Feature/dtypes

.github/workflows/build.yml

@@ -26,8 +26,10 @@ jobs:
             CC: gcc-9
             CXX: g++-9
             CXXFLAGS: -Wno-maybe-uninitialized -Wno-cpp
+            FC: gfortran-9
             GCOV: gcov-9
             OCCA_COVERAGE: 1
+            OCCA_FORTRAN_ENABLED: 1
             useCMake: true
 
           - name: "[Ubuntu] clang-11"
@@ -49,9 +51,10 @@ jobs:
             CC: icx
             CXX: icpx
             CXXFLAGS: -Wno-uninitialized
-            FC: ifx
+            FC: ifort
             GCOV: gcov-9
             OCCA_COVERAGE: 1
+            OCCA_FORTRAN_ENABLED: 1
             useCMake: true
             useoneAPI: true
 
@@ -103,7 +106,8 @@ jobs:
       shell: bash
       run: |
         sudo apt update
-        sudo apt install intel-oneapi-compiler-dpcpp-cpp 
+        sudo apt install intel-oneapi-compiler-dpcpp-cpp
+        sudo apt install intel-oneapi-compiler-fortran
 
     - name: Compiler info
       if: ${{ !matrix.useCMake }}
@@ -115,8 +119,12 @@ jobs:
         cmake -S . -B build \
         -DCMAKE_BUILD_TYPE="RelWithDebInfo" \
         -DCMAKE_INSTALL_PREFIX=install \
+        -DCMAKE_C_COMPILER=${CC} \
+        -DCMAKE_CXX_COMPILER=${CXX} \
+        -DCMAKE_Fortran_COMPILER=${FC} \
         -DENABLE_TESTS=ON \
-        -DENABLE_EXAMPLES=ON
+        -DENABLE_EXAMPLES=ON \
+        -DENABLE_FORTRAN=ON
 
     - name: CMake configure
       if: ${{ matrix.useCMake && matrix.useoneAPI}}
@@ -128,8 +136,12 @@ jobs:
         cmake -S . -B build \
         -DCMAKE_BUILD_TYPE="RelWithDebInfo" \
         -DCMAKE_INSTALL_PREFIX=install \
+        -DCMAKE_C_COMPILER=${CC} \
+        -DCMAKE_CXX_COMPILER=${CXX} \
+        -DCMAKE_Fortran_COMPILER=${FC} \
         -DENABLE_TESTS=ON \
         -DENABLE_EXAMPLES=ON \
+        -DENABLE_FORTRAN=ON \
         -DCMAKE_PREFIX_PATH="/opt/intel/oneapi/compiler/latest/linux;/opt/intel/oneapi/compiler/latest/linux/compiler"
 
     - name: CMake build
@@ -165,7 +177,8 @@ jobs:
     - name: Run CTests
       if: ${{ matrix.useCMake && !matrix.useoneAPI }}
       run: |
-        ctest --test-dir build --progress --output-on-failure --parallel 8 --schedule-random -E "examples_cpp_arrays-opencl|examples_cpp_for_loops-opencl|examples_cpp_generic_inline_kernel-opencl|examples_cpp_shared_memory-opencl|examples_cpp_nonblocking_streams-opencl|examples_cpp_shared_memory-dpcpp|examples_cpp_nonblocking_streams-dpcpp|examples_cpp_for_loops-dpcpp|examples_cpp_arrays-dpcpp"
+        ctest --test-dir build --progress --output-on-failure --parallel 8 --schedule-random -E "examples_cpp_arrays-opencl|examples_cpp_for_loops-opencl|examples_cpp_generic_inline_kernel-opencl|examples_cpp_shared_memory-opencl|examples_cpp_nonblocking_streams-opencl|examples_cpp_for_loops-dpcpp|examples_cpp_arrays-dpcpp|examples_cpp_nonblocking_streams-dpcpp"
+
 
 
     - name: Run CTests
@@ -176,7 +189,7 @@ jobs:
       run: |
         source /opt/intel/oneapi/setvars.sh
         export SYCL_DEVICE_FILTER=opencl.cpu
-        ctest --test-dir build --progress --output-on-failure --parallel 8 --schedule-random -E "examples_cpp_arrays-opencl|examples_cpp_for_loops-opencl|examples_cpp_generic_inline_kernel-opencl|examples_cpp_shared_memory-opencl|examples_cpp_nonblocking_streams-opencl|examples_cpp_shared_memory-dpcpp|examples_cpp_nonblocking_streams-dpcpp|examples_cpp_for_loops-dpcpp|examples_cpp_arrays-dpcpp"
+        ctest --test-dir build --progress --output-on-failure --parallel 8 --schedule-random -E "examples_cpp_arrays-opencl|examples_cpp_for_loops-opencl|examples_cpp_generic_inline_kernel-opencl|examples_cpp_shared_memory-opencl|examples_cpp_nonblocking_streams-opencl|examples_cpp_for_loops-dpcpp|examples_cpp_arrays-dpcpp|examples_cpp_nonblocking_streams-dpcpp"
 
     - name: Upload code coverage
       if: ${{ matrix.OCCA_COVERAGE }}

cmake/FindOpenCLWrapper.cmake

@@ -3,31 +3,42 @@
 # This Find module is also distributed alongside the occa package config file!
 ###############################################################################
 
-# Look in some default places for OpenCL and set OPENCL_ROOT if not already set
-if(NOT OPENCL_ROOT)
-  # Search in user specified path first
-  find_path(OPENCL_ROOT
-    NAMES CL/cl.h
+# Try finding OpenCL. The user should set OpenCL_ROOT if needed.
+find_package(OpenCL QUIET)
+if(NOT OpenCL_FOUND)
+  # Otherwise, look for the headers and library in standard locations
+  find_path(OpenCL_INCLUDE_DIR
+    NAMES CL/cl.h OpenCL/cl.h
     PATHS
-    ENV   OPENCL_PATH
-    DOC   "OPENCL root location"
-    NO_DEFAULT_PATH)
+      ENV CUDA_PATH
+      ENV CUDAToolkit_ROOT
+      ENV ROCM_PATH
+      ENV NVHPC_ROOT
+      ENV SYCL_ROOT
+      /usr/local/cuda
+      /opt/rocm/opencl
+      /opt/intel/oneapi/compiler/latest/linux
+    PATH_SUFFIXES
+      include
+      include/sycl
+  )
 
-  # Now search in default path
-  find_path(OPENCL_ROOT
-    NAMES CL/cl.h
-    PATHS 
-      /usr 
-      /opt/rocm/opencl 
-      /usr/local/cuda 
+  find_library(OpenCL_LIBRARY
+    NAMES OpenCL libOpenCL
+    PATHS
+      ENV CUDA_PATH
+      ENV CUDAToolkit_ROOT
+      ENV ROCM_PATH
+      ENV NVHPC_ROOT
+      ENV SYCL_ROOT
+      /usr/local/cuda
+      /opt/rocm/opencl
       /opt/intel/oneapi/compiler/latest/linux
-    PATH_SUFFIXES sycl
-    DOC   "OPENCL root location")
+    PATH_SUFFIXES
+      lib
+      lib64
+  )
 endif()
-
-# Trick CMake's default OpenCL module to look in our directory
-set(ENV{AMDAPPSDKROOT} ${OPENCL_ROOT})
-
 find_package(OpenCL)
 
 include(FindPackageHandleStandardArgs)

examples/cpp/05_custom_types/main.cpp

@@ -62,10 +62,10 @@ int main(int argc, const char **argv) {
   myFloatDtype.registerType();
 
   // Struct dtype
-  occa::dtype_t myFloat2Dtype;
-  myFloat2Dtype.registerType();
+  occa::dtype_t myFloat2Dtype("myFloat2", sizeof(myFloat2));
   myFloat2Dtype.addField("x", occa::dtype::float_);
   myFloat2Dtype.addField("y", occa::dtype::float_);
+  myFloat2Dtype.registerType();
 
   // Tuple dtype
   occa::dtype_t myFloat4Dtype = occa::dtype_t::tuple(occa::dtype::float_, 4);
@@ -108,6 +108,66 @@ int main(int argc, const char **argv) {
     }
   }
 
+  o_a.free();
+  o_b.free();
+  o_ab.free();
+
+  // Templated dtypes
+  // - occa::memory dtypes can be created and assigned via templated methods
+  // - dtypes registered this way are tracked internally and destroyed when deregistered
+  // - Registered dtype_t objects are treated as singletons and assumed
+  //     to exist while the memory objects are still alive
+
+  // Automatic creation
+  // Using a type that has not been registered with OCCA will be
+  // automatically registered as a byte field (thus all type-checking is disabled)
+  o_a = occa::malloc<myFloat>(entries);
+
+  // Struct dtype
+  occa::dtype::registerType<myFloat2>("myFloat2",
+                                      {"x", "y"},
+                                      {occa::dtype::float_, occa::dtype::float_});
+
+  o_b = occa::malloc<myFloat2>(entries / 2);
+
+  // Tuple dtype
+  occa::dtype::registerType<myFloat4>("myFloat4", occa::dtype::float_, 4);
+
+  o_ab = occa::malloc<myFloat4>(entries / 4);
+
+  // Copy memory to the device
+  o_a.copyFrom(a);
+  o_b.copyFrom(b);
+
+  // Launch device kernel
+  addVectors(entries,
+             o_a.cast(occa::dtype::float_),
+             o_b,
+             o_ab);
+
+  // Copy result to the host
+  o_ab.copyTo(ab);
+
+  // Assert values
+  for (int i = 0; i < (entries / 4); ++i) {
+    for (int j = 0; j < 4; ++j) {
+      std::cout << '(' << i << ',' << j << ") : " << ab[i].values[j] << '\n';
+    }
+  }
+  for (int i = 0; i < entries; ++i) {
+    float a_i  = a[i].value;
+    float b_i  = (i % 2) ? b[i / 2].y : b[i / 2].x;
+    float ab_i = ab[i / 4].values[i % 4];
+    if (!occa::areBitwiseEqual(ab_i, a_i + b_i)) {
+      throw 1;
+    }
+  }
+
+  // Can deregister templated dtypes if they're no longer needed
+  occa::dtype::deRegisterType<myFloat>();
+  occa::dtype::deRegisterType<myFloat2>();
+  occa::dtype::deRegisterType<myFloat4>();
+
   // Free host memory
   delete [] a;
   delete [] b;

examples/cpp/17_memory_pool/CMakeLists.txt

@@ -0,0 +1,4 @@
+compile_cpp_example(memory_pool main.cpp)
+
+add_custom_target(cpp_example_memory_pool_okl ALL COMMAND ${CMAKE_COMMAND} -E copy ${CMAKE_CURRENT_SOURCE_DIR}/addVectors.okl addVectors.okl)
+add_dependencies(examples_cpp_memory_pool cpp_example_memory_pool_okl)

examples/cpp/17_memory_pool/Makefile

@@ -0,0 +1,30 @@
+
+PROJ_DIR := $(dir $(abspath $(lastword $(MAKEFILE_LIST))))
+
+ifndef OCCA_DIR
+  include $(PROJ_DIR)/../../../scripts/build/Makefile
+else
+  include ${OCCA_DIR}/scripts/build/Makefile
+endif
+
+#---[ COMPILATION ]-------------------------------
+headers = $(wildcard $(incPath)/*.hpp) $(wildcard $(incPath)/*.tpp)
+sources = $(wildcard $(srcPath)/*.cpp)
+
+objects  = $(subst $(srcPath)/,$(objPath)/,$(sources:.cpp=.o))
+
+executables: ${PROJ_DIR}/main
+
+${PROJ_DIR}/main: $(objects) $(headers) ${PROJ_DIR}/main.cpp
+	$(compiler) $(compilerFlags) -o ${PROJ_DIR}/main $(flags) $(objects) ${PROJ_DIR}/main.cpp $(paths) $(linkerFlags)
+	@if which install_name_tool > /dev/null 2>&1; then \
+		install_name_tool -add_rpath "${OCCA_DIR}/lib" ${PROJ_DIR}/main; \
+	fi
+
+$(objPath)/%.o:$(srcPath)/%.cpp $(wildcard $(subst $(srcPath)/,$(incPath)/,$(<:.cpp=.hpp))) $(wildcard $(subst $(srcPath)/,$(incPath)/,$(<:.cpp=.tpp)))
+	$(compiler) $(compilerFlags) -o $@ $(flags) -c $(paths) $<
+
+clean:
+	rm -f $(objPath)/*;
+	rm -f ${PROJ_DIR}/main;
+#=================================================

examples/cpp/17_memory_pool/README.md

@@ -0,0 +1,28 @@
+# Example: Memory Pool
+
+A example showing the basics of using OCCA's memory pools
+
+- Creating an OCCA memory
+- Managing memory pool
+- Reserving and using memory from memory pool
+
+# Compiling the Example
+
+```bash
+make
+```
+
+## Usage
+
+```
+> ./main --help
+
+Usage: ./main [OPTIONS]
+
+Example using memory pools
+
+Options:
+  -d, --device     Device properties (default: "{mode: 'Serial'}")
+  -h, --help       Print usage
+  -v, --verbose    Compile kernels in verbose mode
+```

examples/cpp/17_memory_pool/addVectors.okl

@@ -0,0 +1,8 @@
+@kernel void addVectors(const int entries,
+                        const float *a,
+                        const float *b,
+                        float *ab) {
+  for (int i = 0; i < entries; ++i; @tile(4, @outer, @inner)) {
+    ab[i] = a[i] + b[i];
+  }
+}