  insn per cycle         
-       8.530463803 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 4893) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     9.153536 sec
+    26,441,951,782      cycles                           #    2.888 GHz                    
+    81,759,972,796      instructions                     #    3.09  insn per cycle         
+       9.158879879 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 6614) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -127,20 +127,20 @@ OK (relative difference <= 5E-3)
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_d_inl0_hrd0/check.exe -p 64 256 1 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=16384)
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/sse4+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.644154e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.647807e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.647807e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.577595e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.580993e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.580993e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     4.514194 sec
-    12,913,516,064      cycles                           #    2.858 GHz                    
-    39,293,095,640      instructions                     #    3.04  insn per cycle         
-       4.518821334 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:13184) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     4.598491 sec
+    12,916,287,273      cycles                           #    2.806 GHz                    
+    39,254,753,938      instructions                     #    3.04  insn per cycle         
+       4.603937867 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:12814) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -155,20 +155,20 @@ OK (relative difference <= 5E-3)
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_d_inl0_hrd0/check.exe -p 64 256 1 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=16384)
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/avx2+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 8.272317e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 8.289989e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 8.289989e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 7.852795e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.869019e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.869019e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     1.996165 sec
-     5,572,431,089      cycles                           #    2.786 GHz                    
-    13,696,683,094      instructions                     #    2.46  insn per cycle         
-       2.000824809 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:11357) (512y:    0) (512z:    0)
+TOTAL       :     2.103011 sec
+     5,568,678,671      cycles                           #    2.642 GHz                    
+    13,799,771,926      instructions                     #    2.48  insn per cycle         
+       2.108561686 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:11059) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -176,27 +176,27 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.626675e-04
-Avg ME (F77/C++)    = 6.6266731198157320E-004
-Relative difference = 2.837296634927675e-07
+Avg ME (F77/C++)    = 6.6266731198157309E-004
+Relative difference = 2.837296636563793e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_d_inl0_hrd0/check.exe -p 64 256 1 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=16384)
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/512y+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 9.416662e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 9.438491e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 9.438491e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 9.035305e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 9.056800e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.056800e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     1.755183 sec
-     4,929,995,053      cycles                           #    2.803 GHz                    
-    12,351,698,776      instructions                     #    2.51  insn per cycle         
-       1.759689895 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:10266) (512y:   88) (512z:    0)
+TOTAL       :     1.829572 sec
+     4,921,598,332      cycles                           #    2.684 GHz                    
+    12,328,469,851      instructions                     #    2.50  insn per cycle         
+       1.835230648 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 9762) (512y:   94) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -204,27 +204,27 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.626675e-04
-Avg ME (F77/C++)    = 6.6266731198157320E-004
-Relative difference = 2.837296634927675e-07
+Avg ME (F77/C++)    = 6.6266731198157309E-004
+Relative difference = 2.837296636563793e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_d_inl0_hrd0/check.exe -p 64 256 1 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=16384)
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/512z+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.299463e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.313710e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.313710e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.926825e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.939647e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.939647e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     2.260606 sec
-     4,129,998,329      cycles                           #    1.824 GHz                    
-     6,345,886,658      instructions                     #    1.54  insn per cycle         
-       2.265252390 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1821) (512y:  102) (512z: 9375)
+TOTAL       :     2.382359 sec
+     4,075,002,441      cycles                           #    1.707 GHz                    
+     6,297,411,526      instructions                     #    1.55  insn per cycle         
+       2.387952463 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1516) (512y:   94) (512z: 9019)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -232,8 +232,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.626675e-04
-Avg ME (F77/C++)    = 6.6266731198157320E-004
-Relative difference = 2.837296634927675e-07
+Avg ME (F77/C++)    = 6.6266731198157309E-004
+Relative difference = 2.837296636563793e-07
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_d_inl0_hrd0_common.txt b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_d_inl0_hrd0_common.txt
index 22d785bafe..2624aa384f 100644
--- a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_d_inl0_hrd0_common.txt
+++ b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_d_inl0_hrd0_common.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
 
-DATE: 2023-11-24_15:25:10
+DATE: 2024-01-30_06:01:23
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd0/gcheck.exe -p 64 256 1 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:COMMON+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 3.505435e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.534142e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.536887e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.497090e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.524372e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.526818e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 4.197467e-01 +- 3.250467e-01 )  GeV^-4
-TOTAL       :     0.512729 sec
-     2,187,113,796      cycles                           #    2.953 GHz                    
-     3,321,864,960      instructions                     #    1.52  insn per cycle         
-       0.803004683 seconds time elapsed
+TOTAL       :     0.512897 sec
+     2,098,983,374      cycles                           #    2.834 GHz                    
+     3,277,353,449      instructions                     #    1.56  insn per cycle         
+       0.803360908 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd0/gcheck.exe -p 64 256 1 --common
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 255
@@ -65,17 +65,17 @@ WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 .........................................................................
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd0/gcheck.exe -p 2048 256 1 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:COMMON+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 4.143311e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.178194e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.179712e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.141054e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.174803e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.176256e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 1.252232e+02 +- 1.234346e+02 )  GeV^-4
-TOTAL       :     3.127476 sec
-    10,173,806,494      cycles                           #    3.006 GHz                    
-    22,940,336,221      instructions                     #    2.25  insn per cycle         
-       3.440988376 seconds time elapsed
+TOTAL       :     3.134385 sec
+     9,742,325,189      cycles                           #    2.872 GHz                    
+    21,219,396,735      instructions                     #    2.18  insn per cycle         
+       3.451782991 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd0/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd0/fgcheck.exe 2 64 2
@@ -86,20 +86,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd0/check.exe -p 64 256 1 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:COMMON+RMBHST+MESHST/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.944681e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.945656e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.945656e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.789322e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.790133e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.790133e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.197467e-01 +- 3.250467e-01 )  GeV^-4
-TOTAL       :     8.444658 sec
-    25,580,623,755      cycles                           #    3.028 GHz                    
-    78,936,122,811      instructions                     #    3.09  insn per cycle         
-       8.449050059 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 4893) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     9.178912 sec
+    26,467,085,384      cycles                           #    2.885 GHz                    
+    81,758,395,147      instructions                     #    3.09  insn per cycle         
+       9.184185479 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 6614) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -113,20 +113,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_d_inl0_hrd0/check.exe -p 64 256 1 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:COMMON+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.654154e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.657693e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.657693e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.580434e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.583873e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.583873e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.197467e-01 +- 3.250467e-01 )  GeV^-4
-TOTAL       :     4.499225 sec
-    12,919,732,335      cycles                           #    2.870 GHz                    
-    39,280,003,953      instructions                     #    3.04  insn per cycle         
-       4.503503119 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:13184) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     4.592041 sec
+    12,908,303,532      cycles                           #    2.809 GHz                    
+    39,241,301,392      instructions                     #    3.04  insn per cycle         
+       4.597199751 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:12814) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -140,20 +140,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_d_inl0_hrd0/check.exe -p 64 256 1 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:COMMON+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 8.335985e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 8.352449e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 8.352449e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 8.006274e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 8.022952e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.022952e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.197467e-01 +- 3.250467e-01 )  GeV^-4
-TOTAL       :     1.978535 sec
-     5,565,762,005      cycles                           #    2.808 GHz                    
-    13,684,790,936      instructions                     #    2.46  insn per cycle         
-       1.982835811 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:11357) (512y:    0) (512z:    0)
+TOTAL       :     2.060575 sec
+     5,561,277,799      cycles                           #    2.694 GHz                    
+    13,787,529,346      instructions                     #    2.48  insn per cycle         
+       2.065507699 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:11059) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -161,26 +161,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.626675e-04
-Avg ME (F77/C++)    = 6.6266731198157320E-004
-Relative difference = 2.837296634927675e-07
+Avg ME (F77/C++)    = 6.6266731198157309E-004
+Relative difference = 2.837296636563793e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_d_inl0_hrd0/check.exe -p 64 256 1 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:COMMON+RMBHST+MESHST/512y+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 9.460859e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 9.482779e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 9.482779e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 9.108001e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 9.130506e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.130506e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.197467e-01 +- 3.250467e-01 )  GeV^-4
-TOTAL       :     1.744747 sec
-     4,901,917,953      cycles                           #    2.804 GHz                    
-    12,339,116,177      instructions                     #    2.52  insn per cycle         
-       1.749045721 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:10266) (512y:   88) (512z:    0)
+TOTAL       :     1.812494 sec
+     4,903,037,786      cycles                           #    2.699 GHz                    
+    12,315,866,756      instructions                     #    2.51  insn per cycle         
+       1.817504411 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 9762) (512y:   94) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -188,26 +188,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.626675e-04
-Avg ME (F77/C++)    = 6.6266731198157320E-004
-Relative difference = 2.837296634927675e-07
+Avg ME (F77/C++)    = 6.6266731198157309E-004
+Relative difference = 2.837296636563793e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_d_inl0_hrd0/check.exe -p 64 256 1 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:COMMON+RMBHST+MESHST/512z+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.321634e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.334985e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.334985e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.888313e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.900941e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.900941e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.197467e-01 +- 3.250467e-01 )  GeV^-4
-TOTAL       :     2.251368 sec
-     4,121,273,998      cycles                           #    1.828 GHz                    
-     6,332,363,558      instructions                     #    1.54  insn per cycle         
-       2.255694764 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1821) (512y:  102) (512z: 9375)
+TOTAL       :     2.393126 sec
+     4,056,497,728      cycles                           #    1.692 GHz                    
+     6,284,230,028      instructions                     #    1.55  insn per cycle         
+       2.398190383 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1516) (512y:   94) (512z: 9019)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -215,8 +215,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.626675e-04
-Avg ME (F77/C++)    = 6.6266731198157320E-004
-Relative difference = 2.837296634927675e-07
+Avg ME (F77/C++)    = 6.6266731198157309E-004
+Relative difference = 2.837296636563793e-07
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_d_inl0_hrd0_curhst.txt b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_d_inl0_hrd0_curhst.txt
index 0edaf6e67f..711141aac6 100644
--- a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_d_inl0_hrd0_curhst.txt
+++ b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_d_inl0_hrd0_curhst.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
 
-DATE: 2023-11-24_15:21:49
+DATE: 2024-01-30_05:57:54
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd0/gcheck.exe -p 64 256 1 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 3.504189e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.533338e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.535687e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.493459e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.521487e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.524248e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     0.507565 sec
-     2,140,644,436      cycles                           #    2.903 GHz                    
-     3,294,278,423      instructions                     #    1.54  insn per cycle         
-       0.797247733 seconds time elapsed
+TOTAL       :     0.509745 sec
+     2,103,800,649      cycles                           #    2.836 GHz                    
+     3,325,789,020      instructions                     #    1.58  insn per cycle         
+       0.801087014 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd0/gcheck.exe -p 64 256 1 --curhst
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 255
@@ -65,17 +65,17 @@ WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 .........................................................................
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd0/gcheck.exe -p 2048 256 1 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 4.134584e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.168777e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.170193e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.145753e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.180203e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.181702e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 6.665112e+00 +- 5.002651e+00 )  GeV^-4
-TOTAL       :     3.073697 sec
-     9,988,823,734      cycles                           #    2.997 GHz                    
-    22,688,491,791      instructions                     #    2.27  insn per cycle         
-       3.391117560 seconds time elapsed
+TOTAL       :     3.079125 sec
+     9,600,126,952      cycles                           #    2.878 GHz                    
+    21,681,876,510      instructions                     #    2.26  insn per cycle         
+       3.393235673 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd0/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd0/fgcheck.exe 2 64 2
@@ -86,20 +86,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd0/check.exe -p 64 256 1 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.920556e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.921467e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.921467e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.797758e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.798583e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.798583e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     8.549047 sec
-    25,566,043,933      cycles                           #    2.990 GHz                    
-    78,938,822,238      instructions                     #    3.09  insn per cycle         
-       8.553277802 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 4893) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     9.135643 sec
+    26,454,582,305      cycles                           #    2.896 GHz                    
+    81,754,058,548      instructions                     #    3.09  insn per cycle         
+       9.140745485 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 6614) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -113,20 +113,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_d_inl0_hrd0/check.exe -p 64 256 1 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.702947e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.706522e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.706522e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.597207e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.600539e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.600539e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     4.438616 sec
-    12,919,298,910      cycles                           #    2.909 GHz                    
-    39,279,556,753      instructions                     #    3.04  insn per cycle         
-       4.442989895 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:13184) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     4.569387 sec
+    12,892,653,048      cycles                           #    2.819 GHz                    
+    39,241,760,724      instructions                     #    3.04  insn per cycle         
+       4.574378716 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:12814) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -140,20 +140,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_d_inl0_hrd0/check.exe -p 64 256 1 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 8.327524e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 8.345264e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 8.345264e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 7.978625e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.995447e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.995447e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     1.979008 sec
-     5,559,052,885      cycles                           #    2.804 GHz                    
-    13,685,718,134      instructions                     #    2.46  insn per cycle         
-       1.983236201 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:11357) (512y:    0) (512z:    0)
+TOTAL       :     2.065609 sec
+     5,559,302,417      cycles                           #    2.687 GHz                    
+    13,789,202,442      instructions                     #    2.48  insn per cycle         
+       2.071000161 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:11059) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -161,26 +161,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.626675e-04
-Avg ME (F77/C++)    = 6.6266731198157320E-004
-Relative difference = 2.837296634927675e-07
+Avg ME (F77/C++)    = 6.6266731198157309E-004
+Relative difference = 2.837296636563793e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_d_inl0_hrd0/check.exe -p 64 256 1 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 9.547712e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 9.570537e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 9.570537e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 9.097696e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 9.119952e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.119952e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     1.726945 sec
-     4,895,458,327      cycles                           #    2.829 GHz                    
-    12,340,699,922      instructions                     #    2.52  insn per cycle         
-       1.731178551 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:10266) (512y:   88) (512z:    0)
+TOTAL       :     1.812868 sec
+     4,896,837,509      cycles                           #    2.695 GHz                    
+    12,317,770,581      instructions                     #    2.52  insn per cycle         
+       1.818257681 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 9762) (512y:   94) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -188,26 +188,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.626675e-04
-Avg ME (F77/C++)    = 6.6266731198157320E-004
-Relative difference = 2.837296634927675e-07
+Avg ME (F77/C++)    = 6.6266731198157309E-004
+Relative difference = 2.837296636563793e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_d_inl0_hrd0/check.exe -p 64 256 1 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.393984e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.407662e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.407662e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.967466e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.979997e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.979997e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     2.227409 sec
-     4,117,035,910      cycles                           #    1.846 GHz                    
-     6,334,322,153      instructions                     #    1.54  insn per cycle         
-       2.231712529 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1821) (512y:  102) (512z: 9375)
+TOTAL       :     2.364567 sec
+     4,060,623,360      cycles                           #    1.715 GHz                    
+     6,286,167,500      instructions                     #    1.55  insn per cycle         
+       2.369629620 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1516) (512y:   94) (512z: 9019)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -215,8 +215,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.626675e-04
-Avg ME (F77/C++)    = 6.6266731198157320E-004
-Relative difference = 2.837296634927675e-07
+Avg ME (F77/C++)    = 6.6266731198157309E-004
+Relative difference = 2.837296636563793e-07
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_d_inl0_hrd0_rmbhst.txt b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_d_inl0_hrd0_rmbhst.txt
index 3fd3545f79..de6151d7b3 100644
--- a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_d_inl0_hrd0_rmbhst.txt
+++ b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_d_inl0_hrd0_rmbhst.txt
@@ -41,24 +41,24 @@ CUDACPP_BUILDDIR='build.512z_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
 
-DATE: 2023-11-24_15:18:31
+DATE: 2024-01-30_05:54:28
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd0/gcheck.exe -p 64 256 1 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 3.205483e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.519039e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.521349e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.181803e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.496640e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.499302e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     0.513661 sec
-     2,155,662,048      cycles                           #    2.911 GHz                    
-     3,403,110,868      instructions                     #    1.58  insn per cycle         
-       0.802490530 seconds time elapsed
+TOTAL       :     0.513215 sec
+     2,110,243,378      cycles                           #    2.841 GHz                    
+     3,364,158,559      instructions                     #    1.59  insn per cycle         
+       0.803846009 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd0/gcheck.exe -p 64 256 1 --rmbhst
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
@@ -68,17 +68,17 @@ WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd0/gcheck.exe -p 2048 256 1 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 3.731558e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.171074e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.172534e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.724341e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.176977e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.178501e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 6.665112e+00 +- 5.002651e+00 )  GeV^-4
-TOTAL       :     3.204951 sec
-    10,303,132,790      cycles                           #    2.977 GHz                    
-    23,518,550,603      instructions                     #    2.28  insn per cycle         
-       3.516665455 seconds time elapsed
+TOTAL       :     3.211617 sec
+     9,930,008,722      cycles                           #    2.863 GHz                    
+    21,629,593,771      instructions                     #    2.18  insn per cycle         
+       3.536993543 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd0/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd0/fgcheck.exe 2 64 2
@@ -89,20 +89,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd0/check.exe -p 64 256 1 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.916291e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.917249e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.917249e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.795958e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.796814e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.796814e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     8.568789 sec
-    25,583,500,760      cycles                           #    2.987 GHz                    
-    78,939,465,924      instructions                     #    3.09  insn per cycle         
-       8.573085981 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 4893) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     9.141774 sec
+    26,442,082,623      cycles                           #    2.892 GHz                    
+    81,755,899,902      instructions                     #    3.09  insn per cycle         
+       9.146895276 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 6614) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -116,20 +116,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_d_inl0_hrd0/check.exe -p 64 256 1 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.689818e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.693400e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.693400e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.584252e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.587667e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.587667e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     4.453935 sec
-    12,901,075,488      cycles                           #    2.894 GHz                    
-    39,279,441,065      instructions                     #    3.04  insn per cycle         
-       4.458436593 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:13184) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     4.585664 sec
+    12,903,354,074      cycles                           #    2.812 GHz                    
+    39,243,037,589      instructions                     #    3.04  insn per cycle         
+       4.591083081 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:12814) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -143,20 +143,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_d_inl0_hrd0/check.exe -p 64 256 1 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 8.313071e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 8.330635e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 8.330635e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 7.993074e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 8.009513e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.009513e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     1.982693 sec
-     5,557,103,226      cycles                           #    2.798 GHz                    
-    13,686,016,474      instructions                     #    2.46  insn per cycle         
-       1.987071584 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:11357) (512y:    0) (512z:    0)
+TOTAL       :     2.061753 sec
+     5,556,410,491      cycles                           #    2.690 GHz                    
+    13,788,754,708      instructions                     #    2.48  insn per cycle         
+       2.066810636 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:11059) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -164,26 +164,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.626675e-04
-Avg ME (F77/C++)    = 6.6266731198157320E-004
-Relative difference = 2.837296634927675e-07
+Avg ME (F77/C++)    = 6.6266731198157309E-004
+Relative difference = 2.837296636563793e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_d_inl0_hrd0/check.exe -p 64 256 1 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 9.334052e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 9.355862e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 9.355862e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 9.089775e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 9.111272e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.111272e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     1.766820 sec
-     4,905,253,884      cycles                           #    2.771 GHz                    
-    12,341,018,898      instructions                     #    2.52  insn per cycle         
-       1.771164761 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:10266) (512y:   88) (512z:    0)
+TOTAL       :     1.814284 sec
+     4,898,229,262      cycles                           #    2.694 GHz                    
+    12,317,871,193      instructions                     #    2.51  insn per cycle         
+       1.819291757 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 9762) (512y:   94) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -191,26 +191,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.626675e-04
-Avg ME (F77/C++)    = 6.6266731198157320E-004
-Relative difference = 2.837296634927675e-07
+Avg ME (F77/C++)    = 6.6266731198157309E-004
+Relative difference = 2.837296636563793e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_d_inl0_hrd0/check.exe -p 64 256 1 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.271568e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.284665e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.284665e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.893591e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.906421e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.906421e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     2.265152 sec
-     4,121,020,497      cycles                           #    1.817 GHz                    
-     6,334,252,603      instructions                     #    1.54  insn per cycle         
-       2.269401421 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1821) (512y:  102) (512z: 9375)
+TOTAL       :     2.392306 sec
+     4,056,818,337      cycles                           #    1.695 GHz                    
+     6,287,135,022      instructions                     #    1.55  insn per cycle         
+       2.397424437 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1516) (512y:   94) (512z: 9019)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -218,8 +218,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.626675e-04
-Avg ME (F77/C++)    = 6.6266731198157320E-004
-Relative difference = 2.837296634927675e-07
+Avg ME (F77/C++)    = 6.6266731198157309E-004
+Relative difference = 2.837296636563793e-07
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_d_inl0_hrd1.txt b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_d_inl0_hrd1.txt
index 98140f2185..ce8b9bfd9b 100644
--- a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_d_inl0_hrd1.txt
+++ b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_d_inl0_hrd1.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_d_inl0_hrd1'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
 
-DATE: 2023-11-24_14:37:01
+DATE: 2024-01-30_05:01:27
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd1/gcheck.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 3.470716e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.498661e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.501013e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.464704e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.493618e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.496312e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     0.519808 sec
-     2,262,549,095      cycles                           #    3.015 GHz                    
-     3,507,447,767      instructions                     #    1.55  insn per cycle         
-       0.822215819 seconds time elapsed
+TOTAL       :     0.530379 sec
+     2,191,459,528      cycles                           #    2.836 GHz                    
+     3,378,194,635      instructions                     #    1.54  insn per cycle         
+       0.862349447 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd1/gcheck.exe -p 64 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 255
@@ -65,17 +65,17 @@ WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 .........................................................................
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd1/gcheck.exe -p 2048 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 4.130084e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.164129e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.165510e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.136041e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.170363e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.171805e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 6.665112e+00 +- 5.002651e+00 )  GeV^-4
-TOTAL       :     3.030552 sec
-    10,073,054,045      cycles                           #    3.064 GHz                    
-    22,850,649,130      instructions                     #    2.27  insn per cycle         
-       3.344535702 seconds time elapsed
+TOTAL       :     3.033894 sec
+     9,468,330,012      cycles                           #    2.874 GHz                    
+    21,262,061,450      instructions                     #    2.25  insn per cycle         
+       3.350179318 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd1/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd1/fgcheck.exe 2 64 2
@@ -86,20 +86,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd1/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.990080e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.991047e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.991047e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.798336e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.799213e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.799213e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     8.251874 sec
-    25,580,916,000      cycles                           #    3.099 GHz                    
-    78,707,540,592      instructions                     #    3.08  insn per cycle         
-       8.258581252 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 4264) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     9.130694 sec
+    26,439,863,153      cycles                           #    2.895 GHz                    
+    81,781,637,155      instructions                     #    3.09  insn per cycle         
+       9.163718345 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 6589) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -113,20 +113,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_d_inl0_hrd1/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.671887e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.674997e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.674997e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.559639e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.562995e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.562995e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     4.475213 sec
-    12,954,555,110      cycles                           #    2.897 GHz                    
-    39,230,579,465      instructions                     #    3.03  insn per cycle         
-       4.487081742 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:12951) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     4.616847 sec
+    12,919,257,236      cycles                           #    2.796 GHz                    
+    39,249,733,665      instructions                     #    3.04  insn per cycle         
+       4.636578065 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:12771) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_d_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -140,20 +140,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_d_inl0_hrd1/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 8.529282e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 8.546223e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 8.546223e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 8.030089e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 8.046612e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.046612e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     1.932123 sec
-     5,603,928,357      cycles                           #    2.895 GHz                    
-    13,800,807,908      instructions                     #    2.46  insn per cycle         
-       1.945155802 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:11422) (512y:    0) (512z:    0)
+TOTAL       :     2.052281 sec
+     5,556,604,473      cycles                           #    2.701 GHz                    
+    13,805,088,947      instructions                     #    2.48  insn per cycle         
+       2.071717259 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:11048) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_d_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -161,26 +161,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_d_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_d_inl0_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.626675e-04
-Avg ME (F77/C++)    = 6.6266731198157320E-004
-Relative difference = 2.837296634927675e-07
+Avg ME (F77/C++)    = 6.6266731198157309E-004
+Relative difference = 2.837296636563793e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_d_inl0_hrd1/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 9.635530e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 9.657625e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 9.657625e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 9.135265e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 9.157006e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.157006e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     1.712282 sec
-     4,956,507,680      cycles                           #    2.889 GHz                    
-    12,466,691,438      instructions                     #    2.52  insn per cycle         
-       1.725471807 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:10258) (512y:  240) (512z:    0)
+TOTAL       :     1.804981 sec
+     4,885,090,375      cycles                           #    2.700 GHz                    
+    12,330,030,988      instructions                     #    2.52  insn per cycle         
+       1.821790981 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 9736) (512y:   94) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_d_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -188,26 +188,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_d_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_d_inl0_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.626675e-04
-Avg ME (F77/C++)    = 6.6266731198157320E-004
-Relative difference = 2.837296634927675e-07
+Avg ME (F77/C++)    = 6.6266731198157309E-004
+Relative difference = 2.837296636563793e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_d_inl0_hrd1/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.639925e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.653575e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.653575e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.917661e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.930225e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.930225e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     2.155986 sec
-     4,118,224,744      cycles                           #    1.907 GHz                    
-     6,458,752,156      instructions                     #    1.57  insn per cycle         
-       2.169032093 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1647) (512y:  192) (512z: 9375)
+TOTAL       :     2.381269 sec
+     4,053,625,505      cycles                           #    1.699 GHz                    
+     6,293,972,632      instructions                     #    1.55  insn per cycle         
+       2.398074513 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1497) (512y:   94) (512z: 9019)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_d_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -215,8 +215,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_d_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_d_inl0_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.626675e-04
-Avg ME (F77/C++)    = 6.6266731198157320E-004
-Relative difference = 2.837296634927675e-07
+Avg ME (F77/C++)    = 6.6266731198157309E-004
+Relative difference = 2.837296636563793e-07
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_d_inl1_hrd0.txt b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_d_inl1_hrd0.txt
index 9bedb53e70..466f11943e 100644
--- a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_d_inl1_hrd0.txt
+++ b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_d_inl1_hrd0.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_d_inl1_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
 
-DATE: 2023-11-24_15:02:17
+DATE: 2024-01-30_05:37:36
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl1_hrd0/gcheck.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 3.232352e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.258036e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.260266e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.224805e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.249067e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.252384e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     0.533076 sec
-     2,264,463,020      cycles                           #    2.941 GHz                    
-     3,538,774,343      instructions                     #    1.56  insn per cycle         
-       0.826602256 seconds time elapsed
+TOTAL       :     0.539977 sec
+     2,169,108,553      cycles                           #    2.827 GHz                    
+     3,309,870,321      instructions                     #    1.53  insn per cycle         
+       0.827427226 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl1_hrd0/gcheck.exe -p 64 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 255
@@ -65,17 +65,17 @@ WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 .........................................................................
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl1_hrd0/gcheck.exe -p 2048 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 3.769231e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.796782e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.797908e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.771192e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.799798e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.801021e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 6.665112e+00 +- 5.002651e+00 )  GeV^-4
-TOTAL       :     3.296720 sec
-    10,445,857,222      cycles                           #    2.932 GHz                    
-    21,607,313,168      instructions                     #    2.07  insn per cycle         
-       3.618781692 seconds time elapsed
+TOTAL       :     3.309705 sec
+    10,258,793,873      cycles                           #    2.876 GHz                    
+    23,623,503,831      instructions                     #    2.30  insn per cycle         
+       3.624842760 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl1_hrd0/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl1_hrd0/fgcheck.exe 2 64 2
@@ -86,20 +86,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl1_hrd0/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 4.334345e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.334819e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.334819e+02                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.186471e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.186937e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.186937e+02                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :    37.848381 sec
-   113,421,106,616      cycles                           #    2.997 GHz                    
-   144,959,874,698      instructions                     #    1.28  insn per cycle         
-      37.852852480 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:21301) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :    39.185516 sec
+   112,945,518,025      cycles                           #    2.882 GHz                    
+   141,519,786,794      instructions                     #    1.25  insn per cycle         
+      39.190901211 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:21365) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl1_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -107,26 +107,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl1_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl1_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.626675e-04
-Avg ME (F77/C++)    = 6.6266731198140450E-004
-Relative difference = 2.83729918072716e-07
+Avg ME (F77/C++)    = 6.6266731198140461E-004
+Relative difference = 2.8372991790910424e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_d_inl1_hrd0/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.164523e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.167012e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.167012e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.072790e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.075243e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.075243e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     5.191844 sec
-    14,732,312,446      cycles                           #    2.836 GHz                    
-    37,575,149,913      instructions                     #    2.55  insn per cycle         
-       5.196266852 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:68119) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     5.347436 sec
+    14,950,247,924      cycles                           #    2.794 GHz                    
+    37,533,141,644      instructions                     #    2.51  insn per cycle         
+       5.352716029 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:68052) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_d_inl1_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -134,26 +134,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_d_inl1_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_d_inl1_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.626675e-04
-Avg ME (F77/C++)    = 6.6266731198141209E-004
-Relative difference = 2.8372990661989057e-07
+Avg ME (F77/C++)    = 6.6266731198141220E-004
+Relative difference = 2.837299064562788e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_d_inl1_hrd0/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.486864e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.501204e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.501204e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 7.349404e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.363561e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.363561e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     2.200274 sec
-     6,167,482,001      cycles                           #    2.799 GHz                    
-    13,061,737,270      instructions                     #    2.12  insn per cycle         
-       2.204748277 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:46960) (512y:    0) (512z:    0)
+TOTAL       :     2.242056 sec
+     6,032,020,393      cycles                           #    2.685 GHz                    
+    12,947,712,227      instructions                     #    2.15  insn per cycle         
+       2.247452761 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:46593) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_d_inl1_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -161,26 +161,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_d_inl1_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_d_inl1_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.626675e-04
-Avg ME (F77/C++)    = 6.6266731198156789E-004
-Relative difference = 2.837296715097453e-07
+Avg ME (F77/C++)    = 6.6266731198156778E-004
+Relative difference = 2.837296716733571e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_d_inl1_hrd0/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 9.239505e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 9.261064e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 9.261064e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 8.895381e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 8.916043e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.916043e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     1.784853 sec
-     5,062,751,449      cycles                           #    2.830 GHz                    
-    11,440,329,026      instructions                     #    2.26  insn per cycle         
-       1.789370916 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:40434) (512y:  285) (512z:    0)
+TOTAL       :     1.857617 sec
+     4,999,907,297      cycles                           #    2.689 GHz                    
+    11,364,404,504      instructions                     #    2.27  insn per cycle         
+       1.863061758 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:40158) (512y:  279) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_d_inl1_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -188,26 +188,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_d_inl1_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_d_inl1_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.626675e-04
-Avg ME (F77/C++)    = 6.6266731198156789E-004
-Relative difference = 2.837296715097453e-07
+Avg ME (F77/C++)    = 6.6266731198156778E-004
+Relative difference = 2.837296716733571e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_d_inl1_hrd0/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.593982e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.608272e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.608272e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 7.220172e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.234094e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.234094e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     2.169483 sec
-     4,001,360,531      cycles                           #    1.841 GHz                    
-     5,942,704,084      instructions                     #    1.49  insn per cycle         
-       2.173913867 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2455) (512y:  337) (512z:39411)
+TOTAL       :     2.282224 sec
+     3,899,980,695      cycles                           #    1.706 GHz                    
+     5,854,430,419      instructions                     #    1.50  insn per cycle         
+       2.287473513 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2112) (512y:  142) (512z:39211)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_d_inl1_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
diff --git a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_d_inl1_hrd1.txt b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_d_inl1_hrd1.txt
index 3babb1df02..5156a1b6a3 100644
--- a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_d_inl1_hrd1.txt
+++ b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_d_inl1_hrd1.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_d_inl1_hrd1'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
 
-DATE: 2023-11-24_15:03:26
+DATE: 2024-01-30_05:38:48
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl1_hrd1/gcheck.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 3.228708e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.253927e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.255914e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.248555e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.273233e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.276174e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     0.533504 sec
-     2,263,207,731      cycles                           #    2.940 GHz                    
-     3,466,881,858      instructions                     #    1.53  insn per cycle         
-       0.826915946 seconds time elapsed
+TOTAL       :     0.534739 sec
+     2,168,342,582      cycles                           #    2.838 GHz                    
+     3,393,710,794      instructions                     #    1.57  insn per cycle         
+       0.822006178 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl1_hrd1/gcheck.exe -p 64 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 255
@@ -65,17 +65,17 @@ WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 .........................................................................
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl1_hrd1/gcheck.exe -p 2048 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 3.789640e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.817455e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.818645e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.787191e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.816239e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.817488e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 6.665112e+00 +- 5.002651e+00 )  GeV^-4
-TOTAL       :     3.275326 sec
-    10,593,520,447      cycles                           #    2.998 GHz                    
-    22,594,076,221      instructions                     #    2.13  insn per cycle         
-       3.592953181 seconds time elapsed
+TOTAL       :     3.282278 sec
+    10,172,932,501      cycles                           #    2.876 GHz                    
+    20,641,658,708      instructions                     #    2.03  insn per cycle         
+       3.596374566 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl1_hrd1/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl1_hrd1/fgcheck.exe 2 64 2
@@ -86,20 +86,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl1_hrd1/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 4.301382e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.301861e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.301861e+02                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.152053e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.152498e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.152498e+02                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :    38.148296 sec
-   114,295,626,783      cycles                           #    2.997 GHz                    
-   145,697,398,014      instructions                     #    1.27  insn per cycle         
-      38.152633004 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:22559) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :    39.509903 sec
+   113,989,864,763      cycles                           #    2.886 GHz                    
+   141,709,117,860      instructions                     #    1.24  insn per cycle         
+      39.515181315 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:21615) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl1_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -107,26 +107,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl1_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_d_inl1_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.626675e-04
-Avg ME (F77/C++)    = 6.6266731198140450E-004
-Relative difference = 2.83729918072716e-07
+Avg ME (F77/C++)    = 6.6266731198140461E-004
+Relative difference = 2.8372991790910424e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_d_inl1_hrd1/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.078026e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.080354e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.080354e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.077703e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.080226e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.080226e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     5.337092 sec
-    15,158,837,201      cycles                           #    2.838 GHz                    
-    37,762,768,502      instructions                     #    2.49  insn per cycle         
-       5.341471006 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:68447) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     5.341972 sec
+    14,900,472,017      cycles                           #    2.788 GHz                    
+    37,594,155,695      instructions                     #    2.52  insn per cycle         
+       5.347186768 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:68056) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_d_inl1_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -134,26 +134,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_d_inl1_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_d_inl1_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.626675e-04
-Avg ME (F77/C++)    = 6.6266731198141209E-004
-Relative difference = 2.8372990661989057e-07
+Avg ME (F77/C++)    = 6.6266731198141220E-004
+Relative difference = 2.837299064562788e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_d_inl1_hrd1/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.680771e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.695420e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.695420e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 7.479123e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.493428e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.493428e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     2.144872 sec
-     6,018,410,000      cycles                           #    2.801 GHz                    
-    12,896,129,377      instructions                     #    2.14  insn per cycle         
-       2.149367936 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:45929) (512y:    0) (512z:    0)
+TOTAL       :     2.203003 sec
+     5,937,038,542      cycles                           #    2.690 GHz                    
+    12,831,821,287      instructions                     #    2.16  insn per cycle         
+       2.208347742 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:45663) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_d_inl1_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -161,26 +161,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_d_inl1_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_d_inl1_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.626675e-04
-Avg ME (F77/C++)    = 6.6266731198156789E-004
-Relative difference = 2.837296715097453e-07
+Avg ME (F77/C++)    = 6.6266731198156778E-004
+Relative difference = 2.837296716733571e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_d_inl1_hrd1/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 9.099517e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 9.119694e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 9.119694e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 8.959391e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 8.980227e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.980227e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     1.811911 sec
-     5,092,083,515      cycles                           #    2.805 GHz                    
-    11,446,810,123      instructions                     #    2.25  insn per cycle         
-       1.816340568 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:40123) (512y:  219) (512z:    0)
+TOTAL       :     1.840604 sec
+     4,989,362,539      cycles                           #    2.704 GHz                    
+    11,359,801,014      instructions                     #    2.28  insn per cycle         
+       1.846082122 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:39855) (512y:  212) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_d_inl1_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -188,26 +188,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_d_inl1_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_d_inl1_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.626675e-04
-Avg ME (F77/C++)    = 6.6266731198156789E-004
-Relative difference = 2.837296715097453e-07
+Avg ME (F77/C++)    = 6.6266731198156778E-004
+Relative difference = 2.837296716733571e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_d_inl1_hrd1/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.715726e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.730963e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.730963e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 7.264695e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.278525e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.278525e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     2.135403 sec
-     3,949,503,348      cycles                           #    1.847 GHz                    
-     5,897,146,057      instructions                     #    1.49  insn per cycle         
-       2.139763572 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1971) (512y:  259) (512z:38937)
+TOTAL       :     2.267781 sec
+     3,893,427,498      cycles                           #    1.714 GHz                    
+     5,843,815,532      instructions                     #    1.50  insn per cycle         
+       2.273034135 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1687) (512y:  116) (512z:38946)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_d_inl1_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
diff --git a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_f_inl0_hrd0.txt b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_f_inl0_hrd0.txt
index 9acfc6188d..aecab864cd 100644
--- a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_f_inl0_hrd0.txt
+++ b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_f_inl0_hrd0.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_f_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
 
-DATE: 2023-11-24_14:37:38
+DATE: 2024-01-30_05:02:06
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/gcheck.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 6.349743e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 6.400970e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 6.406599e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.329622e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.381296e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.387810e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 4.059596e+00 +- 2.368053e+00 )  GeV^-4
-TOTAL       :     0.479034 sec
-     2,079,658,422      cycles                           #    2.994 GHz                    
-     3,036,293,369      instructions                     #    1.46  insn per cycle         
-       0.774778080 seconds time elapsed
+TOTAL       :     0.486367 sec
+     1,996,254,093      cycles                           #    2.831 GHz                    
+     2,951,017,935      instructions                     #    1.48  insn per cycle         
+       0.792595596 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/gcheck.exe -p 64 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 255
@@ -65,17 +65,17 @@ WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 .........................................................................
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/gcheck.exe -p 2048 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 8.518158e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 8.591813e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 8.595087e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 8.619469e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 8.695026e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.698446e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 6.664703e+00 +- 5.072736e+00 )  GeV^-4
-TOTAL       :     1.721697 sec
-     5,967,810,947      cycles                           #    3.055 GHz                    
-    12,572,582,180      instructions                     #    2.11  insn per cycle         
-       2.009764828 seconds time elapsed
+TOTAL       :     1.718352 sec
+     5,604,348,056      cycles                           #    2.870 GHz                    
+    11,484,891,091      instructions                     #    2.05  insn per cycle         
+       2.010002941 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/fgcheck.exe 2 64 2
@@ -86,128 +86,128 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.064583e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.065690e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.065690e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.963446e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.964435e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.964435e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.060121e+00 +- 2.367902e+00 )  GeV^-4
-TOTAL       :     7.952623 sec
-    24,634,175,599      cycles                           #    3.097 GHz                    
-    78,128,580,676      instructions                     #    3.17  insn per cycle         
-       7.959267887 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 3603) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     8.362127 sec
+    24,202,873,915      cycles                           #    2.893 GHz                    
+    75,878,244,924      instructions                     #    3.14  insn per cycle         
+       8.372784572 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 3898) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 6.627486e-04
-Avg ME (F77/C++)    = 6.6274863266294753E-004
-Relative difference = 4.92840687132121e-08
+Avg ME (C++/C++)    = 6.627487e-04
+Avg ME (F77/C++)    = 6.6274870439686495E-004
+Relative difference = 6.634286759220428e-09
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_f_inl0_hrd0/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.270538e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.283405e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.283405e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 7.122204e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.135618e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.135618e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.060119e+00 +- 2.367901e+00 )  GeV^-4
-TOTAL       :     2.264433 sec
-     6,463,017,562      cycles                           #    2.850 GHz                    
-    20,121,780,291      instructions                     #    3.11  insn per cycle         
-       2.276060029 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:13763) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     2.311249 sec
+     6,498,315,380      cycles                           #    2.806 GHz                    
+    20,115,878,445      instructions                     #    3.10  insn per cycle         
+       2.327706318 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:13237) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_f_inl0_hrd0/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 6.627486e-04
-Avg ME (F77/C++)    = 6.6274861460025036E-004
-Relative difference = 2.2029847170826283e-08
+Avg ME (C++/C++)    = 6.627485e-04
+Avg ME (F77/C++)    = 6.6274853360924479E-004
+Relative difference = 5.071191384964548e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_f_inl0_hrd0/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.637995e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.644543e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.644543e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.585863e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.592266e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.592266e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 4.060560e+00 +- 2.367611e+00 )  GeV^-4
-TOTAL       :     1.009271 sec
-     2,836,657,749      cycles                           #    2.799 GHz                    
-     6,989,190,074      instructions                     #    2.46  insn per cycle         
-       1.021190194 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:11874) (512y:    0) (512z:    0)
+TOTAL       :     1.042848 sec
+     2,820,748,390      cycles                           #    2.693 GHz                    
+     7,038,277,049      instructions                     #    2.50  insn per cycle         
+       1.060611053 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:11604) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_f_inl0_hrd0/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 6.627194e-04
-Avg ME (F77/C++)    = 6.6271938174574524E-004
-Relative difference = 2.7544470208782633e-08
+Avg ME (C++/C++)    = 6.627193e-04
+Avg ME (F77/C++)    = 6.6271927529261421E-004
+Relative difference = 3.728182620967159e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_f_inl0_hrd0/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.933245e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.942230e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.942230e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.805764e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.814413e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.814413e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 4.060560e+00 +- 2.367611e+00 )  GeV^-4
-TOTAL       :     0.856055 sec
-     2,489,722,611      cycles                           #    2.894 GHz                    
-     6,296,795,301      instructions                     #    2.53  insn per cycle         
-       0.871292854 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:10822) (512y:   43) (512z:    0)
+TOTAL       :     0.916917 sec
+     2,479,527,909      cycles                           #    2.691 GHz                    
+     6,280,728,930      instructions                     #    2.53  insn per cycle         
+       0.937569165 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:10320) (512y:   50) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_f_inl0_hrd0/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 6.627194e-04
-Avg ME (F77/C++)    = 6.6271938174574524E-004
-Relative difference = 2.7544470208782633e-08
+Avg ME (C++/C++)    = 6.627193e-04
+Avg ME (F77/C++)    = 6.6271927529261421E-004
+Relative difference = 3.728182620967159e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_f_inl0_hrd0/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[16] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.556155e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.561846e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.561846e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.395801e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.400853e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.400853e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 4.060562e+00 +- 2.367612e+00 )  GeV^-4
-TOTAL       :     1.062224 sec
-     2,048,930,683      cycles                           #    1.922 GHz                    
-     3,267,038,899      instructions                     #    1.59  insn per cycle         
-       1.073737564 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2414) (512y:   46) (512z: 9571)
+TOTAL       :     1.183787 sec
+     2,037,112,677      cycles                           #    1.714 GHz                    
+     3,249,000,234      instructions                     #    1.59  insn per cycle         
+       1.203517458 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2165) (512y:   48) (512z: 9219)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -215,8 +215,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.627195e-04
-Avg ME (F77/C++)    = 6.6271952779373838E-004
-Relative difference = 4.193891735414155e-08
+Avg ME (F77/C++)    = 6.6271952818273971E-004
+Relative difference = 4.252589469696448e-08
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_f_inl0_hrd0_bridge.txt b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_f_inl0_hrd0_bridge.txt
index fa5a863bc1..cfd5bd9f60 100644
--- a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_f_inl0_hrd0_bridge.txt
+++ b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_f_inl0_hrd0_bridge.txt
@@ -41,7 +41,7 @@ CUDACPP_BUILDDIR='build.512z_f_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
 
-DATE: 2023-11-24_15:13:56
+DATE: 2024-01-30_05:49:44
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
@@ -51,17 +51,17 @@ WARNING! Bridge selected: cannot use RamboDevice, will use RamboHost
 WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
 WARNING! Instantiate device Bridge (nevt=16384, gpublocks=64, gputhreads=256, gpublocks*gputhreads=16384)
 WARNING! Set grid in Bridge (nevt=16384, gpublocks=64, gputhreads=256, gpublocks*gputhreads=16384)
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 5.632667e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 6.322906e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 6.322906e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.575134e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.304295e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.304295e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 4.048178e+00 +- 2.364571e+00 )  GeV^-4
-TOTAL       :     0.467438 sec
-     2,006,072,261      cycles                           #    2.897 GHz                    
-     2,979,194,260      instructions                     #    1.49  insn per cycle         
-       0.749996630 seconds time elapsed
+TOTAL       :     0.470892 sec
+     1,938,590,562      cycles                           #    2.832 GHz                    
+     2,932,139,577      instructions                     #    1.51  insn per cycle         
+       0.742517096 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/gcheck.exe -p 64 256 1 --bridge
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Bridge selected: cannot use RamboDevice, will use RamboHost
@@ -77,17 +77,17 @@ WARNING! Bridge selected: cannot use RamboDevice, will use RamboHost
 WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
 WARNING! Instantiate device Bridge (nevt=524288, gpublocks=2048, gputhreads=256, gpublocks*gputhreads=524288)
 WARNING! Set grid in Bridge (nevt=524288, gpublocks=2048, gputhreads=256, gpublocks*gputhreads=524288)
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 7.176934e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 8.461971e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 8.461971e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 7.189558e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 8.483327e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.483327e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 6.641710e+00 +- 4.994249e+00 )  GeV^-4
-TOTAL       :     1.909592 sec
-     6,340,042,952      cycles                           #    2.952 GHz                    
-    13,540,665,643      instructions                     #    2.14  insn per cycle         
-       2.204543754 seconds time elapsed
+TOTAL       :     1.911946 sec
+     6,179,624,048      cycles                           #    2.874 GHz                    
+    12,701,880,125      instructions                     #    2.06  insn per cycle         
+       2.209063416 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/fgcheck.exe 2 64 2
@@ -99,132 +99,132 @@ OK (relative difference <= 5E-3)
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/check.exe -p 64 256 1 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=16384)
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.018583e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.019594e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.019594e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.966573e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.967552e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.967552e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.060121e+00 +- 2.367902e+00 )  GeV^-4
-TOTAL       :     8.135657 sec
-    24,647,759,447      cycles                           #    3.029 GHz                    
-    78,134,979,688      instructions                     #    3.17  insn per cycle         
-       8.140016435 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 3603) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     8.353267 sec
+    24,210,307,332      cycles                           #    2.898 GHz                    
+    75,882,231,103      instructions                     #    3.13  insn per cycle         
+       8.358202878 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 3898) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 6.627486e-04
-Avg ME (F77/C++)    = 6.6274863266294753E-004
-Relative difference = 4.92840687132121e-08
+Avg ME (C++/C++)    = 6.627487e-04
+Avg ME (F77/C++)    = 6.6274870439686495E-004
+Relative difference = 6.634286759220428e-09
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_f_inl0_hrd0/check.exe -p 64 256 1 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=16384)
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/sse4+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.102824e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.116231e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.116231e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 7.010932e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.023878e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.023878e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.060119e+00 +- 2.367901e+00 )  GeV^-4
-TOTAL       :     2.319512 sec
-     6,477,527,484      cycles                           #    2.794 GHz                    
-    20,133,025,469      instructions                     #    3.11  insn per cycle         
-       2.323961953 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:13763) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     2.350780 sec
+     6,507,988,967      cycles                           #    2.764 GHz                    
+    20,124,211,431      instructions                     #    3.09  insn per cycle         
+       2.355993372 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:13237) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_f_inl0_hrd0/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 6.627486e-04
-Avg ME (F77/C++)    = 6.6274861460025036E-004
-Relative difference = 2.2029847170826283e-08
+Avg ME (C++/C++)    = 6.627485e-04
+Avg ME (F77/C++)    = 6.6274853360924479E-004
+Relative difference = 5.071191384964548e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_f_inl0_hrd0/check.exe -p 64 256 1 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=16384)
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/avx2+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.662031e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.669314e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.669314e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.585110e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.591932e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.591932e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 4.060560e+00 +- 2.367611e+00 )  GeV^-4
-TOTAL       :     0.997479 sec
-     2,842,471,628      cycles                           #    2.839 GHz                    
-     6,997,876,677      instructions                     #    2.46  insn per cycle         
-       1.001720557 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:11874) (512y:    0) (512z:    0)
+TOTAL       :     1.046222 sec
+     2,830,060,229      cycles                           #    2.694 GHz                    
+     7,047,238,365      instructions                     #    2.49  insn per cycle         
+       1.051506977 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:11604) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_f_inl0_hrd0/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 6.627194e-04
-Avg ME (F77/C++)    = 6.6271938174574524E-004
-Relative difference = 2.7544470208782633e-08
+Avg ME (C++/C++)    = 6.627193e-04
+Avg ME (F77/C++)    = 6.6271927529261421E-004
+Relative difference = 3.728182620967159e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_f_inl0_hrd0/check.exe -p 64 256 1 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=16384)
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/512y+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.899662e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.908650e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.908650e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.805765e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.814390e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.814390e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 4.060560e+00 +- 2.367611e+00 )  GeV^-4
-TOTAL       :     0.874017 sec
-     2,497,188,263      cycles                           #    2.845 GHz                    
-     6,305,318,806      instructions                     #    2.52  insn per cycle         
-       0.878481296 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:10822) (512y:   43) (512z:    0)
+TOTAL       :     0.919776 sec
+     2,488,595,721      cycles                           #    2.693 GHz                    
+     6,289,461,030      instructions                     #    2.53  insn per cycle         
+       0.925186931 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:10320) (512y:   50) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_f_inl0_hrd0/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 6.627194e-04
-Avg ME (F77/C++)    = 6.6271938174574524E-004
-Relative difference = 2.7544470208782633e-08
+Avg ME (C++/C++)    = 6.627193e-04
+Avg ME (F77/C++)    = 6.6271927529261421E-004
+Relative difference = 3.728182620967159e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_f_inl0_hrd0/check.exe -p 64 256 1 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=16384)
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/512z+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[16] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.510965e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.516631e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.516631e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.390787e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.395884e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.395884e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 4.060562e+00 +- 2.367612e+00 )  GeV^-4
-TOTAL       :     1.096117 sec
-     2,057,326,282      cycles                           #    1.871 GHz                    
-     3,276,156,329      instructions                     #    1.59  insn per cycle         
-       1.100460049 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2414) (512y:   46) (512z: 9571)
+TOTAL       :     1.191044 sec
+     2,045,888,825      cycles                           #    1.712 GHz                    
+     3,258,286,239      instructions                     #    1.59  insn per cycle         
+       1.196330024 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2165) (512y:   48) (512z: 9219)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -232,8 +232,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.627195e-04
-Avg ME (F77/C++)    = 6.6271952779373838E-004
-Relative difference = 4.193891735414155e-08
+Avg ME (F77/C++)    = 6.6271952818273971E-004
+Relative difference = 4.252589469696448e-08
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_f_inl0_hrd0_common.txt b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_f_inl0_hrd0_common.txt
index cde090527d..18818d76f2 100644
--- a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_f_inl0_hrd0_common.txt
+++ b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_f_inl0_hrd0_common.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_f_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
 
-DATE: 2023-11-24_15:25:47
+DATE: 2024-01-30_06:02:02
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/gcheck.exe -p 64 256 1 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:COMMON+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 6.341986e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 6.393716e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 6.399601e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.319163e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.372298e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.378244e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 4.159397e-01 +- 3.238804e-01 )  GeV^-4
-TOTAL       :     0.465286 sec
-     1,996,862,139      cycles                           #    2.938 GHz                    
-     2,998,262,282      instructions                     #    1.50  insn per cycle         
-       0.737241303 seconds time elapsed
+TOTAL       :     0.470285 sec
+     1,953,187,910      cycles                           #    2.826 GHz                    
+     2,879,626,230      instructions                     #    1.47  insn per cycle         
+       0.750725256 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/gcheck.exe -p 64 256 1 --common
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 255
@@ -65,17 +65,17 @@ WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 .........................................................................
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/gcheck.exe -p 2048 256 1 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:COMMON+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 8.562727e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 8.635286e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 8.638678e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 8.571852e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 8.645768e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.649137e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 1.094367e+02 +- 1.071509e+02 )  GeV^-4
-TOTAL       :     1.800198 sec
-     6,027,791,673      cycles                           #    2.972 GHz                    
-    12,131,730,184      instructions                     #    2.01  insn per cycle         
-       2.091221198 seconds time elapsed
+TOTAL       :     1.807519 sec
+     5,850,952,354      cycles                           #    2.861 GHz                    
+    11,909,032,858      instructions                     #    2.04  insn per cycle         
+       2.113707665 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/fgcheck.exe 2 64 2
@@ -86,128 +86,128 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/check.exe -p 64 256 1 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:COMMON+RMBHST+MESHST/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.988613e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.989604e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.989604e+03                 )  sec^-1
-MeanMatrixElemValue         = ( 4.208459e-01 +- 3.253446e-01 )  GeV^-4
-TOTAL       :     8.257430 sec
-    24,655,829,370      cycles                           #    2.985 GHz                    
-    78,128,720,285      instructions                     #    3.17  insn per cycle         
-       8.261783085 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 3603) (avx2:    0) (512y:    0) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 1.964186e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.965186e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.965186e+03                 )  sec^-1
+MeanMatrixElemValue         = ( 4.208458e-01 +- 3.253446e-01 )  GeV^-4
+TOTAL       :     8.362848 sec
+    24,219,340,843      cycles                           #    2.896 GHz                    
+    75,878,803,024      instructions                     #    3.13  insn per cycle         
+       8.367752014 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 3898) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 6.627486e-04
-Avg ME (F77/C++)    = 6.6274863266294753E-004
-Relative difference = 4.92840687132121e-08
+Avg ME (C++/C++)    = 6.627487e-04
+Avg ME (F77/C++)    = 6.6274870439686495E-004
+Relative difference = 6.634286759220428e-09
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_f_inl0_hrd0/check.exe -p 64 256 1 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:COMMON+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.090825e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.103638e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.103638e+03                 )  sec^-1
-MeanMatrixElemValue         = ( 4.208457e-01 +- 3.253445e-01 )  GeV^-4
-TOTAL       :     2.321920 sec
-     6,463,785,266      cycles                           #    2.780 GHz                    
-    20,119,008,709      instructions                     #    3.11  insn per cycle         
-       2.326112409 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:13763) (avx2:    0) (512y:    0) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 7.106063e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.119817e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.119817e+03                 )  sec^-1
+MeanMatrixElemValue         = ( 4.208458e-01 +- 3.253446e-01 )  GeV^-4
+TOTAL       :     2.317534 sec
+     6,502,161,706      cycles                           #    2.801 GHz                    
+    20,113,148,136      instructions                     #    3.09  insn per cycle         
+       2.322610994 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:13237) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_f_inl0_hrd0/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 6.627486e-04
-Avg ME (F77/C++)    = 6.6274861460025036E-004
-Relative difference = 2.2029847170826283e-08
+Avg ME (C++/C++)    = 6.627485e-04
+Avg ME (F77/C++)    = 6.6274853360924479E-004
+Relative difference = 5.071191384964548e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_f_inl0_hrd0/check.exe -p 64 256 1 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:COMMON+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.643399e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.650724e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.650724e+04                 )  sec^-1
-MeanMatrixElemValue         = ( 4.214978e-01 +- 3.255521e-01 )  GeV^-4
-TOTAL       :     1.007824 sec
-     2,846,106,580      cycles                           #    2.814 GHz                    
-     6,988,034,221      instructions                     #    2.46  insn per cycle         
-       1.012119516 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:11874) (512y:    0) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 1.586948e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.593562e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.593562e+04                 )  sec^-1
+MeanMatrixElemValue         = ( 4.214979e-01 +- 3.255522e-01 )  GeV^-4
+TOTAL       :     1.043186 sec
+     2,822,730,977      cycles                           #    2.696 GHz                    
+     7,035,059,102      instructions                     #    2.49  insn per cycle         
+       1.048122577 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:11604) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_f_inl0_hrd0/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 6.627194e-04
-Avg ME (F77/C++)    = 6.6271938174574524E-004
-Relative difference = 2.7544470208782633e-08
+Avg ME (C++/C++)    = 6.627193e-04
+Avg ME (F77/C++)    = 6.6271927529261421E-004
+Relative difference = 3.728182620967159e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_f_inl0_hrd0/check.exe -p 64 256 1 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:COMMON+RMBHST+MESHST/512y+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.842696e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.851591e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.851591e+04                 )  sec^-1
-MeanMatrixElemValue         = ( 4.214978e-01 +- 3.255521e-01 )  GeV^-4
-TOTAL       :     0.900263 sec
-     2,494,778,530      cycles                           #    2.760 GHz                    
-     6,293,767,609      instructions                     #    2.52  insn per cycle         
-       0.904816892 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:10822) (512y:   43) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 1.807119e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.816011e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.816011e+04                 )  sec^-1
+MeanMatrixElemValue         = ( 4.214979e-01 +- 3.255522e-01 )  GeV^-4
+TOTAL       :     0.917444 sec
+     2,481,419,746      cycles                           #    2.693 GHz                    
+     6,275,834,953      instructions                     #    2.53  insn per cycle         
+       0.922842065 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:10320) (512y:   50) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_f_inl0_hrd0/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 6.627194e-04
-Avg ME (F77/C++)    = 6.6271938174574524E-004
-Relative difference = 2.7544470208782633e-08
+Avg ME (C++/C++)    = 6.627193e-04
+Avg ME (F77/C++)    = 6.6271927529261421E-004
+Relative difference = 3.728182620967159e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_f_inl0_hrd0/check.exe -p 64 256 1 --common OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:COMMON+RMBHST+MESHST/512z+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[16] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.422519e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.427987e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.427987e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.399447e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.404609e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.404609e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 4.214981e-01 +- 3.255523e-01 )  GeV^-4
-TOTAL       :     1.162978 sec
-     2,054,015,790      cycles                           #    1.761 GHz                    
-     3,264,160,600      instructions                     #    1.59  insn per cycle         
-       1.167204725 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2414) (512y:   46) (512z: 9571)
+TOTAL       :     1.182503 sec
+     2,042,245,375      cycles                           #    1.722 GHz                    
+     3,246,419,225      instructions                     #    1.59  insn per cycle         
+       1.187753193 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2165) (512y:   48) (512z: 9219)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -215,8 +215,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.627195e-04
-Avg ME (F77/C++)    = 6.6271952779373838E-004
-Relative difference = 4.193891735414155e-08
+Avg ME (F77/C++)    = 6.6271952818273971E-004
+Relative difference = 4.252589469696448e-08
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_f_inl0_hrd0_curhst.txt b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_f_inl0_hrd0_curhst.txt
index 3a3eb3caf7..e0bdb664e1 100644
--- a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_f_inl0_hrd0_curhst.txt
+++ b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_f_inl0_hrd0_curhst.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_f_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
 
-DATE: 2023-11-24_15:22:26
+DATE: 2024-01-30_05:58:32
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/gcheck.exe -p 64 256 1 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 6.336723e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 6.390423e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 6.395725e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.316613e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.368482e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.375052e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 4.059596e+00 +- 2.368053e+00 )  GeV^-4
-TOTAL       :     0.463745 sec
-     1,978,530,916      cycles                           #    2.937 GHz                    
-     2,900,092,077      instructions                     #    1.47  insn per cycle         
-       0.732414298 seconds time elapsed
+TOTAL       :     0.466730 sec
+     1,919,349,851      cycles                           #    2.829 GHz                    
+     2,893,848,641      instructions                     #    1.51  insn per cycle         
+       0.736177730 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/gcheck.exe -p 64 256 1 --curhst
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 255
@@ -65,17 +65,17 @@ WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 .........................................................................
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/gcheck.exe -p 2048 256 1 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 8.551039e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 8.623038e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 8.626383e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 8.573214e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 8.646713e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.650131e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 6.664703e+00 +- 5.072736e+00 )  GeV^-4
-TOTAL       :     1.755684 sec
-     5,914,588,617      cycles                           #    2.977 GHz                    
-    12,946,533,243      instructions                     #    2.19  insn per cycle         
-       2.043185741 seconds time elapsed
+TOTAL       :     1.756106 sec
+     5,695,142,145      cycles                           #    2.868 GHz                    
+    11,326,470,226      instructions                     #    1.99  insn per cycle         
+       2.046387591 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/fgcheck.exe 2 64 2
@@ -86,128 +86,128 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/check.exe -p 64 256 1 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.015505e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.016502e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.016502e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.965319e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.966317e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.966317e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.060121e+00 +- 2.367902e+00 )  GeV^-4
-TOTAL       :     8.145261 sec
-    24,654,200,089      cycles                           #    3.026 GHz                    
-    78,126,847,348      instructions                     #    3.17  insn per cycle         
-       8.149417178 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 3603) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     8.353416 sec
+    24,206,918,909      cycles                           #    2.897 GHz                    
+    75,878,282,077      instructions                     #    3.13  insn per cycle         
+       8.358253425 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 3898) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 6.627486e-04
-Avg ME (F77/C++)    = 6.6274863266294753E-004
-Relative difference = 4.92840687132121e-08
+Avg ME (C++/C++)    = 6.627487e-04
+Avg ME (F77/C++)    = 6.6274870439686495E-004
+Relative difference = 6.634286759220428e-09
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_f_inl0_hrd0/check.exe -p 64 256 1 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.056381e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.069788e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.069788e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.994720e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.007761e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.007761e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.060119e+00 +- 2.367901e+00 )  GeV^-4
-TOTAL       :     2.332327 sec
-     6,459,567,003      cycles                           #    2.766 GHz                    
-    20,120,739,754      instructions                     #    3.11  insn per cycle         
-       2.336953548 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:13763) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     2.353332 sec
+     6,524,875,303      cycles                           #    2.768 GHz                    
+    20,114,868,262      instructions                     #    3.08  insn per cycle         
+       2.358279130 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:13237) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_f_inl0_hrd0/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 6.627486e-04
-Avg ME (F77/C++)    = 6.6274861460025036E-004
-Relative difference = 2.2029847170826283e-08
+Avg ME (C++/C++)    = 6.627485e-04
+Avg ME (F77/C++)    = 6.6274853360924479E-004
+Relative difference = 5.071191384964548e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_f_inl0_hrd0/check.exe -p 64 256 1 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.634359e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.641184e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.641184e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.578556e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.585147e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.585147e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 4.060560e+00 +- 2.367611e+00 )  GeV^-4
-TOTAL       :     1.011258 sec
-     2,838,489,472      cycles                           #    2.797 GHz                    
-     6,988,461,226      instructions                     #    2.46  insn per cycle         
-       1.015381469 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:11874) (512y:    0) (512z:    0)
+TOTAL       :     1.047733 sec
+     2,820,818,870      cycles                           #    2.682 GHz                    
+     7,037,506,961      instructions                     #    2.49  insn per cycle         
+       1.053002937 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:11604) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_f_inl0_hrd0/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 6.627194e-04
-Avg ME (F77/C++)    = 6.6271938174574524E-004
-Relative difference = 2.7544470208782633e-08
+Avg ME (C++/C++)    = 6.627193e-04
+Avg ME (F77/C++)    = 6.6271927529261421E-004
+Relative difference = 3.728182620967159e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_f_inl0_hrd0/check.exe -p 64 256 1 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.722994e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.730625e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.730625e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.765542e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.773827e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.773827e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 4.060560e+00 +- 2.367611e+00 )  GeV^-4
-TOTAL       :     0.960124 sec
-     2,489,397,951      cycles                           #    2.584 GHz                    
-     6,296,004,783      instructions                     #    2.53  insn per cycle         
-       0.964386434 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:10822) (512y:   43) (512z:    0)
+TOTAL       :     0.937560 sec
+     2,478,872,591      cycles                           #    2.633 GHz                    
+     6,279,446,291      instructions                     #    2.53  insn per cycle         
+       0.942558881 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:10320) (512y:   50) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_f_inl0_hrd0/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 6.627194e-04
-Avg ME (F77/C++)    = 6.6271938174574524E-004
-Relative difference = 2.7544470208782633e-08
+Avg ME (C++/C++)    = 6.627193e-04
+Avg ME (F77/C++)    = 6.6271927529261421E-004
+Relative difference = 3.728182620967159e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_f_inl0_hrd0/check.exe -p 64 256 1 --curhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[16] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.483765e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.489132e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.489132e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.394421e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.399630e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.399630e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 4.060562e+00 +- 2.367612e+00 )  GeV^-4
-TOTAL       :     1.113558 sec
-     2,047,316,573      cycles                           #    1.833 GHz                    
-     3,265,815,382      instructions                     #    1.60  insn per cycle         
-       1.117668257 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2414) (512y:   46) (512z: 9571)
+TOTAL       :     1.184942 sec
+     2,037,351,256      cycles                           #    1.714 GHz                    
+     3,247,924,134      instructions                     #    1.59  insn per cycle         
+       1.189828303 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2165) (512y:   48) (512z: 9219)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -215,8 +215,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.627195e-04
-Avg ME (F77/C++)    = 6.6271952779373838E-004
-Relative difference = 4.193891735414155e-08
+Avg ME (F77/C++)    = 6.6271952818273971E-004
+Relative difference = 4.252589469696448e-08
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_f_inl0_hrd0_rmbhst.txt b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_f_inl0_hrd0_rmbhst.txt
index 474faabf1c..d4941d3986 100644
--- a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_f_inl0_hrd0_rmbhst.txt
+++ b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_f_inl0_hrd0_rmbhst.txt
@@ -41,24 +41,24 @@ CUDACPP_BUILDDIR='build.512z_f_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
 
-DATE: 2023-11-24_15:19:08
+DATE: 2024-01-30_05:55:07
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/gcheck.exe -p 64 256 1 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 5.719331e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 6.371583e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 6.377724e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.730552e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.395791e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.401561e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 4.048178e+00 +- 2.364571e+00 )  GeV^-4
-TOTAL       :     0.465990 sec
-     1,978,008,980      cycles                           #    2.919 GHz                    
-     2,922,355,478      instructions                     #    1.48  insn per cycle         
-       0.734927742 seconds time elapsed
+TOTAL       :     0.472770 sec
+     1,942,039,449      cycles                           #    2.839 GHz                    
+     2,914,569,721      instructions                     #    1.50  insn per cycle         
+       0.744083634 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/gcheck.exe -p 64 256 1 --rmbhst
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
@@ -68,17 +68,17 @@ WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/gcheck.exe -p 2048 256 1 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 7.466647e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 8.638185e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 8.641583e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 7.426812e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 8.621213e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.624728e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 6.641710e+00 +- 4.994249e+00 )  GeV^-4
-TOTAL       :     1.827648 sec
-     6,079,488,382      cycles                           #    2.960 GHz                    
-    11,657,968,792      instructions                     #    1.92  insn per cycle         
-       2.117399258 seconds time elapsed
+TOTAL       :     1.841459 sec
+     5,951,272,102      cycles                           #    2.874 GHz                    
+    12,317,260,326      instructions                     #    2.07  insn per cycle         
+       2.133121829 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/fgcheck.exe 2 64 2
@@ -89,128 +89,128 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/check.exe -p 64 256 1 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.995150e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.996157e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.996157e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.960613e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.961562e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.961562e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.060121e+00 +- 2.367902e+00 )  GeV^-4
-TOTAL       :     8.229536 sec
-    24,636,759,314      cycles                           #    2.993 GHz                    
-    78,130,437,860      instructions                     #    3.17  insn per cycle         
-       8.233776212 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 3603) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     8.374119 sec
+    24,216,955,817      cycles                           #    2.891 GHz                    
+    75,878,033,044      instructions                     #    3.13  insn per cycle         
+       8.378947710 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 3898) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd0/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 6.627486e-04
-Avg ME (F77/C++)    = 6.6274863266294753E-004
-Relative difference = 4.92840687132121e-08
+Avg ME (C++/C++)    = 6.627487e-04
+Avg ME (F77/C++)    = 6.6274870439686495E-004
+Relative difference = 6.634286759220428e-09
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_f_inl0_hrd0/check.exe -p 64 256 1 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.265246e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.278746e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.278746e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 7.136107e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.149132e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.149132e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.060119e+00 +- 2.367901e+00 )  GeV^-4
-TOTAL       :     2.265483 sec
-     6,468,059,761      cycles                           #    2.851 GHz                    
-    20,120,655,203      instructions                     #    3.11  insn per cycle         
-       2.269729053 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:13763) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     2.306790 sec
+     6,504,696,579      cycles                           #    2.815 GHz                    
+    20,114,676,918      instructions                     #    3.09  insn per cycle         
+       2.311724672 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:13237) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_f_inl0_hrd0/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 6.627486e-04
-Avg ME (F77/C++)    = 6.6274861460025036E-004
-Relative difference = 2.2029847170826283e-08
+Avg ME (C++/C++)    = 6.627485e-04
+Avg ME (F77/C++)    = 6.6274853360924479E-004
+Relative difference = 5.071191384964548e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_f_inl0_hrd0/check.exe -p 64 256 1 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.626140e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.632839e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.632839e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.585387e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.592052e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.592052e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 4.060560e+00 +- 2.367611e+00 )  GeV^-4
-TOTAL       :     1.016729 sec
-     2,837,492,476      cycles                           #    2.781 GHz                    
-     6,988,268,106      instructions                     #    2.46  insn per cycle         
-       1.020920647 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:11874) (512y:    0) (512z:    0)
+TOTAL       :     1.043340 sec
+     2,821,286,489      cycles                           #    2.694 GHz                    
+     7,037,435,358      instructions                     #    2.49  insn per cycle         
+       1.048505999 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:11604) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_f_inl0_hrd0/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 6.627194e-04
-Avg ME (F77/C++)    = 6.6271938174574524E-004
-Relative difference = 2.7544470208782633e-08
+Avg ME (C++/C++)    = 6.627193e-04
+Avg ME (F77/C++)    = 6.6271927529261421E-004
+Relative difference = 3.728182620967159e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_f_inl0_hrd0/check.exe -p 64 256 1 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.871366e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.880395e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.880395e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.743919e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.751789e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.751789e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 4.060560e+00 +- 2.367611e+00 )  GeV^-4
-TOTAL       :     0.884395 sec
-     2,487,612,796      cycles                           #    2.801 GHz                    
-     6,295,453,513      instructions                     #    2.53  insn per cycle         
-       0.888549870 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:10822) (512y:   43) (512z:    0)
+TOTAL       :     0.949190 sec
+     2,568,265,414      cycles                           #    2.694 GHz                    
+     6,279,620,229      instructions                     #    2.45  insn per cycle         
+       0.954345697 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:10320) (512y:   50) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_f_inl0_hrd0/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 6.627194e-04
-Avg ME (F77/C++)    = 6.6271938174574524E-004
-Relative difference = 2.7544470208782633e-08
+Avg ME (C++/C++)    = 6.627193e-04
+Avg ME (F77/C++)    = 6.6271927529261421E-004
+Relative difference = 3.728182620967159e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_f_inl0_hrd0/check.exe -p 64 256 1 --rmbhst OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[16] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.476534e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.482059e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.482059e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.404393e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.409463e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.409463e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 4.060562e+00 +- 2.367612e+00 )  GeV^-4
-TOTAL       :     1.119059 sec
-     2,048,523,522      cycles                           #    1.825 GHz                    
-     3,265,858,129      instructions                     #    1.59  insn per cycle         
-       1.123246679 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2414) (512y:   46) (512z: 9571)
+TOTAL       :     1.176805 sec
+     2,037,562,738      cycles                           #    1.726 GHz                    
+     3,247,895,210      instructions                     #    1.59  insn per cycle         
+       1.182054069 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2165) (512y:   48) (512z: 9219)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -218,8 +218,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.627195e-04
-Avg ME (F77/C++)    = 6.6271952779373838E-004
-Relative difference = 4.193891735414155e-08
+Avg ME (F77/C++)    = 6.6271952818273971E-004
+Relative difference = 4.252589469696448e-08
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_f_inl0_hrd1.txt b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_f_inl0_hrd1.txt
index 8db08c40cb..391ab3d24f 100644
--- a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_f_inl0_hrd1.txt
+++ b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_f_inl0_hrd1.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_f_inl0_hrd1'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
 
-DATE: 2023-11-24_14:38:07
+DATE: 2024-01-30_05:02:36
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd1/gcheck.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 6.342222e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 6.395573e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 6.401305e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.280133e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.331305e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.337921e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 4.059596e+00 +- 2.368053e+00 )  GeV^-4
-TOTAL       :     0.475377 sec
-     2,071,036,372      cycles                           #    2.995 GHz                    
-     3,032,826,911      instructions                     #    1.46  insn per cycle         
-       0.762514427 seconds time elapsed
+TOTAL       :     0.487324 sec
+     1,984,658,948      cycles                           #    2.819 GHz                    
+     2,919,152,547      instructions                     #    1.47  insn per cycle         
+       0.799038148 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd1/gcheck.exe -p 64 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 255
@@ -65,17 +65,17 @@ WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 .........................................................................
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd1/gcheck.exe -p 2048 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 8.503453e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 8.577079e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 8.580179e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 8.572518e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 8.647175e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.650566e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 6.664703e+00 +- 5.072736e+00 )  GeV^-4
-TOTAL       :     1.721929 sec
-     5,978,281,541      cycles                           #    3.061 GHz                    
-    12,363,091,331      instructions                     #    2.07  insn per cycle         
-       2.009961043 seconds time elapsed
+TOTAL       :     1.731714 sec
+     5,664,416,860      cycles                           #    2.869 GHz                    
+    11,423,818,247      instructions                     #    2.02  insn per cycle         
+       2.033192051 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd1/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd1/fgcheck.exe 2 64 2
@@ -86,20 +86,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd1/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.071210e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.072246e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.072246e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.928583e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.929543e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.929543e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.060121e+00 +- 2.367902e+00 )  GeV^-4
-TOTAL       :     7.926211 sec
-    24,590,863,046      cycles                           #    3.102 GHz                    
-    77,854,642,024      instructions                     #    3.17  insn per cycle         
-       7.932789655 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 3114) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     8.512607 sec
+    24,191,141,745      cycles                           #    2.843 GHz                    
+    75,807,282,467      instructions                     #    3.13  insn per cycle         
+       8.524714483 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 3848) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -107,107 +107,107 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl0_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.627487e-04
-Avg ME (F77/C++)    = 6.6274866268634797E-004
-Relative difference = 5.630135835748959e-08
+Avg ME (F77/C++)    = 6.6274870430095556E-004
+Relative difference = 6.489572191632735e-09
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_f_inl0_hrd1/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.597544e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.611358e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.611358e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 7.113368e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.126874e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.126874e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.060119e+00 +- 2.367901e+00 )  GeV^-4
-TOTAL       :     2.166211 sec
-     6,421,975,036      cycles                           #    2.959 GHz                    
-    20,086,123,706      instructions                     #    3.13  insn per cycle         
-       2.181924706 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:13452) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     2.313934 sec
+     6,500,918,155      cycles                           #    2.804 GHz                    
+    20,111,364,543      instructions                     #    3.09  insn per cycle         
+       2.332783497 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:13231) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_f_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_f_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_f_inl0_hrd1/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 6.627486e-04
-Avg ME (F77/C++)    = 6.6274861465384638E-004
-Relative difference = 2.211071647257023e-08
+Avg ME (C++/C++)    = 6.627485e-04
+Avg ME (F77/C++)    = 6.6274853360924479E-004
+Relative difference = 5.071191384964548e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_f_inl0_hrd1/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.634914e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.641813e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.641813e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.589760e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.596530e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.596530e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 4.060560e+00 +- 2.367611e+00 )  GeV^-4
-TOTAL       :     1.012157 sec
-     2,920,524,129      cycles                           #    2.875 GHz                    
-     7,130,900,907      instructions                     #    2.44  insn per cycle         
-       1.024052491 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:12261) (512y:    0) (512z:    0)
+TOTAL       :     1.040223 sec
+     2,815,442,217      cycles                           #    2.695 GHz                    
+     7,038,519,370      instructions                     #    2.50  insn per cycle         
+       1.057514134 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:11587) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_f_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_f_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_f_inl0_hrd1/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 6.627194e-04
-Avg ME (F77/C++)    = 6.6271939668077068E-004
-Relative difference = 5.008498817890231e-09
+Avg ME (C++/C++)    = 6.627193e-04
+Avg ME (F77/C++)    = 6.6271927529261421E-004
+Relative difference = 3.728182620967159e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_f_inl0_hrd1/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.776439e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.783925e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.783925e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.751311e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.759469e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.759469e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 4.060560e+00 +- 2.367611e+00 )  GeV^-4
-TOTAL       :     0.931369 sec
-     2,592,272,120      cycles                           #    2.772 GHz                    
-     6,439,655,252      instructions                     #    2.48  insn per cycle         
-       0.940797030 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:11276) (512y:   27) (512z:    0)
+TOTAL       :     0.945030 sec
+     2,478,506,957      cycles                           #    2.610 GHz                    
+     6,280,796,881      instructions                     #    2.53  insn per cycle         
+       0.988336476 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:10302) (512y:   50) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_f_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_f_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_f_inl0_hrd1/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 6.627194e-04
-Avg ME (F77/C++)    = 6.6271939668077068E-004
-Relative difference = 5.008498817890231e-09
+Avg ME (C++/C++)    = 6.627193e-04
+Avg ME (F77/C++)    = 6.6271927529261421E-004
+Relative difference = 3.728182620967159e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_f_inl0_hrd1/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[16] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.511530e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.517127e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.517127e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.386273e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.391271e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.391271e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 4.060562e+00 +- 2.367612e+00 )  GeV^-4
-TOTAL       :     1.092797 sec
-     2,119,842,752      cycles                           #    1.933 GHz                    
-     3,428,582,326      instructions                     #    1.62  insn per cycle         
-       1.103781056 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2911) (512y:   22) (512z: 9647)
+TOTAL       :     1.191730 sec
+     2,039,311,665      cycles                           #    1.704 GHz                    
+     3,248,072,614      instructions                     #    1.59  insn per cycle         
+       1.208300824 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2140) (512y:   48) (512z: 9219)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_f_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -215,8 +215,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_f_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_f_inl0_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.627195e-04
-Avg ME (F77/C++)    = 6.6271952032322112E-004
-Relative difference = 3.066639970473621e-08
+Avg ME (F77/C++)    = 6.6271952818273971E-004
+Relative difference = 4.252589469696448e-08
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_f_inl1_hrd0.txt b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_f_inl1_hrd0.txt
index 7e38ef2c7b..77eae3ae9c 100644
--- a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_f_inl1_hrd0.txt
+++ b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_f_inl1_hrd0.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_f_inl1_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
 
-DATE: 2023-11-24_15:04:36
+DATE: 2024-01-30_05:40:00
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl1_hrd0/gcheck.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 5.586198e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 5.628840e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 5.633485e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.547321e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 5.587547e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.593359e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 4.059596e+00 +- 2.368053e+00 )  GeV^-4
-TOTAL       :     0.485828 sec
-     2,082,184,633      cycles                           #    2.938 GHz                    
-     3,131,974,897      instructions                     #    1.50  insn per cycle         
-       0.768933691 seconds time elapsed
+TOTAL       :     0.493771 sec
+     2,067,778,848      cycles                           #    2.808 GHz                    
+     3,079,367,454      instructions                     #    1.49  insn per cycle         
+       0.793803934 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl1_hrd0/gcheck.exe -p 64 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 255
@@ -65,17 +65,17 @@ WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 .........................................................................
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl1_hrd0/gcheck.exe -p 2048 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 7.739364e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.800371e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.802973e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 7.730139e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.790957e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.793762e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 6.664703e+00 +- 5.072736e+00 )  GeV^-4
-TOTAL       :     1.856072 sec
-     6,264,858,693      cycles                           #    2.996 GHz                    
-    12,861,475,785      instructions                     #    2.05  insn per cycle         
-       2.150989164 seconds time elapsed
+TOTAL       :     1.861027 sec
+     6,035,258,548      cycles                           #    2.873 GHz                    
+    13,088,532,370      instructions                     #    2.17  insn per cycle         
+       2.157681364 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl1_hrd0/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl1_hrd0/fgcheck.exe 2 64 2
@@ -86,47 +86,47 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl1_hrd0/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 5.664072e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 5.664878e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 5.664878e+02                 )  sec^-1
-MeanMatrixElemValue         = ( 4.059969e+00 +- 2.367799e+00 )  GeV^-4
-TOTAL       :    28.963182 sec
-    87,150,309,684      cycles                           #    3.009 GHz                    
-   135,629,504,187      instructions                     #    1.56  insn per cycle         
-      28.967350767 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:15563) (avx2:    0) (512y:    0) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 5.418156e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 5.418889e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.418889e+02                 )  sec^-1
+MeanMatrixElemValue         = ( 4.059968e+00 +- 2.367799e+00 )  GeV^-4
+TOTAL       :    30.278912 sec
+    87,193,893,967      cycles                           #    2.880 GHz                    
+   133,999,567,781      instructions                     #    1.54  insn per cycle         
+      30.284052553 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:16123) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl1_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl1_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl1_hrd0/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 6.627534e-04
-Avg ME (F77/C++)    = 6.6275340277317796E-004
-Relative difference = 4.184328521943034e-09
+Avg ME (C++/C++)    = 6.627535e-04
+Avg ME (F77/C++)    = 6.6275354356437610E-004
+Relative difference = 6.573239683366044e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_f_inl1_hrd0/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.045716e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.058481e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.058481e+03                 )  sec^-1
-MeanMatrixElemValue         = ( 4.059962e+00 +- 2.367792e+00 )  GeV^-4
-TOTAL       :     2.336070 sec
-     6,777,659,542      cycles                           #    2.897 GHz                    
-    19,385,758,717      instructions                     #    2.86  insn per cycle         
-       2.340385269 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:69681) (avx2:    0) (512y:    0) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 6.858617e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.871131e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.871131e+03                 )  sec^-1
+MeanMatrixElemValue         = ( 4.059961e+00 +- 2.367791e+00 )  GeV^-4
+TOTAL       :     2.400232 sec
+     6,719,203,240      cycles                           #    2.795 GHz                    
+    19,163,412,782      instructions                     #    2.85  insn per cycle         
+       2.405407499 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:68898) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_f_inl1_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -134,80 +134,80 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_f_inl1_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_f_inl1_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.627486e-04
-Avg ME (F77/C++)    = 6.6274862707273868E-004
-Relative difference = 4.0849182767952624e-08
+Avg ME (F77/C++)    = 6.6274859783433532E-004
+Relative difference = 3.2677016209485094e-09
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_f_inl1_hrd0/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.472442e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.477968e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.477968e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.418642e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.423893e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.423893e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 4.060903e+00 +- 2.367376e+00 )  GeV^-4
-TOTAL       :     1.122029 sec
-     3,176,939,727      cycles                           #    2.822 GHz                    
-     6,808,012,735      instructions                     #    2.14  insn per cycle         
-       1.126367927 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:49077) (512y:    0) (512z:    0)
+TOTAL       :     1.168526 sec
+     3,140,858,608      cycles                           #    2.683 GHz                    
+     6,747,205,943      instructions                     #    2.15  insn per cycle         
+       1.173847287 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:48625) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_f_inl1_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_f_inl1_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_f_inl1_hrd0/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 6.627273e-04
-Avg ME (F77/C++)    = 6.6272731558747466E-004
-Relative difference = 2.3520194007978538e-08
+Avg ME (C++/C++)    = 6.627272e-04
+Avg ME (F77/C++)    = 6.6272724143469353E-004
+Relative difference = 6.252149235286529e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_f_inl1_hrd0/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.771353e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.779044e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.779044e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.703185e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.710717e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.710717e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 4.060903e+00 +- 2.367376e+00 )  GeV^-4
-TOTAL       :     0.933958 sec
-     2,648,978,339      cycles                           #    2.826 GHz                    
-     5,986,332,919      instructions                     #    2.26  insn per cycle         
-       0.938142599 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:42677) (512y:   11) (512z:    0)
+TOTAL       :     0.972109 sec
+     2,610,520,883      cycles                           #    2.675 GHz                    
+     5,931,408,487      instructions                     #    2.27  insn per cycle         
+       0.977161465 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:42219) (512y:   24) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_f_inl1_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_f_inl1_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_f_inl1_hrd0/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 6.627273e-04
-Avg ME (F77/C++)    = 6.6272731558747466E-004
-Relative difference = 2.3520194007978538e-08
+Avg ME (C++/C++)    = 6.627272e-04
+Avg ME (F77/C++)    = 6.6272724143469353E-004
+Relative difference = 6.252149235286529e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_f_inl1_hrd0/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[16] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.477195e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.482900e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.482900e+04                 )  sec^-1
-MeanMatrixElemValue         = ( 4.060904e+00 +- 2.367377e+00 )  GeV^-4
-TOTAL       :     1.118598 sec
-     2,077,867,589      cycles                           #    1.852 GHz                    
-     3,500,933,885      instructions                     #    1.68  insn per cycle         
-       1.122888926 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 5197) (512y:    3) (512z:44822)
+EvtsPerSec[Rmb+ME]     (23) = ( 1.380375e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.385342e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.385342e+04                 )  sec^-1
+MeanMatrixElemValue         = ( 4.060905e+00 +- 2.367377e+00 )  GeV^-4
+TOTAL       :     1.197476 sec
+     2,050,152,648      cycles                           #    1.706 GHz                    
+     3,435,996,672      instructions                     #    1.68  insn per cycle         
+       1.202741015 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 4188) (512y:    9) (512z:44489)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_f_inl1_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -215,8 +215,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_f_inl1_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_f_inl1_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.627275e-04
-Avg ME (F77/C++)    = 6.6272750363879224E-004
-Relative difference = 5.490631193034436e-09
+Avg ME (F77/C++)    = 6.6272748295826550E-004
+Relative difference = 2.5714542480216212e-08
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_f_inl1_hrd1.txt b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_f_inl1_hrd1.txt
index be288b051a..0e738d355a 100644
--- a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_f_inl1_hrd1.txt
+++ b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_f_inl1_hrd1.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_f_inl1_hrd1'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
 
-DATE: 2023-11-24_15:05:29
+DATE: 2024-01-30_05:40:55
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl1_hrd1/gcheck.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 5.522303e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 5.562037e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 5.566663e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.495403e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 5.535166e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.540654e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 4.059596e+00 +- 2.368053e+00 )  GeV^-4
-TOTAL       :     0.487689 sec
-     2,083,511,350      cycles                           #    2.934 GHz                    
-     3,089,371,073      instructions                     #    1.48  insn per cycle         
-       0.770457246 seconds time elapsed
+TOTAL       :     0.492553 sec
+     2,044,553,803      cycles                           #    2.834 GHz                    
+     3,023,997,415      instructions                     #    1.48  insn per cycle         
+       0.781011529 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl1_hrd1/gcheck.exe -p 64 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 255
@@ -65,17 +65,17 @@ WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 .........................................................................
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl1_hrd1/gcheck.exe -p 2048 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 7.692823e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.753155e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.755861e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 7.639095e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.697524e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.700186e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 6.664703e+00 +- 5.072736e+00 )  GeV^-4
-TOTAL       :     1.857687 sec
-     6,225,343,404      cycles                           #    2.971 GHz                    
-    13,475,402,986      instructions                     #    2.16  insn per cycle         
-       2.151878134 seconds time elapsed
+TOTAL       :     1.866900 sec
+     6,069,227,607      cycles                           #    2.871 GHz                    
+    11,631,061,560      instructions                     #    1.92  insn per cycle         
+       2.173672620 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl1_hrd1/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl1_hrd1/fgcheck.exe 2 64 2
@@ -86,47 +86,47 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl1_hrd1/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 5.702624e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 5.703436e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 5.703436e+02                 )  sec^-1
-MeanMatrixElemValue         = ( 4.059969e+00 +- 2.367799e+00 )  GeV^-4
-TOTAL       :    28.768124 sec
-    86,390,205,800      cycles                           #    3.004 GHz                    
-   135,908,144,933      instructions                     #    1.57  insn per cycle         
-      28.772428929 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:15910) (avx2:    0) (512y:    0) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 5.528053e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 5.528817e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.528817e+02                 )  sec^-1
+MeanMatrixElemValue         = ( 4.059968e+00 +- 2.367799e+00 )  GeV^-4
+TOTAL       :    29.676151 sec
+    85,692,453,161      cycles                           #    2.888 GHz                    
+   134,120,579,675      instructions                     #    1.57  insn per cycle         
+      29.681167734 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:16109) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl1_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl1_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_f_inl1_hrd1/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 6.627535e-04
-Avg ME (F77/C++)    = 6.6275352674967369E-004
-Relative difference = 4.0361421941458736e-08
+Avg ME (C++/C++)    = 6.627536e-04
+Avg ME (F77/C++)    = 6.6275357377482830E-004
+Relative difference = 3.95700176737784e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_f_inl1_hrd1/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.022155e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.035059e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.035059e+03                 )  sec^-1
-MeanMatrixElemValue         = ( 4.059962e+00 +- 2.367792e+00 )  GeV^-4
-TOTAL       :     2.343932 sec
-     6,861,828,247      cycles                           #    2.923 GHz                    
-    19,438,837,198      instructions                     #    2.83  insn per cycle         
-       2.348614161 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:69723) (avx2:    0) (512y:    0) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 6.924333e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.936823e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.936823e+03                 )  sec^-1
+MeanMatrixElemValue         = ( 4.059961e+00 +- 2.367791e+00 )  GeV^-4
+TOTAL       :     2.377362 sec
+     6,721,293,685      cycles                           #    2.823 GHz                    
+    19,223,635,236      instructions                     #    2.86  insn per cycle         
+       2.382317911 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:68882) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_f_inl1_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -134,80 +134,80 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_f_inl1_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_f_inl1_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.627486e-04
-Avg ME (F77/C++)    = 6.6274862764021530E-004
-Relative difference = 4.170542995014107e-08
+Avg ME (F77/C++)    = 6.6274859765498573E-004
+Relative difference = 3.538316437387639e-09
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_f_inl1_hrd1/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.496126e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.501849e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.501849e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.449646e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.455242e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.455242e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 4.060903e+00 +- 2.367376e+00 )  GeV^-4
-TOTAL       :     1.104423 sec
-     3,132,664,979      cycles                           #    2.827 GHz                    
-     6,718,780,084      instructions                     #    2.14  insn per cycle         
-       1.108980849 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:47667) (512y:    0) (512z:    0)
+TOTAL       :     1.140025 sec
+     3,079,658,771      cycles                           #    2.692 GHz                    
+     6,686,222,708      instructions                     #    2.17  insn per cycle         
+       1.145080651 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:47416) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_f_inl1_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_f_inl1_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_f_inl1_hrd1/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 6.627273e-04
-Avg ME (F77/C++)    = 6.6272731651051409E-004
-Relative difference = 2.4912983202981302e-08
+Avg ME (C++/C++)    = 6.627272e-04
+Avg ME (F77/C++)    = 6.6272724133897148E-004
+Relative difference = 6.237705578619894e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_f_inl1_hrd1/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.776930e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.784946e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.784946e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.717993e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.725785e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.725785e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 4.060903e+00 +- 2.367376e+00 )  GeV^-4
-TOTAL       :     0.932427 sec
-     2,628,346,991      cycles                           #    2.809 GHz                    
-     5,969,517,035      instructions                     #    2.27  insn per cycle         
-       0.937063186 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:41842) (512y:   13) (512z:    0)
+TOTAL       :     0.963197 sec
+     2,607,305,399      cycles                           #    2.696 GHz                    
+     5,935,632,787      instructions                     #    2.28  insn per cycle         
+       0.968307475 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:41564) (512y:   18) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_f_inl1_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_f_inl1_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_f_inl1_hrd1/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 6.627273e-04
-Avg ME (F77/C++)    = 6.6272731651051409E-004
-Relative difference = 2.4912983202981302e-08
+Avg ME (C++/C++)    = 6.627272e-04
+Avg ME (F77/C++)    = 6.6272724133897148E-004
+Relative difference = 6.237705578619894e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_f_inl1_hrd1/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=1] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=1] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[16] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.476207e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.481770e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.481770e+04                 )  sec^-1
-MeanMatrixElemValue         = ( 4.060904e+00 +- 2.367377e+00 )  GeV^-4
-TOTAL       :     1.119825 sec
-     2,078,177,152      cycles                           #    1.850 GHz                    
-     3,494,307,916      instructions                     #    1.68  insn per cycle         
-       1.124539671 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 4161) (512y:    4) (512z:44465)
+EvtsPerSec[Rmb+ME]     (23) = ( 1.382587e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.387561e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.387561e+04                 )  sec^-1
+MeanMatrixElemValue         = ( 4.060905e+00 +- 2.367377e+00 )  GeV^-4
+TOTAL       :     1.195178 sec
+     2,050,651,524      cycles                           #    1.710 GHz                    
+     3,422,960,187      instructions                     #    1.67  insn per cycle         
+       1.200266882 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3375) (512y:   11) (512z:43966)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_f_inl1_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -215,8 +215,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_f_inl1_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_f_inl1_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.627275e-04
-Avg ME (F77/C++)    = 6.6272750384530066E-004
-Relative difference = 5.80223501432476e-09
+Avg ME (F77/C++)    = 6.6272749650985591E-004
+Relative difference = 5.26633351741962e-09
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_m_inl0_hrd0.txt b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_m_inl0_hrd0.txt
index 5f7b3a9875..7714401e20 100644
--- a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_m_inl0_hrd0.txt
+++ b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_m_inl0_hrd0.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_m_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
 
-DATE: 2023-11-24_14:38:36
+DATE: 2024-01-30_05:03:07
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_m_inl0_hrd0/gcheck.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 3.457884e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.485326e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.487559e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.456900e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.484722e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.487399e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     0.520964 sec
-     2,264,629,245      cycles                           #    3.014 GHz                    
-     3,506,769,468      instructions                     #    1.55  insn per cycle         
-       0.829035526 seconds time elapsed
+TOTAL       :     0.528831 sec
+     2,192,645,567      cycles                           #    2.833 GHz                    
+     3,378,106,633      instructions                     #    1.54  insn per cycle         
+       0.861052908 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_m_inl0_hrd0/gcheck.exe -p 64 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 255
@@ -65,17 +65,17 @@ WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 .........................................................................
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_m_inl0_hrd0/gcheck.exe -p 2048 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 4.121692e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.155469e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.156924e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.113905e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.147620e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.149017e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 6.665112e+00 +- 5.002651e+00 )  GeV^-4
-TOTAL       :     3.040169 sec
-    10,113,736,424      cycles                           #    3.067 GHz                    
-    21,679,010,671      instructions                     #    2.14  insn per cycle         
-       3.353783770 seconds time elapsed
+TOTAL       :     3.049267 sec
+     9,507,642,735      cycles                           #    2.871 GHz                    
+    19,066,132,971      instructions                     #    2.01  insn per cycle         
+       3.371164553 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_m_inl0_hrd0/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_m_inl0_hrd0/fgcheck.exe 2 64 2
@@ -86,20 +86,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_m_inl0_hrd0/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.970949e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.971967e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.971967e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.769606e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.770411e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.770411e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     8.331676 sec
-    25,844,448,081      cycles                           #    3.101 GHz                    
-    79,435,783,122      instructions                     #    3.07  insn per cycle         
-       8.337986155 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 4858) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     9.278224 sec
+    26,812,823,901      cycles                           #    2.889 GHz                    
+    82,462,709,559      instructions                     #    3.08  insn per cycle         
+       9.289930135 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 6623) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_m_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -113,20 +113,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_m_inl0_hrd0/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.738000e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.741428e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.741428e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.509625e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.512894e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.512894e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     4.396976 sec
-    12,662,050,046      cycles                           #    2.877 GHz                    
-    38,549,909,567      instructions                     #    3.04  insn per cycle         
-       4.410716297 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:13163) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     4.686363 sec
+    12,638,766,565      cycles                           #    2.696 GHz                    
+    38,538,047,706      instructions                     #    3.05  insn per cycle         
+       4.708715306 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:12755) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_m_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -140,20 +140,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_m_inl0_hrd0/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 8.572381e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 8.589867e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 8.589867e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 8.005037e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 8.021640e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.021640e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     1.922338 sec
-     5,514,460,304      cycles                           #    2.862 GHz                    
-    13,481,024,869      instructions                     #    2.44  insn per cycle         
-       1.938160725 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:11242) (512y:    0) (512z:    0)
+TOTAL       :     2.058850 sec
+     5,538,789,085      cycles                           #    2.684 GHz                    
+    13,583,257,196      instructions                     #    2.45  insn per cycle         
+       2.079297542 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:10944) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_m_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -161,26 +161,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_m_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_m_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.626675e-04
-Avg ME (F77/C++)    = 6.6266730409276857E-004
-Relative difference = 2.956342832710188e-07
+Avg ME (F77/C++)    = 6.6266730409276836E-004
+Relative difference = 2.9563428359824236e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_m_inl0_hrd0/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 9.241393e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 9.262450e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 9.262450e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 9.175649e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 9.196938e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.196938e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     1.784180 sec
-     4,876,372,587      cycles                           #    2.727 GHz                    
-    12,135,890,910      instructions                     #    2.49  insn per cycle         
-       1.800822098 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:10154) (512y:   79) (512z:    0)
+TOTAL       :     1.797590 sec
+     4,843,535,516      cycles                           #    2.687 GHz                    
+    12,110,039,110      instructions                     #    2.50  insn per cycle         
+       1.813279758 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 9682) (512y:   76) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_m_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -188,26 +188,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_m_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_m_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.626675e-04
-Avg ME (F77/C++)    = 6.6266730409276857E-004
-Relative difference = 2.956342832710188e-07
+Avg ME (F77/C++)    = 6.6266730409276836E-004
+Relative difference = 2.9563428359824236e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_m_inl0_hrd0/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.118322e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.130873e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.130873e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.862805e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.874864e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.874864e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     2.314160 sec
-     4,150,527,960      cycles                           #    1.791 GHz                    
-     6,337,492,716      instructions                     #    1.53  insn per cycle         
-       2.327321030 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1803) (512y:   93) (512z: 9358)
+TOTAL       :     2.399875 sec
+     4,096,013,404      cycles                           #    1.704 GHz                    
+     6,283,624,620      instructions                     #    1.53  insn per cycle         
+       2.418716991 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1528) (512y:   76) (512z: 9010)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_m_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -215,8 +215,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_m_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_m_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.626675e-04
-Avg ME (F77/C++)    = 6.6266730409276857E-004
-Relative difference = 2.956342832710188e-07
+Avg ME (F77/C++)    = 6.6266730409276836E-004
+Relative difference = 2.9563428359824236e-07
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_m_inl0_hrd1.txt b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_m_inl0_hrd1.txt
index d8b5e539f7..9cdb5ea5b9 100644
--- a/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_m_inl0_hrd1.txt
+++ b/epochX/cudacpp/tput/logs_ggttgg_mad/log_ggttgg_mad_m_inl0_hrd1.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_m_inl0_hrd1'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg'
 
-DATE: 2023-11-24_14:39:13
+DATE: 2024-01-30_05:03:46
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_m_inl0_hrd1/gcheck.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 3.493159e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.521560e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.523711e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.463401e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.491582e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.494105e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     0.520820 sec
-     2,242,158,202      cycles                           #    2.985 GHz                    
-     3,540,678,032      instructions                     #    1.58  insn per cycle         
-       0.821178917 seconds time elapsed
+TOTAL       :     0.528345 sec
+     2,191,366,155      cycles                           #    2.835 GHz                    
+     3,376,981,873      instructions                     #    1.54  insn per cycle         
+       0.868249282 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_m_inl0_hrd1/gcheck.exe -p 64 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 255
@@ -65,17 +65,17 @@ WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 .........................................................................
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_m_inl0_hrd1/gcheck.exe -p 2048 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 4.120115e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.154154e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.155531e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.141311e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.175321e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.176779e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 6.665112e+00 +- 5.002651e+00 )  GeV^-4
-TOTAL       :     3.032729 sec
-    10,102,528,341      cycles                           #    3.074 GHz                    
-    22,682,645,762      instructions                     #    2.25  insn per cycle         
-       3.344844132 seconds time elapsed
+TOTAL       :     3.034775 sec
+     9,461,569,572      cycles                           #    2.871 GHz                    
+    21,570,730,622      instructions                     #    2.28  insn per cycle         
+       3.354365055 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_m_inl0_hrd1/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_m_inl0_hrd1/fgcheck.exe 2 64 2
@@ -86,20 +86,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_m_inl0_hrd1/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.963020e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.963993e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.963993e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.763986e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.764820e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.764820e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     8.365135 sec
-    25,986,900,256      cycles                           #    3.106 GHz                    
-    79,446,545,297      instructions                     #    3.06  insn per cycle         
-       8.371739931 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 4505) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     9.308542 sec
+    26,818,191,963      cycles                           #    2.880 GHz                    
+    82,362,969,124      instructions                     #    3.07  insn per cycle         
+       9.331807277 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 6491) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.none_m_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -113,20 +113,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_m_inl0_hrd1/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.587282e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.590429e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.590429e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.494755e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.497969e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.497969e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     4.581482 sec
-    12,652,204,532      cycles                           #    2.760 GHz                    
-    38,520,894,837      instructions                     #    3.04  insn per cycle         
-       4.597176677 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:12930) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     4.702744 sec
+    12,651,856,685      cycles                           #    2.688 GHz                    
+    38,557,643,348      instructions                     #    3.05  insn per cycle         
+       4.723006762 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:12729) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.sse4_m_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -140,20 +140,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_m_inl0_hrd1/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 8.252221e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 8.267695e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 8.267695e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 8.057026e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 8.073448e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.073448e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     1.997543 sec
-     5,563,118,159      cycles                           #    2.781 GHz                    
-    13,606,463,107      instructions                     #    2.45  insn per cycle         
-       2.009289653 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:11327) (512y:    0) (512z:    0)
+TOTAL       :     2.045356 sec
+     5,503,322,263      cycles                           #    2.685 GHz                    
+    13,599,131,001      instructions                     #    2.47  insn per cycle         
+       2.065937163 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:10926) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_m_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -161,26 +161,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_m_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.avx2_m_inl0_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.626675e-04
-Avg ME (F77/C++)    = 6.6266730409276857E-004
-Relative difference = 2.956342832710188e-07
+Avg ME (F77/C++)    = 6.6266730409276836E-004
+Relative difference = 2.9563428359824236e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_m_inl0_hrd1/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 9.355631e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 9.376642e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 9.376642e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 9.173965e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 9.195231e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.195231e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     1.762069 sec
-     4,918,440,362      cycles                           #    2.785 GHz                    
-    12,270,891,194      instructions                     #    2.49  insn per cycle         
-       1.776725149 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:10143) (512y:  239) (512z:    0)
+TOTAL       :     1.797623 sec
+     4,836,406,491      cycles                           #    2.684 GHz                    
+    12,123,840,407      instructions                     #    2.51  insn per cycle         
+       1.816744592 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 9659) (512y:   76) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_m_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -188,26 +188,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_m_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512y_m_inl0_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.626675e-04
-Avg ME (F77/C++)    = 6.6266730409276857E-004
-Relative difference = 2.956342832710188e-07
+Avg ME (F77/C++)    = 6.6266730409276836E-004
+Relative difference = 2.9563428359824236e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_m_inl0_hrd1/check.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.144291e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.157521e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.157521e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.872297e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.884618e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.884618e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 4.063123e+00 +- 2.368970e+00 )  GeV^-4
-TOTAL       :     2.305102 sec
-     4,151,951,816      cycles                           #    1.798 GHz                    
-     6,443,598,853      instructions                     #    1.55  insn per cycle         
-       2.315841771 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1628) (512y:  191) (512z: 9356)
+TOTAL       :     2.396265 sec
+     4,088,419,794      cycles                           #    1.703 GHz                    
+     6,289,480,909      instructions                     #    1.54  insn per cycle         
+       2.414194012 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1508) (512y:   76) (512z: 9009)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_m_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -215,8 +215,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProces
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_m_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/build.512z_m_inl0_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 6.626675e-04
-Avg ME (F77/C++)    = 6.6266730409276857E-004
-Relative difference = 2.956342832710188e-07
+Avg ME (F77/C++)    = 6.6266730409276836E-004
+Relative difference = 2.9563428359824236e-07
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_ggttggg_mad/log_ggttggg_mad_d_inl0_hrd0.txt b/epochX/cudacpp/tput/logs_ggttggg_mad/log_ggttggg_mad_d_inl0_hrd0.txt
index 8be70208e9..10dc25694a 100644
--- a/epochX/cudacpp/tput/logs_ggttggg_mad/log_ggttggg_mad_d_inl0_hrd0.txt
+++ b/epochX/cudacpp/tput/logs_ggttggg_mad/log_ggttggg_mad_d_inl0_hrd0.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg'
 
-DATE: 2023-11-24_14:41:33
+DATE: 2024-01-30_05:06:15
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_d_inl0_hrd0/gcheck.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 4.061812e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.062202e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.062394e+02                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.064289e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.064686e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.064874e+02                 )  sec^-1
 MeanMatrixElemValue         = ( 1.187066e-05 +- 9.825549e-06 )  GeV^-6
-TOTAL       :     2.449920 sec
-     8,136,633,317      cycles                           #    2.967 GHz                    
-    17,299,512,986      instructions                     #    2.13  insn per cycle         
-       2.841850939 seconds time elapsed
+TOTAL       :     2.459219 sec
+     7,914,579,350      cycles                           #    2.876 GHz                    
+    17,414,362,649      instructions                     #    2.20  insn per cycle         
+       2.856648920 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_d_inl0_hrd0/gcheck.exe -p 1 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 255
@@ -65,17 +65,17 @@ WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 .........................................................................
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_d_inl0_hrd0/gcheck.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 9.240506e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 9.242687e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 9.242958e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 9.261836e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 9.264181e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.264456e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 1.856249e-04 +- 8.329951e-05 )  GeV^-6
-TOTAL       :     3.988007 sec
-    13,311,291,301      cycles                           #    3.086 GHz                    
-    31,141,162,586      instructions                     #    2.34  insn per cycle         
-       4.368862794 seconds time elapsed
+TOTAL       :     4.001930 sec
+    12,466,660,301      cycles                           #    2.881 GHz                    
+    28,598,806,424      instructions                     #    2.29  insn per cycle         
+       4.385332309 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_d_inl0_hrd0/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_d_inl0_hrd0/fgcheck.exe 2 64 2
@@ -86,20 +86,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_d_inl0_hrd0/check.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 8.074588e+01                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 8.074809e+01                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 8.074809e+01                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 7.667093e+01                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.667308e+01                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.667308e+01                 )  sec^-1
 MeanMatrixElemValue         = ( 1.187066e-05 +- 9.825549e-06 )  GeV^-6
-TOTAL       :     6.548778 sec
-    18,889,945,267      cycles                           #    2.886 GHz                    
-    53,916,181,025      instructions                     #    2.85  insn per cycle         
-       6.559444585 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:32448) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     6.887821 sec
+    18,997,365,246      cycles                           #    2.759 GHz                    
+    55,182,817,229      instructions                     #    2.90  insn per cycle         
+       6.894930966 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:44874) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -113,20 +113,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.sse4_d_inl0_hrd0/check.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.664916e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.665007e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.665007e+02                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.565125e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.565211e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.565211e+02                 )  sec^-1
 MeanMatrixElemValue         = ( 1.187066e-05 +- 9.825549e-06 )  GeV^-6
-TOTAL       :     3.179365 sec
-     9,817,296,962      cycles                           #    3.087 GHz                    
-    27,093,187,078      instructions                     #    2.76  insn per cycle         
-       3.190791457 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:96443) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     3.384131 sec
+     9,789,568,447      cycles                           #    2.893 GHz                    
+    27,057,217,068      instructions                     #    2.76  insn per cycle         
+       3.398188002 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:97234) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.sse4_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -140,20 +140,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.avx2_d_inl0_hrd0/check.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.619576e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.620013e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.620013e+02                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.331784e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.332213e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.332213e+02                 )  sec^-1
 MeanMatrixElemValue         = ( 1.187066e-05 +- 9.825549e-06 )  GeV^-6
-TOTAL       :     1.467898 sec
-     4,254,332,720      cycles                           #    2.899 GHz                    
-     9,562,072,100      instructions                     #    2.25  insn per cycle         
-       1.478868560 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:84390) (512y:    0) (512z:    0)
+TOTAL       :     1.593056 sec
+     4,251,132,724      cycles                           #    2.667 GHz                    
+     9,566,982,441      instructions                     #    2.25  insn per cycle         
+       1.603318722 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:84279) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.avx2_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -167,20 +167,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512y_d_inl0_hrd0/check.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 4.155197e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.155766e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.155766e+02                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.782288e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.782847e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.782847e+02                 )  sec^-1
 MeanMatrixElemValue         = ( 1.187066e-05 +- 9.825549e-06 )  GeV^-6
-TOTAL       :     1.277897 sec
-     3,721,805,443      cycles                           #    2.907 GHz                    
-     8,486,279,618      instructions                     #    2.28  insn per cycle         
-       1.292716754 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:79991) (512y:   91) (512z:    0)
+TOTAL       :     1.405349 sec
+     3,719,980,949      cycles                           #    2.646 GHz                    
+     8,451,730,597      instructions                     #    2.27  insn per cycle         
+       1.418908281 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:79441) (512y:   90) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512y_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -194,20 +194,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512z_d_inl0_hrd0/check.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.734112e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.734785e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.734785e+02                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.332107e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.332611e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.332611e+02                 )  sec^-1
 MeanMatrixElemValue         = ( 1.187066e-05 +- 9.825549e-06 )  GeV^-6
-TOTAL       :     1.420502 sec
-     2,698,172,498      cycles                           #    1.898 GHz                    
-     4,274,439,269      instructions                     #    1.58  insn per cycle         
-       1.431694585 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2284) (512y:  105) (512z:79105)
+TOTAL       :     1.593467 sec
+     2,690,971,905      cycles                           #    1.687 GHz                    
+     4,249,909,932      instructions                     #    1.58  insn per cycle         
+       1.609272621 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2166) (512y:   90) (512z:78318)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512z_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
diff --git a/epochX/cudacpp/tput/logs_ggttggg_mad/log_ggttggg_mad_d_inl0_hrd0_bridge.txt b/epochX/cudacpp/tput/logs_ggttggg_mad/log_ggttggg_mad_d_inl0_hrd0_bridge.txt
index cc408356f3..14598d99fd 100644
--- a/epochX/cudacpp/tput/logs_ggttggg_mad/log_ggttggg_mad_d_inl0_hrd0_bridge.txt
+++ b/epochX/cudacpp/tput/logs_ggttggg_mad/log_ggttggg_mad_d_inl0_hrd0_bridge.txt
@@ -41,7 +41,7 @@ CUDACPP_BUILDDIR='build.512z_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg'
 
-DATE: 2023-11-24_15:14:25
+DATE: 2024-01-30_05:50:15
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
@@ -51,17 +51,17 @@ WARNING! Bridge selected: cannot use RamboDevice, will use RamboHost
 WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
 WARNING! Instantiate device Bridge (nevt=256, gpublocks=1, gputhreads=256, gpublocks*gputhreads=256)
 WARNING! Set grid in Bridge (nevt=256, gpublocks=1, gputhreads=256, gpublocks*gputhreads=256)
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 4.059911e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.060821e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.060821e+02                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.062580e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.063573e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.063573e+02                 )  sec^-1
 MeanMatrixElemValue         = ( 1.187066e-05 +- 9.825549e-06 )  GeV^-6
-TOTAL       :     2.369780 sec
-     8,074,775,034      cycles                           #    3.005 GHz                    
-    17,556,507,233      instructions                     #    2.17  insn per cycle         
-       2.745666195 seconds time elapsed
+TOTAL       :     2.393250 sec
+     7,805,787,223      cycles                           #    2.878 GHz                    
+    17,759,546,689      instructions                     #    2.28  insn per cycle         
+       2.771767839 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_d_inl0_hrd0/gcheck.exe -p 1 256 1 --bridge
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Bridge selected: cannot use RamboDevice, will use RamboHost
@@ -77,17 +77,17 @@ WARNING! Bridge selected: cannot use RamboDevice, will use RamboHost
 WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
 WARNING! Instantiate device Bridge (nevt=16384, gpublocks=64, gputhreads=256, gpublocks*gputhreads=16384)
 WARNING! Set grid in Bridge (nevt=16384, gpublocks=64, gputhreads=256, gpublocks*gputhreads=16384)
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 9.206968e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 9.239855e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 9.239855e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 9.205412e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 9.241153e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.241153e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 1.856249e-04 +- 8.329951e-05 )  GeV^-6
-TOTAL       :     3.981224 sec
-    12,936,091,202      cycles                           #    3.008 GHz                    
-    29,929,520,175      instructions                     #    2.31  insn per cycle         
-       4.359803213 seconds time elapsed
+TOTAL       :     3.999133 sec
+    12,487,046,648      cycles                           #    2.887 GHz                    
+    29,181,392,973      instructions                     #    2.34  insn per cycle         
+       4.379902707 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_d_inl0_hrd0/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_d_inl0_hrd0/fgcheck.exe 2 64 2
@@ -99,20 +99,20 @@ OK (relative difference <= 5E-3)
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_d_inl0_hrd0/check.exe -p 1 256 2 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=256)
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.880447e+01                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.880664e+01                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.880664e+01                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 7.924049e+01                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.924280e+01                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.924280e+01                 )  sec^-1
 MeanMatrixElemValue         = ( 1.187066e-05 +- 9.825549e-06 )  GeV^-6
-TOTAL       :     6.704716 sec
-    18,908,384,031      cycles                           #    2.819 GHz                    
-    53,916,851,861      instructions                     #    2.85  insn per cycle         
-       6.708873310 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:32448) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     6.668799 sec
+    18,978,883,548      cycles                           #    2.845 GHz                    
+    55,181,310,686      instructions                     #    2.91  insn per cycle         
+       6.673958990 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:44874) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -127,20 +127,20 @@ OK (relative difference <= 5E-3)
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.sse4_d_inl0_hrd0/check.exe -p 1 256 2 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=256)
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/sse4+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.623567e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.623659e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.623659e+02                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.558442e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.558530e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.558530e+02                 )  sec^-1
 MeanMatrixElemValue         = ( 1.187066e-05 +- 9.825549e-06 )  GeV^-6
-TOTAL       :     3.257235 sec
-     9,753,060,796      cycles                           #    2.991 GHz                    
-    27,093,339,492      instructions                     #    2.78  insn per cycle         
-       3.261809851 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:96443) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     3.394020 sec
+     9,815,752,501      cycles                           #    2.889 GHz                    
+    27,056,612,659      instructions                     #    2.76  insn per cycle         
+       3.399148950 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:97234) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.sse4_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -155,20 +155,20 @@ OK (relative difference <= 5E-3)
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.avx2_d_inl0_hrd0/check.exe -p 1 256 2 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=256)
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/avx2+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.503641e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.504075e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.504075e+02                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.345002e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.345461e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.345461e+02                 )  sec^-1
 MeanMatrixElemValue         = ( 1.187066e-05 +- 9.825549e-06 )  GeV^-6
-TOTAL       :     1.512954 sec
-     4,263,837,997      cycles                           #    2.812 GHz                    
-     9,562,348,938      instructions                     #    2.24  insn per cycle         
-       1.517455505 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:84390) (512y:    0) (512z:    0)
+TOTAL       :     1.587609 sec
+     4,248,692,453      cycles                           #    2.674 GHz                    
+     9,567,437,136      instructions                     #    2.25  insn per cycle         
+       1.592590793 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:84279) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.avx2_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -183,20 +183,20 @@ OK (relative difference <= 5E-3)
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512y_d_inl0_hrd0/check.exe -p 1 256 2 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=256)
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/512y+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 4.030228e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.030821e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.030821e+02                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.873515e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.874138e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.874138e+02                 )  sec^-1
 MeanMatrixElemValue         = ( 1.187066e-05 +- 9.825549e-06 )  GeV^-6
-TOTAL       :     1.317705 sec
-     3,745,717,606      cycles                           #    2.836 GHz                    
-     8,485,950,351      instructions                     #    2.27  insn per cycle         
-       1.322163213 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:79991) (512y:   91) (512z:    0)
+TOTAL       :     1.369431 sec
+     3,692,449,005      cycles                           #    2.689 GHz                    
+     8,450,968,058      instructions                     #    2.29  insn per cycle         
+       1.374284426 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:79441) (512y:   90) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512y_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -211,20 +211,20 @@ OK (relative difference <= 5E-3)
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512z_d_inl0_hrd0/check.exe -p 1 256 2 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=256)
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/512z+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.640278e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.640943e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.640943e+02                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.369341e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.369854e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.369854e+02                 )  sec^-1
 MeanMatrixElemValue         = ( 1.187066e-05 +- 9.825549e-06 )  GeV^-6
-TOTAL       :     1.457064 sec
-     2,701,126,575      cycles                           #    1.849 GHz                    
-     4,273,675,572      instructions                     #    1.58  insn per cycle         
-       1.461485660 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2284) (512y:  105) (512z:79105)
+TOTAL       :     1.574041 sec
+     2,686,211,452      cycles                           #    1.702 GHz                    
+     4,249,274,815      instructions                     #    1.58  insn per cycle         
+       1.579137128 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2166) (512y:   90) (512z:78318)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512z_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
diff --git a/epochX/cudacpp/tput/logs_ggttggg_mad/log_ggttggg_mad_d_inl0_hrd1.txt b/epochX/cudacpp/tput/logs_ggttggg_mad/log_ggttggg_mad_d_inl0_hrd1.txt
index 5875d438d6..869fccfa2f 100644
--- a/epochX/cudacpp/tput/logs_ggttggg_mad/log_ggttggg_mad_d_inl0_hrd1.txt
+++ b/epochX/cudacpp/tput/logs_ggttggg_mad/log_ggttggg_mad_d_inl0_hrd1.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_d_inl0_hrd1'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg'
 
-DATE: 2023-11-24_14:42:37
+DATE: 2024-01-30_05:07:21
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_d_inl0_hrd1/gcheck.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 4.059999e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.060392e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.060490e+02                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.062893e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.063296e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.063519e+02                 )  sec^-1
 MeanMatrixElemValue         = ( 1.187066e-05 +- 9.825549e-06 )  GeV^-6
-TOTAL       :     2.458037 sec
-     8,400,321,652      cycles                           #    3.063 GHz                    
-    17,432,347,823      instructions                     #    2.08  insn per cycle         
-       2.856914697 seconds time elapsed
+TOTAL       :     2.463696 sec
+     7,904,124,830      cycles                           #    2.867 GHz                    
+    17,962,469,169      instructions                     #    2.27  insn per cycle         
+       2.863348110 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_d_inl0_hrd1/gcheck.exe -p 1 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 255
@@ -65,17 +65,17 @@ WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 .........................................................................
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_d_inl0_hrd1/gcheck.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 9.242262e+03                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 9.244491e+03                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 9.244679e+03                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 9.275434e+03                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 9.277655e+03                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.278066e+03                 )  sec^-1
 MeanMatrixElemValue         = ( 1.856249e-04 +- 8.329951e-05 )  GeV^-6
-TOTAL       :     4.000355 sec
-    13,346,085,971      cycles                           #    3.087 GHz                    
-    30,375,563,442      instructions                     #    2.28  insn per cycle         
-       4.381330754 seconds time elapsed
+TOTAL       :     4.004406 sec
+    12,472,043,203      cycles                           #    2.872 GHz                    
+    27,476,431,943      instructions                     #    2.20  insn per cycle         
+       4.397866058 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_d_inl0_hrd1/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_d_inl0_hrd1/fgcheck.exe 2 64 2
@@ -86,20 +86,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_d_inl0_hrd1/check.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 8.428708e+01                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 8.428938e+01                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 8.428938e+01                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 7.993181e+01                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.993429e+01                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.993429e+01                 )  sec^-1
 MeanMatrixElemValue         = ( 1.187066e-05 +- 9.825549e-06 )  GeV^-6
-TOTAL       :     6.266392 sec
-    18,760,509,477      cycles                           #    2.994 GHz                    
-    53,924,653,614      instructions                     #    2.87  insn per cycle         
-       6.272821275 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:32063) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     6.616864 sec
+    18,937,214,023      cycles                           #    2.863 GHz                    
+    55,162,675,285      instructions                     #    2.91  insn per cycle         
+       6.624084944 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:44747) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_d_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -113,20 +113,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.sse4_d_inl0_hrd1/check.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.667610e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.667700e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.667700e+02                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.560244e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.560337e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.560337e+02                 )  sec^-1
 MeanMatrixElemValue         = ( 1.187066e-05 +- 9.825549e-06 )  GeV^-6
-TOTAL       :     3.175383 sec
-     9,846,531,791      cycles                           #    3.102 GHz                    
-    27,090,667,624      instructions                     #    2.75  insn per cycle         
-       3.186565682 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:96286) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     3.391033 sec
+     9,810,909,577      cycles                           #    2.891 GHz                    
+    27,064,931,751      instructions                     #    2.76  insn per cycle         
+       3.404410372 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:97230) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.sse4_d_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -140,20 +140,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.avx2_d_inl0_hrd1/check.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.613188e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.613622e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.613622e+02                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.366743e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.367151e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.367151e+02                 )  sec^-1
 MeanMatrixElemValue         = ( 1.187066e-05 +- 9.825549e-06 )  GeV^-6
-TOTAL       :     1.470967 sec
-     4,269,806,021      cycles                           #    2.905 GHz                    
-     9,562,364,627      instructions                     #    2.24  insn per cycle         
-       1.483852740 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:84478) (512y:    0) (512z:    0)
+TOTAL       :     1.577213 sec
+     4,241,194,499      cycles                           #    2.687 GHz                    
+     9,570,392,055      instructions                     #    2.26  insn per cycle         
+       1.590680511 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:84249) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.avx2_d_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -167,20 +167,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512y_d_inl0_hrd1/check.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 4.136889e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.137474e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.137474e+02                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.823083e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.823621e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.823621e+02                 )  sec^-1
 MeanMatrixElemValue         = ( 1.187066e-05 +- 9.825549e-06 )  GeV^-6
-TOTAL       :     1.283162 sec
-     3,736,243,344      cycles                           #    2.908 GHz                    
-     8,485,942,736      instructions                     #    2.27  insn per cycle         
-       1.303492183 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:80014) (512y:  241) (512z:    0)
+TOTAL       :     1.389663 sec
+     3,742,544,913      cycles                           #    2.690 GHz                    
+     8,455,558,047      instructions                     #    2.26  insn per cycle         
+       1.401942381 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:79386) (512y:   90) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512y_d_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -194,20 +194,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512z_d_inl0_hrd1/check.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.757031e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.757649e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.757649e+02                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.367545e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.368096e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.368096e+02                 )  sec^-1
 MeanMatrixElemValue         = ( 1.187066e-05 +- 9.825549e-06 )  GeV^-6
-TOTAL       :     1.412617 sec
-     2,696,840,598      cycles                           #    1.907 GHz                    
-     4,277,582,192      instructions                     #    1.59  insn per cycle         
-       1.424037126 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2169) (512y:  187) (512z:79110)
+TOTAL       :     1.578445 sec
+     2,686,793,480      cycles                           #    1.702 GHz                    
+     4,251,847,609      instructions                     #    1.58  insn per cycle         
+       1.591347897 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2130) (512y:   90) (512z:78289)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512z_d_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
diff --git a/epochX/cudacpp/tput/logs_ggttggg_mad/log_ggttggg_mad_f_inl0_hrd0.txt b/epochX/cudacpp/tput/logs_ggttggg_mad/log_ggttggg_mad_f_inl0_hrd0.txt
index bdc444302c..a75bd83e48 100644
--- a/epochX/cudacpp/tput/logs_ggttggg_mad/log_ggttggg_mad_f_inl0_hrd0.txt
+++ b/epochX/cudacpp/tput/logs_ggttggg_mad/log_ggttggg_mad_f_inl0_hrd0.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_f_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg'
 
-DATE: 2023-11-24_14:43:40
+DATE: 2024-01-30_05:08:27
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_f_inl0_hrd0/gcheck.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 6.755368e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 6.756413e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 6.756816e+02                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.769847e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.770754e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.771164e+02                 )  sec^-1
 MeanMatrixElemValue         = ( 1.186984e-05 +- 9.824899e-06 )  GeV^-6
-TOTAL       :     1.683293 sec
-     5,751,994,986      cycles                           #    2.961 GHz                    
-    11,825,851,354      instructions                     #    2.06  insn per cycle         
-       2.048483718 seconds time elapsed
+TOTAL       :     1.703094 sec
+     5,571,181,653      cycles                           #    2.867 GHz                    
+    11,974,166,174      instructions                     #    2.15  insn per cycle         
+       2.057946232 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_f_inl0_hrd0/gcheck.exe -p 1 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 255
@@ -65,17 +65,17 @@ WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 .........................................................................
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_f_inl0_hrd0/gcheck.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 2.349504e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.350297e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.350390e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.318486e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.319261e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.319430e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 1.856829e-04 +- 8.333435e-05 )  GeV^-6
-TOTAL       :     1.907896 sec
-     6,721,068,256      cycles                           #    3.077 GHz                    
-    14,203,232,269      instructions                     #    2.11  insn per cycle         
-       2.241134413 seconds time elapsed
+TOTAL       :     1.904733 sec
+     6,266,697,659      cycles                           #    2.868 GHz                    
+    13,596,680,456      instructions                     #    2.17  insn per cycle         
+       2.241129899 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_f_inl0_hrd0/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_f_inl0_hrd0/fgcheck.exe 2 64 2
@@ -86,20 +86,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_f_inl0_hrd0/check.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 9.138313e+01                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 9.138596e+01                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 9.138596e+01                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 8.651013e+01                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 8.651286e+01                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.651286e+01                 )  sec^-1
 MeanMatrixElemValue         = ( 1.187013e-05 +- 9.825040e-06 )  GeV^-6
-TOTAL       :     5.786618 sec
-    17,878,554,532      cycles                           #    3.089 GHz                    
-    53,588,537,371      instructions                     #    3.00  insn per cycle         
-       5.793567200 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:20208) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     6.115216 sec
+    17,580,950,028      cycles                           #    2.876 GHz                    
+    51,788,424,956      instructions                     #    2.95  insn per cycle         
+       6.122234952 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:27812) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -107,26 +107,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProce
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 9.847961e-03
-Avg ME (F77/C++)    = 9.8479612087541066E-003
-Relative difference = 2.1197698286506752e-08
+Avg ME (F77/C++)    = 9.8479612087330436E-003
+Relative difference = 2.119555946686223e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.sse4_f_inl0_hrd0/check.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.595673e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.596126e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.596126e+02                 )  sec^-1
-MeanMatrixElemValue         = ( 1.187013e-05 +- 9.825037e-06 )  GeV^-6
-TOTAL       :     1.481350 sec
-     4,569,631,683      cycles                           #    3.086 GHz                    
-    13,762,990,034      instructions                     #    3.01  insn per cycle         
-       1.491903170 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:96986) (avx2:    0) (512y:    0) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 3.365857e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.366295e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.366295e+02                 )  sec^-1
+MeanMatrixElemValue         = ( 1.187013e-05 +- 9.825038e-06 )  GeV^-6
+TOTAL       :     1.576617 sec
+     4,544,162,423      cycles                           #    2.878 GHz                    
+    13,760,085,205      instructions                     #    3.03  insn per cycle         
+       1.587566374 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:97762) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.sse4_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -134,26 +134,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProce
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.sse4_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.sse4_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 9.847955e-03
-Avg ME (F77/C++)    = 9.8479546896527003E-003
-Relative difference = 3.151388282563952e-08
+Avg ME (F77/C++)    = 9.8479546894727158E-003
+Relative difference = 3.1532159158088894e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.avx2_f_inl0_hrd0/check.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.220064e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.221950e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.221950e+02                 )  sec^-1
-MeanMatrixElemValue         = ( 1.187188e-05 +- 9.826767e-06 )  GeV^-6
-TOTAL       :     0.739476 sec
-     2,140,174,105      cycles                           #    2.889 GHz                    
-     4,817,678,891      instructions                     #    2.25  insn per cycle         
-       0.765666592 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:84904) (512y:    0) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 6.652038e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.653755e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.653755e+02                 )  sec^-1
+MeanMatrixElemValue         = ( 1.187187e-05 +- 9.826763e-06 )  GeV^-6
+TOTAL       :     0.803941 sec
+     2,147,173,176      cycles                           #    2.667 GHz                    
+     4,827,637,015      instructions                     #    2.25  insn per cycle         
+       0.818354401 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:84831) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.avx2_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -161,26 +161,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProce
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.avx2_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.avx2_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 9.892973e-03
-Avg ME (F77/C++)    = 9.8929728161070551E-003
-Relative difference = 1.858823877057982e-08
+Avg ME (F77/C++)    = 9.8929728159608508E-003
+Relative difference = 1.8603017364363385e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512y_f_inl0_hrd0/check.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 8.208843e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 8.211063e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 8.211063e+02                 )  sec^-1
-MeanMatrixElemValue         = ( 1.187188e-05 +- 9.826767e-06 )  GeV^-6
-TOTAL       :     0.650167 sec
-     1,891,036,782      cycles                           #    2.902 GHz                    
-     4,274,768,942      instructions                     #    2.26  insn per cycle         
-       0.663284347 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:80610) (512y:   46) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 7.264093e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.266084e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.266084e+02                 )  sec^-1
+MeanMatrixElemValue         = ( 1.187187e-05 +- 9.826763e-06 )  GeV^-6
+TOTAL       :     0.735161 sec
+     1,890,652,826      cycles                           #    2.565 GHz                    
+     4,260,215,320      instructions                     #    2.25  insn per cycle         
+       0.752160652 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:80038) (512y:   46) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512y_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -188,35 +188,35 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProce
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512y_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512y_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 9.892973e-03
-Avg ME (F77/C++)    = 9.8929728161070551E-003
-Relative difference = 1.858823877057982e-08
+Avg ME (F77/C++)    = 9.8929728159608508E-003
+Relative difference = 1.8603017364363385e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512z_f_inl0_hrd0/check.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[16] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.569007e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.571564e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.571564e+02                 )  sec^-1
-MeanMatrixElemValue         = ( 1.187188e-05 +- 9.826771e-06 )  GeV^-6
-TOTAL       :     0.710143 sec
-     1,360,534,299      cycles                           #    1.926 GHz                    
-     2,159,690,774      instructions                     #    1.59  insn per cycle         
-       0.724501882 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2877) (512y:   49) (512z:79298)
+EvtsPerSec[Rmb+ME]     (23) = ( 6.595587e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.597618e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.597618e+02                 )  sec^-1
+MeanMatrixElemValue         = ( 1.187188e-05 +- 9.826770e-06 )  GeV^-6
+TOTAL       :     0.809620 sec
+     1,357,631,253      cycles                           #    1.673 GHz                    
+     2,149,171,041      instructions                     #    1.58  insn per cycle         
+       0.843747051 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2820) (512y:   44) (512z:78510)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512z_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512z_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512z_f_inl0_hrd0/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 9.892981e-03
-Avg ME (F77/C++)    = 9.8929811982958280E-003
-Relative difference = 2.0044092642523172e-08
+Avg ME (C++/C++)    = 9.892980e-03
+Avg ME (F77/C++)    = 9.8929802670331551E-003
+Relative difference = 2.699218597469717e-08
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_ggttggg_mad/log_ggttggg_mad_f_inl0_hrd0_bridge.txt b/epochX/cudacpp/tput/logs_ggttggg_mad/log_ggttggg_mad_f_inl0_hrd0_bridge.txt
index b005f4754c..dd846fe890 100644
--- a/epochX/cudacpp/tput/logs_ggttggg_mad/log_ggttggg_mad_f_inl0_hrd0_bridge.txt
+++ b/epochX/cudacpp/tput/logs_ggttggg_mad/log_ggttggg_mad_f_inl0_hrd0_bridge.txt
@@ -41,7 +41,7 @@ CUDACPP_BUILDDIR='build.512z_f_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg'
 
-DATE: 2023-11-24_15:15:29
+DATE: 2024-01-30_05:51:20
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
@@ -51,17 +51,17 @@ WARNING! Bridge selected: cannot use RamboDevice, will use RamboHost
 WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
 WARNING! Instantiate device Bridge (nevt=256, gpublocks=1, gputhreads=256, gpublocks*gputhreads=256)
 WARNING! Set grid in Bridge (nevt=256, gpublocks=1, gputhreads=256, gpublocks*gputhreads=256)
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 6.768306e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 6.770186e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 6.770186e+02                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.783457e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.785575e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.785575e+02                 )  sec^-1
 MeanMatrixElemValue         = ( 1.187094e-05 +- 9.825664e-06 )  GeV^-6
-TOTAL       :     1.618405 sec
-     5,667,284,995      cycles                           #    2.995 GHz                    
-    11,929,859,554      instructions                     #    2.11  insn per cycle         
-       1.951873227 seconds time elapsed
+TOTAL       :     1.618103 sec
+     5,426,392,715      cycles                           #    2.867 GHz                    
+    11,041,442,286      instructions                     #    2.03  insn per cycle         
+       1.951599799 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_f_inl0_hrd0/gcheck.exe -p 1 256 1 --bridge
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Bridge selected: cannot use RamboDevice, will use RamboHost
@@ -77,17 +77,17 @@ WARNING! Bridge selected: cannot use RamboDevice, will use RamboHost
 WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
 WARNING! Instantiate device Bridge (nevt=16384, gpublocks=64, gputhreads=256, gpublocks*gputhreads=16384)
 WARNING! Set grid in Bridge (nevt=16384, gpublocks=64, gputhreads=256, gpublocks*gputhreads=16384)
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 2.327456e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.341794e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.341794e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.306053e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.319762e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.319762e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 1.856441e-04 +- 8.331096e-05 )  GeV^-6
-TOTAL       :     1.910757 sec
-     6,253,041,183      cycles                           #    2.847 GHz                    
-    13,520,927,317      instructions                     #    2.16  insn per cycle         
-       2.256378130 seconds time elapsed
+TOTAL       :     1.925915 sec
+     6,319,647,912      cycles                           #    2.872 GHz                    
+    13,785,417,374      instructions                     #    2.18  insn per cycle         
+       2.259607907 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_f_inl0_hrd0/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_f_inl0_hrd0/fgcheck.exe 2 64 2
@@ -99,20 +99,20 @@ OK (relative difference <= 5E-3)
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_f_inl0_hrd0/check.exe -p 1 256 2 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=256)
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 8.875191e+01                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 8.875509e+01                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 8.875509e+01                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 8.635189e+01                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 8.635467e+01                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.635467e+01                 )  sec^-1
 MeanMatrixElemValue         = ( 1.187013e-05 +- 9.825040e-06 )  GeV^-6
-TOTAL       :     5.953864 sec
-    17,981,915,231      cycles                           #    3.018 GHz                    
-    53,589,180,868      instructions                     #    2.98  insn per cycle         
-       5.958154287 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:20208) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     6.119375 sec
+    17,637,027,949      cycles                           #    2.881 GHz                    
+    51,787,792,256      instructions                     #    2.94  insn per cycle         
+       6.124243714 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:27812) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -120,27 +120,27 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProce
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 9.847961e-03
-Avg ME (F77/C++)    = 9.8479612087541066E-003
-Relative difference = 2.1197698286506752e-08
+Avg ME (F77/C++)    = 9.8479612087330436E-003
+Relative difference = 2.119555946686223e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.sse4_f_inl0_hrd0/check.exe -p 1 256 2 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=256)
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/sse4+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.525468e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.525921e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.525921e+02                 )  sec^-1
-MeanMatrixElemValue         = ( 1.187013e-05 +- 9.825037e-06 )  GeV^-6
-TOTAL       :     1.503844 sec
-     4,572,076,788      cycles                           #    3.033 GHz                    
-    13,762,724,678      instructions                     #    3.01  insn per cycle         
-       1.508341488 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:96986) (avx2:    0) (512y:    0) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 3.362357e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.362789e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.362789e+02                 )  sec^-1
+MeanMatrixElemValue         = ( 1.187013e-05 +- 9.825038e-06 )  GeV^-6
+TOTAL       :     1.576085 sec
+     4,544,551,937      cycles                           #    2.877 GHz                    
+    13,759,350,934      instructions                     #    3.03  insn per cycle         
+       1.581388093 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:97762) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.sse4_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -148,27 +148,27 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProce
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.sse4_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.sse4_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 9.847955e-03
-Avg ME (F77/C++)    = 9.8479546896527003E-003
-Relative difference = 3.151388282563952e-08
+Avg ME (F77/C++)    = 9.8479546894727158E-003
+Relative difference = 3.1532159158088894e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.avx2_f_inl0_hrd0/check.exe -p 1 256 2 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=256)
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/avx2+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.089959e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.091818e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.091818e+02                 )  sec^-1
-MeanMatrixElemValue         = ( 1.187188e-05 +- 9.826767e-06 )  GeV^-6
-TOTAL       :     0.750792 sec
-     2,139,360,697      cycles                           #    2.835 GHz                    
-     4,817,322,213      instructions                     #    2.25  insn per cycle         
-       0.755255088 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:84904) (512y:    0) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 6.701025e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.702845e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.702845e+02                 )  sec^-1
+MeanMatrixElemValue         = ( 1.187187e-05 +- 9.826763e-06 )  GeV^-6
+TOTAL       :     0.794089 sec
+     2,138,661,629      cycles                           #    2.680 GHz                    
+     4,826,930,405      instructions                     #    2.26  insn per cycle         
+       0.798991405 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:84831) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.avx2_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -176,27 +176,27 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProce
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.avx2_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.avx2_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 9.892973e-03
-Avg ME (F77/C++)    = 9.8929728161070551E-003
-Relative difference = 1.858823877057982e-08
+Avg ME (F77/C++)    = 9.8929728159608508E-003
+Relative difference = 1.8603017364363385e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512y_f_inl0_hrd0/check.exe -p 1 256 2 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=256)
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/512y+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 8.083345e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 8.085639e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 8.085639e+02                 )  sec^-1
-MeanMatrixElemValue         = ( 1.187188e-05 +- 9.826767e-06 )  GeV^-6
-TOTAL       :     0.658894 sec
-     1,877,804,985      cycles                           #    2.834 GHz                    
-     4,274,294,161      instructions                     #    2.28  insn per cycle         
-       0.663396758 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:80610) (512y:   46) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 7.613418e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.615510e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.615510e+02                 )  sec^-1
+MeanMatrixElemValue         = ( 1.187187e-05 +- 9.826763e-06 )  GeV^-6
+TOTAL       :     0.699716 sec
+     1,882,009,512      cycles                           #    2.675 GHz                    
+     4,259,439,384      instructions                     #    2.26  insn per cycle         
+       0.704552121 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:80038) (512y:   46) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512y_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -204,36 +204,36 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProce
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512y_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512y_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 9.892973e-03
-Avg ME (F77/C++)    = 9.8929728161070551E-003
-Relative difference = 1.858823877057982e-08
+Avg ME (F77/C++)    = 9.8929728159608508E-003
+Relative difference = 1.8603017364363385e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512z_f_inl0_hrd0/check.exe -p 1 256 2 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=256)
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/512z+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[16] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.030887e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.033160e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.033160e+02                 )  sec^-1
-MeanMatrixElemValue         = ( 1.187188e-05 +- 9.826771e-06 )  GeV^-6
-TOTAL       :     0.757777 sec
-     1,359,709,959      cycles                           #    1.785 GHz                    
-     2,159,410,769      instructions                     #    1.59  insn per cycle         
-       0.762416842 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2877) (512y:   49) (512z:79298)
+EvtsPerSec[Rmb+ME]     (23) = ( 6.688489e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.690546e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.690546e+02                 )  sec^-1
+MeanMatrixElemValue         = ( 1.187188e-05 +- 9.826770e-06 )  GeV^-6
+TOTAL       :     0.795848 sec
+     1,355,819,871      cycles                           #    1.696 GHz                    
+     2,148,215,879      instructions                     #    1.58  insn per cycle         
+       0.800761416 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2820) (512y:   44) (512z:78510)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512z_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512z_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512z_f_inl0_hrd0/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 9.892981e-03
-Avg ME (F77/C++)    = 9.8929811982958280E-003
-Relative difference = 2.0044092642523172e-08
+Avg ME (C++/C++)    = 9.892980e-03
+Avg ME (F77/C++)    = 9.8929802670331551E-003
+Relative difference = 2.699218597469717e-08
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_ggttggg_mad/log_ggttggg_mad_f_inl0_hrd1.txt b/epochX/cudacpp/tput/logs_ggttggg_mad/log_ggttggg_mad_f_inl0_hrd1.txt
index a6a593e43c..90b9187b98 100644
--- a/epochX/cudacpp/tput/logs_ggttggg_mad/log_ggttggg_mad_f_inl0_hrd1.txt
+++ b/epochX/cudacpp/tput/logs_ggttggg_mad/log_ggttggg_mad_f_inl0_hrd1.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_f_inl0_hrd1'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg'
 
-DATE: 2023-11-24_14:44:27
+DATE: 2024-01-30_05:09:16
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_f_inl0_hrd1/gcheck.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 6.747119e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 6.748060e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 6.748340e+02                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.764318e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.765250e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.765666e+02                 )  sec^-1
 MeanMatrixElemValue         = ( 1.186984e-05 +- 9.824899e-06 )  GeV^-6
-TOTAL       :     1.659570 sec
-     5,920,808,418      cycles                           #    3.063 GHz                    
-    12,044,538,664      instructions                     #    2.03  insn per cycle         
-       1.991771483 seconds time elapsed
+TOTAL       :     1.705167 sec
+     5,556,067,435      cycles                           #    2.852 GHz                    
+    10,985,634,618      instructions                     #    1.98  insn per cycle         
+       2.060310498 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_f_inl0_hrd1/gcheck.exe -p 1 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 255
@@ -65,17 +65,17 @@ WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 .........................................................................
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_f_inl0_hrd1/gcheck.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 2.325940e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.326739e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.326842e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.344230e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.345038e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.345205e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 1.856829e-04 +- 8.333435e-05 )  GeV^-6
-TOTAL       :     1.907449 sec
-     6,693,274,890      cycles                           #    3.061 GHz                    
-    12,994,667,842      instructions                     #    1.94  insn per cycle         
-       2.242820084 seconds time elapsed
+TOTAL       :     1.928465 sec
+     6,365,812,176      cycles                           #    2.870 GHz                    
+    12,742,048,160      instructions                     #    2.00  insn per cycle         
+       2.275067290 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_f_inl0_hrd1/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_f_inl0_hrd1/fgcheck.exe 2 64 2
@@ -86,20 +86,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_f_inl0_hrd1/check.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 9.023898e+01                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 9.024201e+01                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 9.024201e+01                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 8.700294e+01                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 8.700564e+01                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 8.700564e+01                 )  sec^-1
 MeanMatrixElemValue         = ( 1.187013e-05 +- 9.825040e-06 )  GeV^-6
-TOTAL       :     5.861507 sec
-    17,865,192,522      cycles                           #    3.046 GHz                    
-    53,579,122,611      instructions                     #    3.00  insn per cycle         
-       5.865565703 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:20207) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     6.077926 sec
+    17,558,502,709      cycles                           #    2.889 GHz                    
+    51,759,109,121      instructions                     #    2.95  insn per cycle         
+       6.085026833 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:27678) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_f_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -107,26 +107,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProce
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_f_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_f_inl0_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 9.847961e-03
-Avg ME (F77/C++)    = 9.8479612087582491E-003
-Relative difference = 2.1198118933954545e-08
+Avg ME (F77/C++)    = 9.8479612087313262E-003
+Relative difference = 2.1195385077844924e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.sse4_f_inl0_hrd1/check.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.621823e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.622258e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.622258e+02                 )  sec^-1
-MeanMatrixElemValue         = ( 1.187013e-05 +- 9.825037e-06 )  GeV^-6
-TOTAL       :     1.463287 sec
-     4,558,245,342      cycles                           #    3.108 GHz                    
-    13,754,988,539      instructions                     #    3.02  insn per cycle         
-       1.467375257 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:96606) (avx2:    0) (512y:    0) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 3.376771e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.377174e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.377174e+02                 )  sec^-1
+MeanMatrixElemValue         = ( 1.187013e-05 +- 9.825038e-06 )  GeV^-6
+TOTAL       :     1.572289 sec
+     4,548,603,521      cycles                           #    2.891 GHz                    
+    13,758,604,883      instructions                     #    3.02  insn per cycle         
+       1.583710945 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:97728) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.sse4_f_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -134,26 +134,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProce
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.sse4_f_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.sse4_f_inl0_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 9.847955e-03
-Avg ME (F77/C++)    = 9.8479546896225560E-003
-Relative difference = 3.151694379513441e-08
+Avg ME (F77/C++)    = 9.8479546894727158E-003
+Relative difference = 3.1532159158088894e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.avx2_f_inl0_hrd1/check.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.105620e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.107481e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.107481e+02                 )  sec^-1
-MeanMatrixElemValue         = ( 1.187188e-05 +- 9.826767e-06 )  GeV^-6
-TOTAL       :     0.747732 sec
-     2,126,324,616      cycles                           #    2.830 GHz                    
-     4,818,424,886      instructions                     #    2.27  insn per cycle         
-       0.751960174 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:85359) (512y:    0) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 6.592179e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.593820e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.593820e+02                 )  sec^-1
+MeanMatrixElemValue         = ( 1.187187e-05 +- 9.826763e-06 )  GeV^-6
+TOTAL       :     0.808909 sec
+     2,140,416,404      cycles                           #    2.637 GHz                    
+     4,826,824,873      instructions                     #    2.26  insn per cycle         
+       0.906681144 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:84793) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.avx2_f_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -161,26 +161,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProce
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.avx2_f_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.avx2_f_inl0_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 9.892973e-03
-Avg ME (F77/C++)    = 9.8929728161070967E-003
-Relative difference = 1.8588234562202478e-08
+Avg ME (F77/C++)    = 9.8929728159608508E-003
+Relative difference = 1.8603017364363385e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512y_f_inl0_hrd1/check.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 8.303282e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 8.305517e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 8.305517e+02                 )  sec^-1
-MeanMatrixElemValue         = ( 1.187188e-05 +- 9.826767e-06 )  GeV^-6
-TOTAL       :     0.641081 sec
-     1,865,331,641      cycles                           #    2.893 GHz                    
-     4,275,221,232      instructions                     #    2.29  insn per cycle         
-       0.645264707 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:81075) (512y:   26) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 7.677326e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.679741e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.679741e+02                 )  sec^-1
+MeanMatrixElemValue         = ( 1.187187e-05 +- 9.826763e-06 )  GeV^-6
+TOTAL       :     0.695609 sec
+     1,868,752,206      cycles                           #    2.678 GHz                    
+     4,259,067,854      instructions                     #    2.28  insn per cycle         
+       0.708960929 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:79978) (512y:   46) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512y_f_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -188,35 +188,35 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProce
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512y_f_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512y_f_inl0_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 9.892973e-03
-Avg ME (F77/C++)    = 9.8929728161070967E-003
-Relative difference = 1.8588234562202478e-08
+Avg ME (F77/C++)    = 9.8929728159608508E-003
+Relative difference = 1.8603017364363385e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512z_f_inl0_hrd1/check.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[16] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.530995e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.533488e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.533488e+02                 )  sec^-1
-MeanMatrixElemValue         = ( 1.187188e-05 +- 9.826771e-06 )  GeV^-6
-TOTAL       :     0.706631 sec
-     1,357,570,001      cycles                           #    1.912 GHz                    
-     2,164,337,331      instructions                     #    1.59  insn per cycle         
-       0.710707193 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3474) (512y:   34) (512z:79492)
+EvtsPerSec[Rmb+ME]     (23) = ( 6.775075e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.777182e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.777182e+02                 )  sec^-1
+MeanMatrixElemValue         = ( 1.187188e-05 +- 9.826770e-06 )  GeV^-6
+TOTAL       :     0.788693 sec
+     1,354,650,321      cycles                           #    1.715 GHz                    
+     2,148,091,187      instructions                     #    1.59  insn per cycle         
+       0.801177717 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2776) (512y:   44) (512z:78501)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512z_f_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512z_f_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512z_f_inl0_hrd1/fcheck.exe 2 64 2
-Avg ME (C++/C++)    = 9.892981e-03
-Avg ME (F77/C++)    = 9.8929811982955140E-003
-Relative difference = 2.0044060904369713e-08
+Avg ME (C++/C++)    = 9.892980e-03
+Avg ME (F77/C++)    = 9.8929802670331551E-003
+Relative difference = 2.699218597469717e-08
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_ggttggg_mad/log_ggttggg_mad_m_inl0_hrd0.txt b/epochX/cudacpp/tput/logs_ggttggg_mad/log_ggttggg_mad_m_inl0_hrd0.txt
index 73f6f01a0a..4eda45e114 100644
--- a/epochX/cudacpp/tput/logs_ggttggg_mad/log_ggttggg_mad_m_inl0_hrd0.txt
+++ b/epochX/cudacpp/tput/logs_ggttggg_mad/log_ggttggg_mad_m_inl0_hrd0.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_m_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg'
 
-DATE: 2023-11-24_14:45:14
+DATE: 2024-01-30_05:10:05
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_m_inl0_hrd0/gcheck.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 4.691784e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.692281e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.692424e+02                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.692959e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.693612e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.693848e+02                 )  sec^-1
 MeanMatrixElemValue         = ( 1.187066e-05 +- 9.825548e-06 )  GeV^-6
-TOTAL       :     2.167507 sec
-     7,673,265,360      cycles                           #    3.083 GHz                    
-    16,847,311,173      instructions                     #    2.20  insn per cycle         
-       2.544843357 seconds time elapsed
+TOTAL       :     2.179406 sec
+     7,155,207,889      cycles                           #    2.861 GHz                    
+    14,615,335,571      instructions                     #    2.04  insn per cycle         
+       2.559881855 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_m_inl0_hrd0/gcheck.exe -p 1 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 255
@@ -65,17 +65,17 @@ WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 .........................................................................
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_m_inl0_hrd0/gcheck.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 1.109611e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.109929e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.109960e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.111470e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.111782e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.111825e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 1.856249e-04 +- 8.329951e-05 )  GeV^-6
-TOTAL       :     3.399118 sec
-    11,512,402,509      cycles                           #    3.091 GHz                    
-    25,362,588,706      instructions                     #    2.20  insn per cycle         
-       3.780006761 seconds time elapsed
+TOTAL       :     3.413893 sec
+    10,746,707,284      cycles                           #    2.875 GHz                    
+    23,674,149,917      instructions                     #    2.20  insn per cycle         
+       3.796927749 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_m_inl0_hrd0/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_m_inl0_hrd0/fgcheck.exe 2 64 2
@@ -86,20 +86,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_m_inl0_hrd0/check.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 7.998480e+01                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.998694e+01                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.998694e+01                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 7.884803e+01                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.885022e+01                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.885022e+01                 )  sec^-1
 MeanMatrixElemValue         = ( 1.187066e-05 +- 9.825549e-06 )  GeV^-6
-TOTAL       :     6.606791 sec
-    19,173,459,975      cycles                           #    2.901 GHz                    
-    54,151,769,179      instructions                     #    2.82  insn per cycle         
-       6.611030036 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:32067) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     6.704107 sec
+    19,257,123,030      cycles                           #    2.874 GHz                    
+    55,394,447,460      instructions                     #    2.88  insn per cycle         
+       6.709385430 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:44898) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_m_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -113,20 +113,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.sse4_m_inl0_hrd0/check.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.627384e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.627470e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.627470e+02                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.509946e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.510039e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.510039e+02                 )  sec^-1
 MeanMatrixElemValue         = ( 1.187066e-05 +- 9.825548e-06 )  GeV^-6
-TOTAL       :     3.249018 sec
-     9,449,828,715      cycles                           #    2.906 GHz                    
-    26,158,243,948      instructions                     #    2.77  insn per cycle         
-       3.253176983 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:96007) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     3.502177 sec
+     9,384,694,038      cycles                           #    2.677 GHz                    
+    25,874,743,625      instructions                     #    2.76  insn per cycle         
+       3.507349921 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:96804) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.sse4_m_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -140,20 +140,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.avx2_m_inl0_hrd0/check.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.818750e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.819217e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.819217e+02                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.557555e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.558062e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.558062e+02                 )  sec^-1
 MeanMatrixElemValue         = ( 1.187066e-05 +- 9.825548e-06 )  GeV^-6
-TOTAL       :     1.387571 sec
-     4,051,003,122      cycles                           #    2.912 GHz                    
-     9,227,088,631      instructions                     #    2.28  insn per cycle         
-       1.391733244 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:84155) (512y:    0) (512z:    0)
+TOTAL       :     1.490188 sec
+     4,000,749,453      cycles                           #    2.678 GHz                    
+     9,119,038,902      instructions                     #    2.28  insn per cycle         
+       1.495279789 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:83820) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.avx2_m_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -167,20 +167,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512y_m_inl0_hrd0/check.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 4.365519e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.366141e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.366141e+02                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.057405e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.058069e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.058069e+02                 )  sec^-1
 MeanMatrixElemValue         = ( 1.187066e-05 +- 9.825548e-06 )  GeV^-6
-TOTAL       :     1.214524 sec
-     3,540,300,515      cycles                           #    2.907 GHz                    
-     8,174,532,807      instructions                     #    2.31  insn per cycle         
-       1.218602300 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:79844) (512y:   79) (512z:    0)
+TOTAL       :     1.307627 sec
+     3,513,640,690      cycles                           #    2.679 GHz                    
+     8,029,011,845      instructions                     #    2.29  insn per cycle         
+       1.312711431 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:79028) (512y:   70) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512y_m_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -194,20 +194,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512z_m_inl0_hrd0/check.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.824767e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.825436e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.825436e+02                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.350506e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.351010e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.351010e+02                 )  sec^-1
 MeanMatrixElemValue         = ( 1.187066e-05 +- 9.825548e-06 )  GeV^-6
-TOTAL       :     1.386311 sec
-     2,658,660,002      cycles                           #    1.913 GHz                    
-     4,154,162,009      instructions                     #    1.56  insn per cycle         
-       1.390371345 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2046) (512y:   93) (512z:78760)
+TOTAL       :     1.581908 sec
+     2,606,864,065      cycles                           #    1.673 GHz                    
+     4,077,382,976      instructions                     #    1.56  insn per cycle         
+       1.587144818 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1903) (512y:   70) (512z:78042)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512z_m_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
diff --git a/epochX/cudacpp/tput/logs_ggttggg_mad/log_ggttggg_mad_m_inl0_hrd1.txt b/epochX/cudacpp/tput/logs_ggttggg_mad/log_ggttggg_mad_m_inl0_hrd1.txt
index d4fe0b979f..328b61834e 100644
--- a/epochX/cudacpp/tput/logs_ggttggg_mad/log_ggttggg_mad_m_inl0_hrd1.txt
+++ b/epochX/cudacpp/tput/logs_ggttggg_mad/log_ggttggg_mad_m_inl0_hrd1.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_m_inl0_hrd1'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg'
 
-DATE: 2023-11-24_14:46:15
+DATE: 2024-01-30_05:11:08
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_m_inl0_hrd1/gcheck.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 4.682422e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.682989e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.683143e+02                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.684370e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.684951e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.685153e+02                 )  sec^-1
 MeanMatrixElemValue         = ( 1.187066e-05 +- 9.825548e-06 )  GeV^-6
-TOTAL       :     2.171969 sec
-     7,673,087,653      cycles                           #    3.078 GHz                    
-    17,236,869,851      instructions                     #    2.25  insn per cycle         
-       2.550325638 seconds time elapsed
+TOTAL       :     2.181079 sec
+     7,148,088,261      cycles                           #    2.853 GHz                    
+    14,239,530,947      instructions                     #    1.99  insn per cycle         
+       2.562146879 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_m_inl0_hrd1/gcheck.exe -p 1 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 255
@@ -65,17 +65,17 @@ WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 .........................................................................
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_m_inl0_hrd1/gcheck.exe -p 64 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGGG_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 1.106682e+04                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.106997e+04                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.107029e+04                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.111591e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.111914e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.111956e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 1.856249e-04 +- 8.329951e-05 )  GeV^-6
-TOTAL       :     3.401616 sec
-    11,548,737,077      cycles                           #    3.091 GHz                    
-    25,085,297,801      instructions                     #    2.17  insn per cycle         
-       3.792911760 seconds time elapsed
+TOTAL       :     3.413150 sec
+    10,755,861,454      cycles                           #    2.876 GHz                    
+    23,518,245,564      instructions                     #    2.19  insn per cycle         
+       3.796500341 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_m_inl0_hrd1/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_m_inl0_hrd1/fgcheck.exe 2 64 2
@@ -86,20 +86,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_m_inl0_hrd1/check.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 8.474149e+01                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 8.474398e+01                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 8.474398e+01                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 7.912565e+01                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.912803e+01                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.912803e+01                 )  sec^-1
 MeanMatrixElemValue         = ( 1.187066e-05 +- 9.825549e-06 )  GeV^-6
-TOTAL       :     6.238723 sec
-    19,082,506,717      cycles                           #    3.058 GHz                    
-    54,152,929,714      instructions                     #    2.84  insn per cycle         
-       6.242654091 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:32244) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     6.680088 sec
+    19,228,329,737      cycles                           #    2.877 GHz                    
+    55,419,296,273      instructions                     #    2.88  insn per cycle         
+       6.685533383 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:44806) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.none_m_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -113,20 +113,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.sse4_m_inl0_hrd1/check.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.640779e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.640877e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.640877e+02                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.515454e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.515537e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.515537e+02                 )  sec^-1
 MeanMatrixElemValue         = ( 1.187066e-05 +- 9.825548e-06 )  GeV^-6
-TOTAL       :     3.222880 sec
-     9,391,551,742      cycles                           #    2.911 GHz                    
-    26,077,333,462      instructions                     #    2.78  insn per cycle         
-       3.227016085 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:95901) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     3.490021 sec
+     9,348,051,078      cycles                           #    2.676 GHz                    
+    25,823,110,897      instructions                     #    2.76  insn per cycle         
+       3.495053121 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:96765) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.sse4_m_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -140,20 +140,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.avx2_m_inl0_hrd1/check.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.773302e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.773766e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.773766e+02                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.556805e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.557285e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.557285e+02                 )  sec^-1
 MeanMatrixElemValue         = ( 1.187066e-05 +- 9.825548e-06 )  GeV^-6
-TOTAL       :     1.403808 sec
-     4,073,842,959      cycles                           #    2.895 GHz                    
-     9,213,198,457      instructions                     #    2.26  insn per cycle         
-       1.407971002 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:83776) (512y:    0) (512z:    0)
+TOTAL       :     1.490221 sec
+     4,003,060,439      cycles                           #    2.680 GHz                    
+     9,098,942,911      instructions                     #    2.27  insn per cycle         
+       1.495311791 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:83378) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.avx2_m_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -167,20 +167,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512y_m_inl0_hrd1/check.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 4.338283e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.338969e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.338969e+02                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.083203e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.083821e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.083821e+02                 )  sec^-1
 MeanMatrixElemValue         = ( 1.187066e-05 +- 9.825548e-06 )  GeV^-6
-TOTAL       :     1.222714 sec
-     3,541,596,074      cycles                           #    2.889 GHz                    
-     8,167,599,536      instructions                     #    2.31  insn per cycle         
-       1.226930952 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2:79373) (512y:  229) (512z:    0)
+TOTAL       :     1.299137 sec
+     3,488,850,980      cycles                           #    2.678 GHz                    
+     8,010,474,997      instructions                     #    2.30  insn per cycle         
+       1.304443015 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2:78540) (512y:   70) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512y_m_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -194,20 +194,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512z_m_inl0_hrd1/check.exe -p 1 256 2 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GG_TTXGGG_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.888294e+02                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.888912e+02                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.888912e+02                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.440905e+02                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.441442e+02                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.441442e+02                 )  sec^-1
 MeanMatrixElemValue         = ( 1.187066e-05 +- 9.825548e-06 )  GeV^-6
-TOTAL       :     1.363883 sec
-     2,618,199,272      cycles                           #    1.915 GHz                    
-     4,152,923,340      instructions                     #    1.59  insn per cycle         
-       1.368029549 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1493) (512y:  175) (512z:78776)
+TOTAL       :     1.541232 sec
+     2,598,862,718      cycles                           #    1.682 GHz                    
+     4,064,975,706      instructions                     #    1.56  insn per cycle         
+       1.546247038 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1420) (512y:   70) (512z:78026)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_ttggg.mad/SubProcesses/P1_gg_ttxggg/build.512z_m_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
diff --git a/epochX/cudacpp/tput/logs_gqttq_mad/log_gqttq_mad_d_inl0_hrd0.txt b/epochX/cudacpp/tput/logs_gqttq_mad/log_gqttq_mad_d_inl0_hrd0.txt
index 293e7e906a..5667ce458e 100644
--- a/epochX/cudacpp/tput/logs_gqttq_mad/log_gqttq_mad_d_inl0_hrd0.txt
+++ b/epochX/cudacpp/tput/logs_gqttq_mad/log_gqttq_mad_d_inl0_hrd0.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux'
 
-DATE: 2023-11-24_14:39:50
+DATE: 2024-01-30_05:04:25
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_d_inl0_hrd0/gcheck.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 2.679868e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 5.317378e+07                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 5.685862e+07                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.650880e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 5.304183e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.677107e+07                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018083e+01 +- 1.429474e+01 )  GeV^-2
-TOTAL       :     0.442326 sec
-     1,985,114,674      cycles                           #    3.012 GHz                    
-     2,801,190,832      instructions                     #    1.41  insn per cycle         
-       0.731797225 seconds time elapsed
+TOTAL       :     0.453043 sec
+     1,889,864,608      cycles                           #    2.824 GHz                    
+     2,684,689,341      instructions                     #    1.42  insn per cycle         
+       0.749142975 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_d_inl0_hrd0/gcheck.exe -p 64 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 255
@@ -65,17 +65,17 @@ WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 .........................................................................
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_d_inl0_hrd0/gcheck.exe -p 2048 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 3.245024e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.095249e+07                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.522546e+07                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.266493e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.111955e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.526543e+07                 )  sec^-1
 MeanMatrixElemValue         = ( 2.602505e+02 +- 2.116328e+02 )  GeV^-2
-TOTAL       :     0.523645 sec
-     2,298,323,986      cycles                           #    3.021 GHz                    
-     3,281,368,879      instructions                     #    1.43  insn per cycle         
-       0.818048829 seconds time elapsed
+TOTAL       :     0.538455 sec
+     2,216,644,376      cycles                           #    2.828 GHz                    
+     3,102,394,165      instructions                     #    1.40  insn per cycle         
+       0.841378524 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_d_inl0_hrd0/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_d_inl0_hrd0/fgcheck.exe 2 64 2
@@ -86,20 +86,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_d_inl0_hrd0/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.100295e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.122827e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.122827e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 9.822300e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.003024e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.003024e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018083e+01 +- 1.429474e+01 )  GeV^-2
-TOTAL       :     1.510825 sec
-     4,703,874,541      cycles                           #    3.106 GHz                    
-    13,462,408,396      instructions                     #    2.86  insn per cycle         
-       1.520640560 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  860) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     1.695633 sec
+     4,892,910,077      cycles                           #    2.883 GHz                    
+    13,801,787,359      instructions                     #    2.82  insn per cycle         
+       1.705964185 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 1166) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -107,26 +107,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.424749e-01
-Avg ME (F77/C++)    = 0.14247482467499484
-Relative difference = 5.286896509487005e-07
+Avg ME (F77/C++)    = 0.14247482467499481
+Relative difference = 5.286896511435107e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.sse4_d_inl0_hrd0/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.958580e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.032826e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.032826e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.896648e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.972375e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.972375e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018083e+01 +- 1.429474e+01 )  GeV^-2
-TOTAL       :     0.858107 sec
-     2,618,495,660      cycles                           #    3.035 GHz                    
-     7,553,259,233      instructions                     #    2.88  insn per cycle         
-       0.870670449 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 3099) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     0.887238 sec
+     2,571,261,116      cycles                           #    2.883 GHz                    
+     7,401,200,610      instructions                     #    2.88  insn per cycle         
+       0.906229412 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 2895) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.sse4_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -134,26 +134,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.sse4_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.sse4_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.424749e-01
-Avg ME (F77/C++)    = 0.14247482467499478
-Relative difference = 5.28689651338321e-07
+Avg ME (F77/C++)    = 0.14247482467499475
+Relative difference = 5.286896515331313e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.avx2_d_inl0_hrd0/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.413914e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.637783e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.637783e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.154928e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.367723e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.367723e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018083e+01 +- 1.429474e+01 )  GeV^-2
-TOTAL       :     0.501429 sec
-     1,476,425,844      cycles                           #    2.921 GHz                    
-     3,120,626,738      instructions                     #    2.11  insn per cycle         
-       0.516624124 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2917) (512y:    0) (512z:    0)
+TOTAL       :     0.543357 sec
+     1,480,133,709      cycles                           #    2.701 GHz                    
+     3,136,765,286      instructions                     #    2.12  insn per cycle         
+       0.561297241 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2890) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.avx2_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -167,20 +167,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512y_d_inl0_hrd0/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.779052e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.050087e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.050087e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.571891e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.844626e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.844626e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018083e+01 +- 1.429474e+01 )  GeV^-2
-TOTAL       :     0.454903 sec
-     1,341,062,737      cycles                           #    2.922 GHz                    
-     2,982,346,892      instructions                     #    2.22  insn per cycle         
-       0.471560702 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2694) (512y:  104) (512z:    0)
+TOTAL       :     0.482249 sec
+     1,314,348,676      cycles                           #    2.699 GHz                    
+     2,923,288,921      instructions                     #    2.22  insn per cycle         
+       0.498803372 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2543) (512y:   93) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512y_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -194,20 +194,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512z_d_inl0_hrd0/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.570252e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.695598e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.695598e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.408041e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.532332e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.532332e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018083e+01 +- 1.429474e+01 )  GeV^-2
-TOTAL       :     0.661079 sec
-     1,328,110,043      cycles                           #    1.997 GHz                    
-     1,954,412,178      instructions                     #    1.47  insn per cycle         
-       0.674908188 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1372) (512y:  106) (512z: 2173)
+TOTAL       :     0.706292 sec
+     1,273,944,985      cycles                           #    1.792 GHz                    
+     1,900,262,296      instructions                     #    1.49  insn per cycle         
+       0.723222352 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1135) (512y:   62) (512z: 2165)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512z_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
diff --git a/epochX/cudacpp/tput/logs_gqttq_mad/log_gqttq_mad_d_inl0_hrd0_bridge.txt b/epochX/cudacpp/tput/logs_gqttq_mad/log_gqttq_mad_d_inl0_hrd0_bridge.txt
index d85e5ad544..7b59743406 100644
--- a/epochX/cudacpp/tput/logs_gqttq_mad/log_gqttq_mad_d_inl0_hrd0_bridge.txt
+++ b/epochX/cudacpp/tput/logs_gqttq_mad/log_gqttq_mad_d_inl0_hrd0_bridge.txt
@@ -41,7 +41,7 @@ CUDACPP_BUILDDIR='build.512z_d_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux'
 
-DATE: 2023-11-24_15:12:44
+DATE: 2024-01-30_05:48:29
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
@@ -51,17 +51,17 @@ WARNING! Bridge selected: cannot use RamboDevice, will use RamboHost
 WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
 WARNING! Instantiate device Bridge (nevt=16384, gpublocks=64, gputhreads=256, gpublocks*gputhreads=16384)
 WARNING! Set grid in Bridge (nevt=16384, gpublocks=64, gputhreads=256, gpublocks*gputhreads=16384)
-Process                     = SIGMA_SM_GUX_TTXUX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 3.546178e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.153075e+07                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.153075e+07                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.408359e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.101986e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.101986e+07                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018083e+01 +- 1.429474e+01 )  GeV^-2
-TOTAL       :     0.474399 sec
-     1,996,543,423      cycles                           #    2.914 GHz                    
-     2,976,710,186      instructions                     #    1.49  insn per cycle         
-       0.743100653 seconds time elapsed
+TOTAL       :     0.481872 sec
+     1,962,034,459      cycles                           #    2.824 GHz                    
+     2,925,170,965      instructions                     #    1.49  insn per cycle         
+       0.753942373 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_d_inl0_hrd0/gcheck.exe -p 64 256 1 --bridge
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Bridge selected: cannot use RamboDevice, will use RamboHost
@@ -77,17 +77,17 @@ WARNING! Bridge selected: cannot use RamboDevice, will use RamboHost
 WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
 WARNING! Instantiate device Bridge (nevt=524288, gpublocks=2048, gputhreads=256, gpublocks*gputhreads=524288)
 WARNING! Set grid in Bridge (nevt=524288, gpublocks=2048, gputhreads=256, gpublocks*gputhreads=524288)
-Process                     = SIGMA_SM_GUX_TTXUX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:DBL+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 3.225202e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.276966e+07                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.276966e+07                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.119182e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.257748e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.257748e+07                 )  sec^-1
 MeanMatrixElemValue         = ( 2.602505e+02 +- 2.116328e+02 )  GeV^-2
-TOTAL       :     0.757945 sec
-     2,951,779,199      cycles                           #    2.926 GHz                    
-     4,535,994,570      instructions                     #    1.54  insn per cycle         
-       1.067442306 seconds time elapsed
+TOTAL       :     0.770186 sec
+     2,924,566,680      cycles                           #    2.837 GHz                    
+     4,475,846,392      instructions                     #    1.53  insn per cycle         
+       1.089161093 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_d_inl0_hrd0/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_d_inl0_hrd0/fgcheck.exe 2 64 2
@@ -99,20 +99,20 @@ OK (relative difference <= 5E-3)
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_d_inl0_hrd0/check.exe -p 64 256 10 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=16384)
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.063685e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.086099e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.086099e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 9.824024e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.002926e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.002926e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018083e+01 +- 1.429474e+01 )  GeV^-2
-TOTAL       :     1.568160 sec
-     4,739,058,277      cycles                           #    3.015 GHz                    
-    13,467,544,042      instructions                     #    2.84  insn per cycle         
-       1.572567564 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  860) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     1.698586 sec
+     4,927,814,709      cycles                           #    2.894 GHz                    
+    13,806,118,322      instructions                     #    2.80  insn per cycle         
+       1.704123738 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 1166) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -120,27 +120,27 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.424749e-01
-Avg ME (F77/C++)    = 0.14247482467499484
-Relative difference = 5.286896509487005e-07
+Avg ME (F77/C++)    = 0.14247482467499481
+Relative difference = 5.286896511435107e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.sse4_d_inl0_hrd0/check.exe -p 64 256 10 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=16384)
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/sse4+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.915883e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.989942e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.989942e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.886173e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.963508e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.963508e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018083e+01 +- 1.429474e+01 )  GeV^-2
-TOTAL       :     0.884700 sec
-     2,669,440,783      cycles                           #    3.003 GHz                    
-     7,602,375,116      instructions                     #    2.85  insn per cycle         
-       0.889666740 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 3099) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     0.900867 sec
+     2,618,017,951      cycles                           #    2.892 GHz                    
+     7,450,102,141      instructions                     #    2.85  insn per cycle         
+       0.906367581 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 2895) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.sse4_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -148,27 +148,27 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.sse4_d_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.sse4_d_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.424749e-01
-Avg ME (F77/C++)    = 0.14247482467499478
-Relative difference = 5.28689651338321e-07
+Avg ME (F77/C++)    = 0.14247482467499475
+Relative difference = 5.286896515331313e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.avx2_d_inl0_hrd0/check.exe -p 64 256 10 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=16384)
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/avx2+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.224315e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.444185e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.444185e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.122916e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.345144e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.345144e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018083e+01 +- 1.429474e+01 )  GeV^-2
-TOTAL       :     0.539421 sec
-     1,532,896,663      cycles                           #    2.819 GHz                    
-     3,168,696,255      instructions                     #    2.07  insn per cycle         
-       0.544640536 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2917) (512y:    0) (512z:    0)
+TOTAL       :     0.557525 sec
+     1,528,674,468      cycles                           #    2.721 GHz                    
+     3,187,083,360      instructions                     #    2.08  insn per cycle         
+       0.563020024 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2890) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.avx2_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -183,20 +183,20 @@ OK (relative difference <= 5E-3)
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512y_d_inl0_hrd0/check.exe -p 64 256 10 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=16384)
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/512y+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.619993e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.881959e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.881959e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.528840e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.810605e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.810605e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018083e+01 +- 1.429474e+01 )  GeV^-2
-TOTAL       :     0.481010 sec
-     1,371,011,914      cycles                           #    2.828 GHz                    
-     3,030,736,025      instructions                     #    2.21  insn per cycle         
-       0.485465230 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2694) (512y:  104) (512z:    0)
+TOTAL       :     0.496872 sec
+     1,359,999,193      cycles                           #    2.712 GHz                    
+     2,973,904,476      instructions                     #    2.19  insn per cycle         
+       0.502643224 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2543) (512y:   93) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512y_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -211,20 +211,20 @@ OK (relative difference <= 5E-3)
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512z_d_inl0_hrd0/check.exe -p 64 256 10 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=16384)
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+BRDHST/512z+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.415824e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.530854e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.530854e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.332416e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.457397e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.457397e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018083e+01 +- 1.429474e+01 )  GeV^-2
-TOTAL       :     0.708748 sec
-     1,355,688,416      cycles                           #    1.903 GHz                    
-     1,991,539,116      instructions                     #    1.47  insn per cycle         
-       0.713346181 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1372) (512y:  106) (512z: 2173)
+TOTAL       :     0.738182 sec
+     1,327,509,915      cycles                           #    1.788 GHz                    
+     1,939,124,841      instructions                     #    1.46  insn per cycle         
+       0.743808066 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1135) (512y:   62) (512z: 2165)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512z_d_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
diff --git a/epochX/cudacpp/tput/logs_gqttq_mad/log_gqttq_mad_d_inl0_hrd1.txt b/epochX/cudacpp/tput/logs_gqttq_mad/log_gqttq_mad_d_inl0_hrd1.txt
index 46adce5ba7..4deacb88f2 100644
--- a/epochX/cudacpp/tput/logs_gqttq_mad/log_gqttq_mad_d_inl0_hrd1.txt
+++ b/epochX/cudacpp/tput/logs_gqttq_mad/log_gqttq_mad_d_inl0_hrd1.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_d_inl0_hrd1'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux'
 
-DATE: 2023-11-24_14:40:07
+DATE: 2024-01-30_05:04:44
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_d_inl0_hrd1/gcheck.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GUX_TTXUX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 2.664367e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 5.214901e+07                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 5.575772e+07                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.642894e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 5.200887e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.567165e+07                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018083e+01 +- 1.429474e+01 )  GeV^-2
-TOTAL       :     0.441294 sec
-     1,975,560,580      cycles                           #    3.010 GHz                    
-     2,812,494,086      instructions                     #    1.42  insn per cycle         
-       0.721240835 seconds time elapsed
+TOTAL       :     0.451244 sec
+     1,883,873,657      cycles                           #    2.821 GHz                    
+     2,671,262,226      instructions                     #    1.42  insn per cycle         
+       0.747348766 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_d_inl0_hrd1/gcheck.exe -p 64 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 255
@@ -65,17 +65,17 @@ WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 .........................................................................
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_d_inl0_hrd1/gcheck.exe -p 2048 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GUX_TTXUX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CUD:DBL+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 3.215656e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 6.999841e+07                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.407602e+07                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.228371e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.990649e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.395918e+07                 )  sec^-1
 MeanMatrixElemValue         = ( 2.602505e+02 +- 2.116328e+02 )  GeV^-2
-TOTAL       :     0.523997 sec
-     2,293,718,680      cycles                           #    3.009 GHz                    
-     3,268,283,620      instructions                     #    1.42  insn per cycle         
-       0.819195532 seconds time elapsed
+TOTAL       :     0.540844 sec
+     2,218,030,903      cycles                           #    2.829 GHz                    
+     3,154,136,532      instructions                     #    1.42  insn per cycle         
+       0.843504278 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_d_inl0_hrd1/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_d_inl0_hrd1/fgcheck.exe 2 64 2
@@ -86,20 +86,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_d_inl0_hrd1/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.074061e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.096252e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.096252e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 9.831536e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.003712e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.003712e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018083e+01 +- 1.429474e+01 )  GeV^-2
-TOTAL       :     1.547162 sec
-     4,714,230,759      cycles                           #    3.040 GHz                    
-    13,457,176,123      instructions                     #    2.85  insn per cycle         
-       1.553898636 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  849) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     1.690067 sec
+     4,884,610,591      cycles                           #    2.883 GHz                    
+    13,807,943,276      instructions                     #    2.83  insn per cycle         
+       1.700194727 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 1161) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_d_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -107,26 +107,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_d_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_d_inl0_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.424749e-01
-Avg ME (F77/C++)    = 0.14247482467499484
-Relative difference = 5.286896509487005e-07
+Avg ME (F77/C++)    = 0.14247482467499481
+Relative difference = 5.286896511435107e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.sse4_d_inl0_hrd1/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.993758e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.068724e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.068724e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.876876e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.953061e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.953061e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018083e+01 +- 1.429474e+01 )  GeV^-2
-TOTAL       :     0.842619 sec
-     2,622,041,010      cycles                           #    3.096 GHz                    
-     7,552,388,899      instructions                     #    2.88  insn per cycle         
-       0.855874832 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 3092) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     0.896918 sec
+     2,573,000,483      cycles                           #    2.854 GHz                    
+     7,407,132,972      instructions                     #    2.88  insn per cycle         
+       0.971480588 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 2892) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.sse4_d_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -134,26 +134,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.sse4_d_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.sse4_d_inl0_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.424749e-01
-Avg ME (F77/C++)    = 0.14247482467499478
-Relative difference = 5.28689651338321e-07
+Avg ME (F77/C++)    = 0.14247482467499475
+Relative difference = 5.286896515331313e-07
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.avx2_d_inl0_hrd1/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.386860e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.608787e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.608787e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.133331e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.344053e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.344053e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018083e+01 +- 1.429474e+01 )  GeV^-2
-TOTAL       :     0.505002 sec
-     1,482,553,157      cycles                           #    2.917 GHz                    
-     3,119,022,546      instructions                     #    2.10  insn per cycle         
-       0.516023532 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2900) (512y:    0) (512z:    0)
+TOTAL       :     0.546739 sec
+     1,486,856,812      cycles                           #    2.696 GHz                    
+     3,137,676,944      instructions                     #    2.11  insn per cycle         
+       0.563341736 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2875) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.avx2_d_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -167,20 +167,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512y_d_inl0_hrd1/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.786208e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.060798e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.060798e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.567673e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.839669e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.839669e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018083e+01 +- 1.429474e+01 )  GeV^-2
-TOTAL       :     0.453959 sec
-     1,338,340,378      cycles                           #    2.921 GHz                    
-     2,979,731,487      instructions                     #    2.23  insn per cycle         
-       0.465473907 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2670) (512y:  104) (512z:    0)
+TOTAL       :     0.482732 sec
+     1,314,507,412      cycles                           #    2.697 GHz                    
+     2,925,746,939      instructions                     #    2.23  insn per cycle         
+       0.501062508 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2527) (512y:   93) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512y_d_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -194,20 +194,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512z_d_inl0_hrd1/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:DBL+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = DOUBLE (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.592500e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.720014e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.720014e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.394430e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.516439e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.516439e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018083e+01 +- 1.429474e+01 )  GeV^-2
-TOTAL       :     0.655376 sec
-     1,326,766,373      cycles                           #    2.012 GHz                    
-     1,952,310,196      instructions                     #    1.47  insn per cycle         
-       0.669844706 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1348) (512y:  106) (512z: 2173)
+TOTAL       :     0.710071 sec
+     1,273,890,672      cycles                           #    1.782 GHz                    
+     1,899,944,131      instructions                     #    1.49  insn per cycle         
+       0.727352268 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1118) (512y:   62) (512z: 2165)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512z_d_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
diff --git a/epochX/cudacpp/tput/logs_gqttq_mad/log_gqttq_mad_f_inl0_hrd0.txt b/epochX/cudacpp/tput/logs_gqttq_mad/log_gqttq_mad_f_inl0_hrd0.txt
index 271711c5ee..1362a87ac8 100644
--- a/epochX/cudacpp/tput/logs_gqttq_mad/log_gqttq_mad_f_inl0_hrd0.txt
+++ b/epochX/cudacpp/tput/logs_gqttq_mad/log_gqttq_mad_f_inl0_hrd0.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_f_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux'
 
-DATE: 2023-11-24_14:40:25
+DATE: 2024-01-30_05:05:03
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_f_inl0_hrd0/gcheck.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 5.342241e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.209752e+08                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.345410e+08                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.327203e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.210086e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.349272e+08                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018174e+01 +- 1.429492e+01 )  GeV^-2
-TOTAL       :     0.439415 sec
-     1,959,210,302      cycles                           #    2.998 GHz                    
-     2,778,383,612      instructions                     #    1.42  insn per cycle         
-       0.742350357 seconds time elapsed
+TOTAL       :     0.446159 sec
+     1,908,363,704      cycles                           #    2.829 GHz                    
+     2,678,040,252      instructions                     #    1.40  insn per cycle         
+       0.749417997 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_f_inl0_hrd0/gcheck.exe -p 64 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 167
@@ -65,17 +65,17 @@ WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 .........................................................................
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_f_inl0_hrd0/gcheck.exe -p 2048 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 7.198954e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.802596e+08                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.953593e+08                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 7.267889e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.817352e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.969269e+08                 )  sec^-1
 MeanMatrixElemValue         = ( 2.571361e+02 +- 2.114021e+02 )  GeV^-2
-TOTAL       :     0.476253 sec
-     2,095,785,186      cycles                           #    2.994 GHz                    
-     2,918,085,144      instructions                     #    1.39  insn per cycle         
-       0.759290235 seconds time elapsed
+TOTAL       :     0.483640 sec
+     2,013,701,507      cycles                           #    2.833 GHz                    
+     2,869,047,503      instructions                     #    1.42  insn per cycle         
+       0.770237631 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_f_inl0_hrd0/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_f_inl0_hrd0/fgcheck.exe 2 64 2
@@ -86,20 +86,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_f_inl0_hrd0/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.156035e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.182076e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.182076e+05                 )  sec^-1
-MeanMatrixElemValue         = ( 2.018563e+01 +- 1.429903e+01 )  GeV^-2
-TOTAL       :     1.437672 sec
-     4,454,268,487      cycles                           #    3.090 GHz                    
-    13,048,356,445      instructions                     #    2.93  insn per cycle         
-       1.444257610 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  745) (avx2:    0) (512y:    0) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 1.109983e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.136218e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.136218e+05                 )  sec^-1
+MeanMatrixElemValue         = ( 2.018563e+01 +- 1.429902e+01 )  GeV^-2
+TOTAL       :     1.498447 sec
+     4,345,988,139      cycles                           #    2.893 GHz                    
+    12,596,967,872      instructions                     #    2.90  insn per cycle         
+       1.511882134 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  773) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -107,26 +107,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.424686e-01
-Avg ME (F77/C++)    = 0.14246857540270419
-Relative difference = 1.7265064590569047e-07
+Avg ME (F77/C++)    = 0.14246860569653919
+Relative difference = 3.998452420257791e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.sse4_f_inl0_hrd0/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.036696e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.233511e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.233511e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.116392e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.330955e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.330955e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018563e+01 +- 1.429902e+01 )  GeV^-2
-TOTAL       :     0.559663 sec
-     1,702,113,504      cycles                           #    3.019 GHz                    
-     4,513,068,480      instructions                     #    2.65  insn per cycle         
-       0.574904754 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 3600) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     0.547469 sec
+     1,595,191,710      cycles                           #    2.889 GHz                    
+     4,246,785,925      instructions                     #    2.66  insn per cycle         
+       0.566121323 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 3265) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.sse4_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -134,26 +134,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.sse4_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.sse4_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.424686e-01
-Avg ME (F77/C++)    = 0.14246859631675157
-Relative difference = 2.5853054135974944e-08
+Avg ME (F77/C++)    = 0.14246860808920836
+Relative difference = 5.677888572434963e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.avx2_f_inl0_hrd0/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 6.119769e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 6.899660e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 6.899660e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.705122e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.431194e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.431194e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018828e+01 +- 1.429922e+01 )  GeV^-2
-TOTAL       :     0.287512 sec
-       851,994,706      cycles                           #    2.923 GHz                    
-     1,897,184,726      instructions                     #    2.23  insn per cycle         
-       0.299631508 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3491) (512y:    0) (512z:    0)
+TOTAL       :     0.309154 sec
+       853,106,357      cycles                           #    2.719 GHz                    
+     1,916,236,758      instructions                     #    2.25  insn per cycle         
+       0.322202646 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3488) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.avx2_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -161,26 +161,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.avx2_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.avx2_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.424749e-01
-Avg ME (F77/C++)    = 0.14247489318272599
-Relative difference = 4.784894739577799e-08
+Avg ME (F77/C++)    = 0.14247490815036912
+Relative difference = 5.7205649062398515e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512y_f_inl0_hrd0/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 6.144841e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 6.980131e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 6.980131e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.291153e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.186493e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.186493e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018828e+01 +- 1.429922e+01 )  GeV^-2
-TOTAL       :     0.287863 sec
-       803,737,999      cycles                           #    2.749 GHz                    
-     1,820,191,781      instructions                     #    2.26  insn per cycle         
-       0.309536059 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3335) (512y:   22) (512z:    0)
+TOTAL       :     0.282114 sec
+       781,605,305      cycles                           #    2.726 GHz                    
+     1,797,850,243      instructions                     #    2.30  insn per cycle         
+       0.301017972 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3186) (512y:   15) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512y_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -188,26 +188,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512y_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512y_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.424749e-01
-Avg ME (F77/C++)    = 0.14247489318272599
-Relative difference = 4.784894739577799e-08
+Avg ME (F77/C++)    = 0.14247490815036912
+Relative difference = 5.7205649062398515e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512z_f_inl0_hrd0/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[16] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 4.909283e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 5.426697e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 5.426697e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.544342e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.998908e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.998908e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018829e+01 +- 1.429922e+01 )  GeV^-2
-TOTAL       :     0.356091 sec
-       735,639,321      cycles                           #    2.041 GHz                    
-     1,305,394,838      instructions                     #    1.77  insn per cycle         
-       0.370547154 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1968) (512y:   32) (512z: 2383)
+TOTAL       :     0.384301 sec
+       720,859,118      cycles                           #    1.854 GHz                    
+     1,288,039,773      instructions                     #    1.79  insn per cycle         
+       0.402338897 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1730) (512y:   24) (512z: 2387)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512z_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -215,8 +215,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512z_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512z_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.424749e-01
-Avg ME (F77/C++)    = 0.14247489383243206
-Relative difference = 4.32888033512879e-08
+Avg ME (F77/C++)    = 0.14247490450137867
+Relative difference = 3.159418737238044e-08
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_gqttq_mad/log_gqttq_mad_f_inl0_hrd0_bridge.txt b/epochX/cudacpp/tput/logs_gqttq_mad/log_gqttq_mad_f_inl0_hrd0_bridge.txt
index 9afdfb410c..8cb59221d4 100644
--- a/epochX/cudacpp/tput/logs_gqttq_mad/log_gqttq_mad_f_inl0_hrd0_bridge.txt
+++ b/epochX/cudacpp/tput/logs_gqttq_mad/log_gqttq_mad_f_inl0_hrd0_bridge.txt
@@ -41,7 +41,7 @@ CUDACPP_BUILDDIR='build.512z_f_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux'
 
-DATE: 2023-11-24_15:13:02
+DATE: 2024-01-30_05:48:48
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
@@ -51,17 +51,17 @@ WARNING! Bridge selected: cannot use RamboDevice, will use RamboHost
 WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
 WARNING! Instantiate device Bridge (nevt=16384, gpublocks=64, gputhreads=256, gpublocks*gputhreads=16384)
 WARNING! Set grid in Bridge (nevt=16384, gpublocks=64, gputhreads=256, gpublocks*gputhreads=16384)
-Process                     = SIGMA_SM_GUX_TTXUX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 5.454543e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.985559e+07                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.985559e+07                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.444132e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.000397e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.000397e+07                 )  sec^-1
 MeanMatrixElemValue         = ( 2.017654e+01 +- 1.429184e+01 )  GeV^-2
-TOTAL       :     0.454607 sec
-     1,923,330,877      cycles                           #    2.893 GHz                    
-     2,869,826,397      instructions                     #    1.49  insn per cycle         
-       0.721979990 seconds time elapsed
+TOTAL       :     0.460393 sec
+     1,902,959,760      cycles                           #    2.835 GHz                    
+     2,813,040,217      instructions                     #    1.48  insn per cycle         
+       0.731144371 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_f_inl0_hrd0/gcheck.exe -p 64 256 1 --bridge
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Bridge selected: cannot use RamboDevice, will use RamboHost
@@ -77,17 +77,17 @@ WARNING! Bridge selected: cannot use RamboDevice, will use RamboHost
 WARNING! RamboHost selected: cannot use CurandDevice, will use CurandHost
 WARNING! Instantiate device Bridge (nevt=524288, gpublocks=2048, gputhreads=256, gpublocks*gputhreads=524288)
 WARNING! Set grid in Bridge (nevt=524288, gpublocks=2048, gputhreads=256, gpublocks*gputhreads=524288)
-Process                     = SIGMA_SM_GUX_TTXUX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:FLT+THX:CURHST+RMBHST+BRDDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 5.025147e+06                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.561466e+07                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.561466e+07                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.962207e+06                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.533260e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.533260e+07                 )  sec^-1
 MeanMatrixElemValue         = ( 2.609942e+02 +- 2.115590e+02 )  GeV^-2
-TOTAL       :     0.624986 sec
-     2,513,731,167      cycles                           #    2.923 GHz                    
-     3,800,964,888      instructions                     #    1.51  insn per cycle         
-       0.917368993 seconds time elapsed
+TOTAL       :     0.631701 sec
+     2,471,933,418      cycles                           #    2.836 GHz                    
+     3,725,494,141      instructions                     #    1.51  insn per cycle         
+       0.929474422 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_f_inl0_hrd0/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_f_inl0_hrd0/fgcheck.exe 2 64 2
@@ -99,20 +99,20 @@ OK (relative difference <= 5E-3)
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_f_inl0_hrd0/check.exe -p 64 256 10 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=16384)
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.088101e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.112561e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.112561e+05                 )  sec^-1
-MeanMatrixElemValue         = ( 2.018563e+01 +- 1.429903e+01 )  GeV^-2
-TOTAL       :     1.529985 sec
-     4,473,562,903      cycles                           #    2.917 GHz                    
-    13,053,120,792      instructions                     #    2.92  insn per cycle         
-       1.534351278 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  745) (avx2:    0) (512y:    0) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 1.095228e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.121318e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.121318e+05                 )  sec^-1
+MeanMatrixElemValue         = ( 2.018563e+01 +- 1.429902e+01 )  GeV^-2
+TOTAL       :     1.522701 sec
+     4,367,827,655      cycles                           #    2.862 GHz                    
+    12,601,331,452      instructions                     #    2.89  insn per cycle         
+       1.527862957 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  773) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -120,27 +120,27 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.424686e-01
-Avg ME (F77/C++)    = 0.14246857540270419
-Relative difference = 1.7265064590569047e-07
+Avg ME (F77/C++)    = 0.14246860569653919
+Relative difference = 3.998452420257791e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.sse4_f_inl0_hrd0/check.exe -p 64 256 10 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=16384)
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/sse4+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.988941e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.180615e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.180615e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.075499e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.292736e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.292736e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018563e+01 +- 1.429902e+01 )  GeV^-2
-TOTAL       :     0.573010 sec
-     1,718,819,301      cycles                           #    2.980 GHz                    
-     4,560,439,401      instructions                     #    2.65  insn per cycle         
-       0.577426659 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 3600) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     0.559885 sec
+     1,623,222,211      cycles                           #    2.878 GHz                    
+     4,293,732,841      instructions                     #    2.65  insn per cycle         
+       0.565168184 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 3265) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.sse4_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -148,27 +148,27 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.sse4_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.sse4_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.424686e-01
-Avg ME (F77/C++)    = 0.14246859631675157
-Relative difference = 2.5853054135974944e-08
+Avg ME (F77/C++)    = 0.14246860808920836
+Relative difference = 5.677888572434963e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.avx2_f_inl0_hrd0/check.exe -p 64 256 10 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=16384)
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/avx2+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 5.851561e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 6.587029e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 6.587029e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.618798e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.338072e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.338072e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018828e+01 +- 1.429922e+01 )  GeV^-2
-TOTAL       :     0.304354 sec
-       870,497,984      cycles                           #    2.826 GHz                    
-     1,933,292,389      instructions                     #    2.22  insn per cycle         
-       0.308728297 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3491) (512y:    0) (512z:    0)
+TOTAL       :     0.317968 sec
+       874,954,516      cycles                           #    2.715 GHz                    
+     1,952,010,632      instructions                     #    2.23  insn per cycle         
+       0.323135602 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3488) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.avx2_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -176,27 +176,27 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.avx2_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.avx2_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.424749e-01
-Avg ME (F77/C++)    = 0.14247489318272599
-Relative difference = 4.784894739577799e-08
+Avg ME (F77/C++)    = 0.14247490815036912
+Relative difference = 5.7205649062398515e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512y_f_inl0_hrd0/check.exe -p 64 256 10 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=16384)
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/512y+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 6.204206e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.044292e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.044292e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.140069e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.015278e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.015278e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018828e+01 +- 1.429922e+01 )  GeV^-2
-TOTAL       :     0.288758 sec
-       819,562,722      cycles                           #    2.802 GHz                    
-     1,856,394,583      instructions                     #    2.27  insn per cycle         
-       0.293079680 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3335) (512y:   22) (512z:    0)
+TOTAL       :     0.296217 sec
+       805,080,990      cycles                           #    2.697 GHz                    
+     1,834,280,964      instructions                     #    2.28  insn per cycle         
+       0.301462842 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3186) (512y:   15) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512y_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -204,27 +204,27 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512y_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512y_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.424749e-01
-Avg ME (F77/C++)    = 0.14247489318272599
-Relative difference = 4.784894739577799e-08
+Avg ME (F77/C++)    = 0.14247490815036912
+Relative difference = 5.7205649062398515e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512z_f_inl0_hrd0/check.exe -p 64 256 10 --bridge OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 WARNING! Instantiate host Bridge (nevt=16384)
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+BRDHST/512z+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[16] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 4.631074e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 5.091279e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 5.091279e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.472935e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.920053e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.920053e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018829e+01 +- 1.429922e+01 )  GeV^-2
-TOTAL       :     0.380781 sec
-       752,756,759      cycles                           #    1.958 GHz                    
-     1,346,012,282      instructions                     #    1.79  insn per cycle         
-       0.385182072 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1968) (512y:   32) (512z: 2383)
+TOTAL       :     0.395002 sec
+       745,120,207      cycles                           #    1.866 GHz                    
+     1,329,072,598      instructions                     #    1.78  insn per cycle         
+       0.400211929 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1730) (512y:   24) (512z: 2387)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512z_f_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -232,8 +232,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512z_f_inl0_hrd0/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512z_f_inl0_hrd0/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.424749e-01
-Avg ME (F77/C++)    = 0.14247489383243206
-Relative difference = 4.32888033512879e-08
+Avg ME (F77/C++)    = 0.14247490450137867
+Relative difference = 3.159418737238044e-08
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_gqttq_mad/log_gqttq_mad_f_inl0_hrd1.txt b/epochX/cudacpp/tput/logs_gqttq_mad/log_gqttq_mad_f_inl0_hrd1.txt
index 8032039f3c..a71ead3e03 100644
--- a/epochX/cudacpp/tput/logs_gqttq_mad/log_gqttq_mad_f_inl0_hrd1.txt
+++ b/epochX/cudacpp/tput/logs_gqttq_mad/log_gqttq_mad_f_inl0_hrd1.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_f_inl0_hrd1'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux'
 
-DATE: 2023-11-24_14:40:42
+DATE: 2024-01-30_05:05:20
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_f_inl0_hrd1/gcheck.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GUX_TTXUX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 5.261838e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.178248e+08                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.310022e+08                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.328749e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.215965e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.352409e+08                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018174e+01 +- 1.429492e+01 )  GeV^-2
-TOTAL       :     0.437409 sec
-     1,963,259,812      cycles                           #    3.015 GHz                    
-     2,775,580,273      instructions                     #    1.41  insn per cycle         
-       0.717773083 seconds time elapsed
+TOTAL       :     0.447516 sec
+     1,904,038,107      cycles                           #    2.819 GHz                    
+     2,679,740,960      instructions                     #    1.41  insn per cycle         
+       0.754557698 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_f_inl0_hrd1/gcheck.exe -p 64 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 167
@@ -65,17 +65,17 @@ WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 .........................................................................
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_f_inl0_hrd1/gcheck.exe -p 2048 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GUX_TTXUX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CUD:FLT+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 7.175724e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.774857e+08                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.916629e+08                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 7.182679e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.774687e+08                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.914662e+08                 )  sec^-1
 MeanMatrixElemValue         = ( 2.571361e+02 +- 2.114021e+02 )  GeV^-2
-TOTAL       :     0.475371 sec
-     2,078,846,085      cycles                           #    2.960 GHz                    
-     2,876,831,626      instructions                     #    1.38  insn per cycle         
-       0.759711014 seconds time elapsed
+TOTAL       :     0.483893 sec
+     2,007,596,287      cycles                           #    2.824 GHz                    
+     2,863,986,921      instructions                     #    1.43  insn per cycle         
+       0.770182944 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_f_inl0_hrd1/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_f_inl0_hrd1/fgcheck.exe 2 64 2
@@ -86,20 +86,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_f_inl0_hrd1/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.164325e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.190669e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.190669e+05                 )  sec^-1
-MeanMatrixElemValue         = ( 2.018563e+01 +- 1.429903e+01 )  GeV^-2
-TOTAL       :     1.426803 sec
-     4,442,867,476      cycles                           #    3.106 GHz                    
-    13,028,535,311      instructions                     #    2.93  insn per cycle         
-       1.433195231 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  727) (avx2:    0) (512y:    0) (512z:    0)
+EvtsPerSec[Rmb+ME]     (23) = ( 1.104449e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.131163e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.131163e+05                 )  sec^-1
+MeanMatrixElemValue         = ( 2.018563e+01 +- 1.429902e+01 )  GeV^-2
+TOTAL       :     1.505306 sec
+     4,350,737,729      cycles                           #    2.883 GHz                    
+    12,588,700,465      instructions                     #    2.89  insn per cycle         
+       1.517040580 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:  759) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_f_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -107,26 +107,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_f_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_f_inl0_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.424686e-01
-Avg ME (F77/C++)    = 0.14246857540270419
-Relative difference = 1.7265064590569047e-07
+Avg ME (F77/C++)    = 0.14246860569653919
+Relative difference = 3.998452420257791e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.sse4_f_inl0_hrd1/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.121532e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.320092e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.320092e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.107801e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.322563e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.322563e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018563e+01 +- 1.429902e+01 )  GeV^-2
-TOTAL       :     0.543798 sec
-     1,695,653,127      cycles                           #    3.095 GHz                    
-     4,509,334,487      instructions                     #    2.66  insn per cycle         
-       0.556325987 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 3588) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     0.548678 sec
+     1,589,053,041      cycles                           #    2.872 GHz                    
+     4,241,478,972      instructions                     #    2.67  insn per cycle         
+       0.565533397 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 3248) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.sse4_f_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -134,26 +134,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.sse4_f_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.sse4_f_inl0_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.424686e-01
-Avg ME (F77/C++)    = 0.14246859631675157
-Relative difference = 2.5853054135974944e-08
+Avg ME (F77/C++)    = 0.14246860808920836
+Relative difference = 5.677888572434963e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.avx2_f_inl0_hrd1/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 6.122817e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 6.890273e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 6.890273e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 5.682195e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 6.406347e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 6.406347e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018828e+01 +- 1.429922e+01 )  GeV^-2
-TOTAL       :     0.287226 sec
-       852,393,433      cycles                           #    2.926 GHz                    
-     1,893,938,143      instructions                     #    2.22  insn per cycle         
-       0.301768010 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3461) (512y:    0) (512z:    0)
+TOTAL       :     0.310286 sec
+       851,032,417      cycles                           #    2.702 GHz                    
+     1,913,907,734      instructions                     #    2.25  insn per cycle         
+       0.327654627 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3463) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.avx2_f_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -161,26 +161,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.avx2_f_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.avx2_f_inl0_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.424749e-01
-Avg ME (F77/C++)    = 0.14247489318272599
-Relative difference = 4.784894739577799e-08
+Avg ME (F77/C++)    = 0.14247490815036912
+Relative difference = 5.7205649062398515e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512y_f_inl0_hrd1/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 6.582926e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.478799e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.478799e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 6.251030e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.131063e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.131063e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018828e+01 +- 1.429922e+01 )  GeV^-2
-TOTAL       :     0.268312 sec
-       800,191,089      cycles                           #    2.937 GHz                    
-     1,816,120,065      instructions                     #    2.27  insn per cycle         
-       0.281438188 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3298) (512y:   22) (512z:    0)
+TOTAL       :     0.283621 sec
+       779,432,148      cycles                           #    2.704 GHz                    
+     1,795,928,128      instructions                     #    2.30  insn per cycle         
+       0.301196370 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3164) (512y:   15) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512y_f_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -188,26 +188,26 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512y_f_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512y_f_inl0_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.424749e-01
-Avg ME (F77/C++)    = 0.14247489318272599
-Relative difference = 4.784894739577799e-08
+Avg ME (F77/C++)    = 0.14247490815036912
+Relative difference = 5.7205649062398515e-08
 OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512z_f_inl0_hrd1/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:FLT+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = FLOAT (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[16] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 4.957580e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 5.456912e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 5.456912e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 4.530328e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 4.979352e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 4.979352e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018829e+01 +- 1.429922e+01 )  GeV^-2
-TOTAL       :     0.351938 sec
-       734,061,618      cycles                           #    2.062 GHz                    
-     1,302,951,487      instructions                     #    1.77  insn per cycle         
-       0.362652515 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1931) (512y:   32) (512z: 2383)
+TOTAL       :     0.386557 sec
+       722,333,254      cycles                           #    1.844 GHz                    
+     1,287,373,146      instructions                     #    1.78  insn per cycle         
+       0.407217093 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1709) (512y:   24) (512z: 2387)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512z_f_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -215,8 +215,8 @@ runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcess
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512z_f_inl0_hrd1/check.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512z_f_inl0_hrd1/fcheck.exe 2 64 2
 Avg ME (C++/C++)    = 1.424749e-01
-Avg ME (F77/C++)    = 0.14247489383243206
-Relative difference = 4.32888033512879e-08
+Avg ME (F77/C++)    = 0.14247490450137867
+Relative difference = 3.159418737238044e-08
 OK (relative difference <= 5E-3)
 =========================================================================
 
diff --git a/epochX/cudacpp/tput/logs_gqttq_mad/log_gqttq_mad_m_inl0_hrd0.txt b/epochX/cudacpp/tput/logs_gqttq_mad/log_gqttq_mad_m_inl0_hrd0.txt
index 1f5cde87e5..3f17b073e2 100644
--- a/epochX/cudacpp/tput/logs_gqttq_mad/log_gqttq_mad_m_inl0_hrd0.txt
+++ b/epochX/cudacpp/tput/logs_gqttq_mad/log_gqttq_mad_m_inl0_hrd0.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_m_inl0_hrd0'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux'
 
-DATE: 2023-11-24_14:40:58
+DATE: 2024-01-30_05:05:38
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_m_inl0_hrd0/gcheck.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 2.593637e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 5.233261e+07                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 5.638999e+07                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.696364e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 5.334716e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.710197e+07                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018083e+01 +- 1.429474e+01 )  GeV^-2
-TOTAL       :     0.449003 sec
-     1,949,950,502      cycles                           #    2.943 GHz                    
-     2,732,368,699      instructions                     #    1.40  insn per cycle         
-       0.743857131 seconds time elapsed
+TOTAL       :     0.455743 sec
+     1,899,569,009      cycles                           #    2.822 GHz                    
+     2,690,270,670      instructions                     #    1.42  insn per cycle         
+       0.752301124 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_m_inl0_hrd0/gcheck.exe -p 64 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 255
@@ -65,17 +65,17 @@ WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 .........................................................................
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_m_inl0_hrd0/gcheck.exe -p 2048 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 3.250012e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 7.124195e+07                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.543411e+07                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.256330e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.134663e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.562668e+07                 )  sec^-1
 MeanMatrixElemValue         = ( 2.602505e+02 +- 2.116328e+02 )  GeV^-2
-TOTAL       :     0.532370 sec
-     2,306,545,413      cycles                           #    3.002 GHz                    
-     3,257,931,207      instructions                     #    1.41  insn per cycle         
-       0.827800812 seconds time elapsed
+TOTAL       :     0.544746 sec
+     2,203,075,600      cycles                           #    2.810 GHz                    
+     3,150,811,707      instructions                     #    1.43  insn per cycle         
+       0.843284004 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_m_inl0_hrd0/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_m_inl0_hrd0/fgcheck.exe 2 64 2
@@ -86,20 +86,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_m_inl0_hrd0/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.090450e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.112908e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.112908e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 9.796791e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.000139e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.000139e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018083e+01 +- 1.429474e+01 )  GeV^-2
-TOTAL       :     1.524359 sec
-     4,736,940,349      cycles                           #    3.101 GHz                    
-    13,465,176,292      instructions                     #    2.84  insn per cycle         
-       1.531673516 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  840) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     1.696226 sec
+     4,903,205,903      cycles                           #    2.884 GHz                    
+    13,824,553,372      instructions                     #    2.82  insn per cycle         
+       1.707005330 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 1135) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_m_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -113,20 +113,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.sse4_m_inl0_hrd0/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.007098e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.083721e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.083721e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.870381e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.944831e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.944831e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018083e+01 +- 1.429474e+01 )  GeV^-2
-TOTAL       :     0.837584 sec
-     2,604,823,117      cycles                           #    3.095 GHz                    
-     7,385,606,203      instructions                     #    2.84  insn per cycle         
-       0.848813750 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 3073) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     0.899365 sec
+     2,603,553,029      cycles                           #    2.880 GHz                    
+     7,349,607,266      instructions                     #    2.82  insn per cycle         
+       0.916195330 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 2967) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.sse4_m_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -140,20 +140,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.avx2_m_inl0_hrd0/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.452149e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.684416e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.684416e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.167537e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.382178e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.382178e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018083e+01 +- 1.429474e+01 )  GeV^-2
-TOTAL       :     0.496103 sec
-     1,465,319,733      cycles                           #    2.929 GHz                    
-     3,056,071,451      instructions                     #    2.09  insn per cycle         
-       0.510799326 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3013) (512y:    0) (512z:    0)
+TOTAL       :     0.541013 sec
+     1,471,630,021      cycles                           #    2.697 GHz                    
+     3,084,577,547      instructions                     #    2.10  insn per cycle         
+       0.558891839 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 3008) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.avx2_m_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -167,20 +167,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512y_m_inl0_hrd0/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.858923e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.141930e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.141930e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.661938e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.948590e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.948590e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018083e+01 +- 1.429474e+01 )  GeV^-2
-TOTAL       :     0.446088 sec
-     1,306,981,940      cycles                           #    2.902 GHz                    
-     2,931,113,574      instructions                     #    2.24  insn per cycle         
-       0.461243420 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2799) (512y:  110) (512z:    0)
+TOTAL       :     0.471097 sec
+     1,285,426,170      cycles                           #    2.700 GHz                    
+     2,873,286,331      instructions                     #    2.24  insn per cycle         
+       0.489244149 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2653) (512y:   96) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512y_m_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
@@ -194,20 +194,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512z_m_inl0_hrd0/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=0]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=0]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.504343e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.619127e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.619127e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.322096e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.437722e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.437722e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018083e+01 +- 1.429474e+01 )  GeV^-2
-TOTAL       :     0.677792 sec
-     1,365,651,157      cycles                           #    2.003 GHz                    
-     1,970,379,541      instructions                     #    1.44  insn per cycle         
-       0.689595363 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1700) (512y:  114) (512z: 2171)
+TOTAL       :     0.731479 sec
+     1,311,962,532      cycles                           #    1.782 GHz                    
+     1,915,335,630      instructions                     #    1.46  insn per cycle         
+       0.746286183 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1493) (512y:   70) (512z: 2164)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512z_m_inl0_hrd0/runTest.exe
 [  PASSED  ] 6 tests.
diff --git a/epochX/cudacpp/tput/logs_gqttq_mad/log_gqttq_mad_m_inl0_hrd1.txt b/epochX/cudacpp/tput/logs_gqttq_mad/log_gqttq_mad_m_inl0_hrd1.txt
index a8d85dd2f3..7294ddea09 100644
--- a/epochX/cudacpp/tput/logs_gqttq_mad/log_gqttq_mad_m_inl0_hrd1.txt
+++ b/epochX/cudacpp/tput/logs_gqttq_mad/log_gqttq_mad_m_inl0_hrd1.txt
@@ -41,23 +41,23 @@ CUDACPP_BUILDDIR='build.512z_m_inl0_hrd1'
 make[1]: Nothing to be done for 'all'.
 make[1]: Leaving directory '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux'
 
-DATE: 2023-11-24_14:41:16
+DATE: 2024-01-30_05:05:56
 
 On itscrd90.cern.ch [CPU: Intel(R) Xeon(R) Silver 4216 CPU] [GPU: 1x Tesla V100S-PCIE-32GB]:
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_m_inl0_hrd1/gcheck.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GUX_TTXUX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 2.654184e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 5.189821e+07                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 5.540640e+07                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.635631e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 5.151573e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 5.502163e+07                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018083e+01 +- 1.429474e+01 )  GeV^-2
-TOTAL       :     0.444065 sec
-     2,018,867,268      cycles                           #    3.015 GHz                    
-     2,835,456,943      instructions                     #    1.40  insn per cycle         
-       0.738016347 seconds time elapsed
+TOTAL       :     0.454560 sec
+     1,887,319,720      cycles                           #    2.810 GHz                    
+     2,686,521,155      instructions                     #    1.42  insn per cycle         
+       0.777570467 seconds time elapsed
 runNcu /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_m_inl0_hrd1/gcheck.exe -p 64 256 1
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 ==PROF== Profiling "sigmaKin": launch__registers_per_thread 255
@@ -65,17 +65,17 @@ WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
 .........................................................................
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_m_inl0_hrd1/gcheck.exe -p 2048 256 1 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CUDA [nvcc 12.0.140 (gcc 11.3.1)] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GUX_TTXUX_CUDA [nvcc 12.0.140 (gcc 12.1.0)] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CUD:MIX+THX:CURDEV+RMBDEV+MESDEV/none+NAVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
-EvtsPerSec[Rmb+ME]     (23) = ( 3.223829e+07                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 6.984831e+07                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 7.384004e+07                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.262333e+07                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 7.007147e+07                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 7.410963e+07                 )  sec^-1
 MeanMatrixElemValue         = ( 2.602505e+02 +- 2.116328e+02 )  GeV^-2
-TOTAL       :     0.536961 sec
-     2,246,742,038      cycles                           #    2.906 GHz                    
-     3,169,759,663      instructions                     #    1.41  insn per cycle         
-       0.832088924 seconds time elapsed
+TOTAL       :     0.538673 sec
+     2,205,224,099      cycles                           #    2.822 GHz                    
+     3,150,366,927      instructions                     #    1.43  insn per cycle         
+       0.838863876 seconds time elapsed
 -------------------------------------------------------------------------
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_m_inl0_hrd1/gcheck.exe --common -p 2 64 2
 cmpExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_m_inl0_hrd1/fgcheck.exe 2 64 2
@@ -86,20 +86,20 @@ OK (relative difference <= 5E-3)
 =========================================================================
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_m_inl0_hrd1/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/none+NAVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = SCALAR ('none': ~vector[1], no SIMD)
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 1.089936e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 1.112348e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 1.112348e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 9.769998e+04                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 9.971532e+04                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 9.971532e+04                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018083e+01 +- 1.429474e+01 )  GeV^-2
-TOTAL       :     1.524608 sec
-     4,732,046,885      cycles                           #    3.096 GHz                    
-    13,451,186,768      instructions                     #    2.84  insn per cycle         
-       1.531974912 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:  827) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     1.700454 sec
+     4,910,062,395      cycles                           #    2.880 GHz                    
+    13,831,764,171      instructions                     #    2.82  insn per cycle         
+       1.712052278 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 1130) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.none_m_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -113,20 +113,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.sse4_m_inl0_hrd1/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/sse4+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[2] ('sse4': SSE4.2, 128bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.003818e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.078884e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.078884e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 1.857842e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 1.932046e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 1.932046e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018083e+01 +- 1.429474e+01 )  GeV^-2
-TOTAL       :     0.838343 sec
-     2,601,051,226      cycles                           #    3.087 GHz                    
-     7,389,258,616      instructions                     #    2.84  insn per cycle         
-       0.853180948 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4: 3062) (avx2:    0) (512y:    0) (512z:    0)
+TOTAL       :     0.905414 sec
+     2,615,099,772      cycles                           #    2.873 GHz                    
+     7,353,136,311      instructions                     #    2.81  insn per cycle         
+       0.925236073 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4: 2957) (avx2:    0) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.sse4_m_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -140,20 +140,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.avx2_m_inl0_hrd1/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/avx2+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('avx2': AVX2, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.440885e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 3.665503e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 3.665503e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.160999e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.374264e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.374264e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018083e+01 +- 1.429474e+01 )  GeV^-2
-TOTAL       :     0.497494 sec
-     1,467,074,103      cycles                           #    2.924 GHz                    
-     3,056,319,919      instructions                     #    2.08  insn per cycle         
-       0.512504484 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2990) (512y:    0) (512z:    0)
+TOTAL       :     0.541919 sec
+     1,475,084,747      cycles                           #    2.698 GHz                    
+     3,084,915,220      instructions                     #    2.09  insn per cycle         
+       0.559487031 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2986) (512y:    0) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.avx2_m_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -167,20 +167,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512y_m_inl0_hrd1/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512y+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[4] ('512y': AVX512, 256bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 3.870867e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 4.157659e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 4.157659e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 3.676411e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 3.967587e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 3.967587e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018083e+01 +- 1.429474e+01 )  GeV^-2
-TOTAL       :     0.444644 sec
-     1,306,078,783      cycles                           #    2.911 GHz                    
-     2,931,883,300      instructions                     #    2.24  insn per cycle         
-       0.459998118 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2775) (512y:  110) (512z:    0)
+TOTAL       :     0.469154 sec
+     1,285,211,957      cycles                           #    2.712 GHz                    
+     2,875,140,516      instructions                     #    2.24  insn per cycle         
+       0.485058196 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 2636) (512y:   96) (512z:    0)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512y_m_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
@@ -194,20 +194,20 @@ OK (relative difference <= 5E-3)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512z_m_inl0_hrd1/check.exe -p 64 256 10 OMP=
 WARNING! CUDACPP_RUNTIME_ENABLEFPE is set: enable Floating Point Exceptions
-Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 11.3.1] [inlineHel=0] [hardcodePARAM=1]
+Process                     = SIGMA_SM_GUX_TTXUX_CPP [gcc 12.1.0] [inlineHel=0] [hardcodePARAM=1]
 Workflow summary            = CPP:MIX+CXS:CURHST+RMBHST+MESHST/512z+CXVBRK
 FP precision                = MIXED (NaN/abnormal=0, zero=0)
 Internal loops fptype_sv    = VECTOR[8] ('512z': AVX512, 512bit) [cxtype_ref=YES]
 OMP threads / `nproc --all` = 1 / 4
-EvtsPerSec[Rmb+ME]     (23) = ( 2.456308e+05                 )  sec^-1
-EvtsPerSec[MatrixElems] (3) = ( 2.570811e+05                 )  sec^-1
-EvtsPerSec[MECalcOnly] (3a) = ( 2.570811e+05                 )  sec^-1
+EvtsPerSec[Rmb+ME]     (23) = ( 2.334432e+05                 )  sec^-1
+EvtsPerSec[MatrixElems] (3) = ( 2.451352e+05                 )  sec^-1
+EvtsPerSec[MECalcOnly] (3a) = ( 2.451352e+05                 )  sec^-1
 MeanMatrixElemValue         = ( 2.018083e+01 +- 1.429474e+01 )  GeV^-2
-TOTAL       :     0.690659 sec
-     1,366,599,933      cycles                           #    1.966 GHz                    
-     1,970,195,836      instructions                     #    1.44  insn per cycle         
-       0.702450567 seconds time elapsed
-=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1676) (512y:  114) (512z: 2171)
+TOTAL       :     0.728206 sec
+     1,313,839,367      cycles                           #    1.794 GHz                    
+     1,915,620,790      instructions                     #    1.46  insn per cycle         
+       0.743678029 seconds time elapsed
+=Symbols in CPPProcess.o= (~sse4:    0) (avx2: 1476) (512y:   70) (512z: 2164)
 -------------------------------------------------------------------------
 runExe /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gux_ttxux/build.512z_m_inl0_hrd1/runTest.exe
 [  PASSED  ] 6 tests.
diff --git a/epochX/cudacpp/tput/teeThroughputX.sh b/epochX/cudacpp/tput/teeThroughputX.sh
index bd478452ac..de0a1e912a 100755
--- a/epochX/cudacpp/tput/teeThroughputX.sh
+++ b/epochX/cudacpp/tput/teeThroughputX.sh
@@ -93,6 +93,8 @@ for arg in $*; do
     rndgen=$arg
   elif [ "$arg" == "-curhst" ]; then
     rndgen=$arg
+  elif [ "$arg" == "-rorhst" ]; then
+    rndgen=$arg
   elif [ "$arg" == "-rmbhst" ]; then
     rmbsmp=$arg
   elif [ "$arg" == "-bridge" ]; then
diff --git a/epochX/cudacpp/tput/throughputX.sh b/epochX/cudacpp/tput/throughputX.sh
index 8160f7fbb9..503d060237 100755
--- a/epochX/cudacpp/tput/throughputX.sh
+++ b/epochX/cudacpp/tput/throughputX.sh
@@ -187,6 +187,9 @@ while [ "$1" != "" ]; do
   elif [ "$1" == "-curhst" ]; then
     rndgen=" -${1}"
     shift
+  elif [ "$1" == "-rorhst" ]; then
+    rndgen=" -${1}"
+    shift
   elif [ "$1" == "-rmbhst" ]; then
     rmbsmp=" -${1}"
     shift
@@ -523,6 +526,7 @@ function cmpExe() {
 
 # Profile #registers and %divergence only
 function runNcu() {
+  if ! ncu -v > /dev/null 2>&1; then return; fi
   if [ "${maketype}" == "-dryrun" ]; then return; fi
   exe=$1
   args="$2"
@@ -545,6 +549,7 @@ function runNcu() {
 # See https://docs.nvidia.com/gameworks/content/developertools/desktop/analysis/report/cudaexperiments/kernellevel/branchstatistics.htm
 # See https://docs.nvidia.com/gameworks/content/developertools/desktop/analysis/report/cudaexperiments/sourcelevel/divergentbranch.htm
 function runNcuDiv() {
+  if ! ncu -v > /dev/null 2>&1; then return; fi
   if [ "${maketype}" == "-dryrun" ]; then return; fi
   exe=$1
   args="-p 1 32 1"
@@ -567,6 +572,7 @@ function runNcuDiv() {
 
 # Profiles sectors and requests
 function runNcuReq() {
+  if ! ncu -v > /dev/null 2>&1; then return; fi
   if [ "${maketype}" == "-dryrun" ]; then return; fi
   exe=$1
   ncuArgs="$2"
@@ -580,7 +586,13 @@ function runNcuReq() {
   set +x
 }
 
-if nvidia-smi -L > /dev/null 2>&1; then gpuTxt="$(nvidia-smi -L | wc -l)x $(nvidia-smi -L | awk '{print $3,$4}' | sort -u)"; else gpuTxt=none; fi
+if nvidia-smi -L > /dev/null 2>&1; then
+  gpuTxt="$(nvidia-smi -L | wc -l)x $(nvidia-smi -L | awk '{print $3,$4}' | sort -u)"
+elif rocm-smi -i > /dev/null 2>&1; then
+  gpuTxt="$(rocm-smi --showproductname | grep 'Card series' | awk '{print $5,$6,$7}')"
+else
+  gpuTxt=none
+fi
 if [ "${unames}" == "Darwin" ]; then 
   cpuTxt=$(sysctl -h machdep.cpu.brand_string)
   cpuTxt=${cpuTxt/machdep.cpu.brand_string: }