Merge branch 'main' of github.com:microsoft/DirectXShaderCompiler into devshFixes

Author: AnastaZIuk

.github/CODEOWNERS

@@ -0,0 +1 @@
+* @microsoft/hlsl-release

.github/ISSUE_TEMPLATE/support.md

@@ -0,0 +1,28 @@
+---
+name: User support
+about: Ask the maintainers a question or for support.
+title: ''
+labels: ['user-support', 'needs-triage']
+assignees: ''
+
+---
+
+**What is the issue you're encountering?**
+<!--- Please provide a few sentences describing the issue you encountered. --->
+
+**Steps to Reproduce (if applicable)**
+<!--- If you're encountering a problem, please provide a description of how you
+got to the state you're in. If possible please provide source and tool command
+line options. If the issue reproduces on Compiler Explorer
+(https://godbolt.org/) or Shader Playground
+(https://shader-playground.timjones.io/) please provide a link. If the source is
+split across multiple files, please preprocess into a single file using DXC's 
+command line `-P -Fi <path>`.--->
+
+
+**Environment**
+- DXC version <!-- replace with the output of 'dxc --version' -->
+- Host Operating System <!--- Host operating system and version --->
+
+<!--- If this is a build issue please include CMake version, compiler and
+version, and the CMake invocation used to configure your build directory. --->

README.md

@@ -34,58 +34,7 @@ As an example of community contribution, this project can also target the [SPIR-
 
 ## Building Sources
 
-Building DXC requires:
-
-* [Git](http://git-scm.com/downloads).
-* [Python](https://www.python.org/downloads/) - version 3.x is required
-* [CMake](https://cmake.org/download/) - version >= 3.17.2
-    * The bundled version with Visual Studio works for Windows.
-* The C++ 14 compiler and runtime of your choosing.
-    * DXC is known to compile with recent versions of GCC, Clang and MSVC.
-
-Building on windows additionally requires:
-
-* [Visual Studio 2019 or later](https://www.visualstudio.com/downloads) - select the following workloads: 
-    * Universal Windows Platform Development
-    * Desktop Development with C++
-* [Windows SDK](https://developer.microsoft.com/en-US/windows/downloads/windows-10-sdk) - version 10.0.18362.0 or newer
-* [Windows Driver Kit](https://docs.microsoft.com/en-us/windows-hardware/drivers/download-the-wdk) - same version as the SDK
-
-> A new experimental simplified build and test workflow is documented [here](docs/BuildingAndTestingDXC.rst).
-
-
-Before you build, you will need to have some additional software installed. This is the most straightforward path - see [Building Sources](https://github.com/microsoft/DirectXShaderCompiler/wiki/Building-Sources) on the Wiki for more options, including Visual Studio 2015 and Ninja support.
-
-* [Git](http://git-scm.com/downloads).
-* [Python](https://www.python.org/downloads/) - version 3.x is required
-* [Visual Studio 2019](https://www.visualstudio.com/downloads) - select the following workloads: 
-    * Universal Windows Platform Development
-    * Desktop Development with C++
-* [Windows SDK](https://developer.microsoft.com/en-US/windows/downloads/windows-10-sdk) - version 10.0.18362.0 or newer
-* [Windows Driver Kit](https://docs.microsoft.com/en-us/windows-hardware/drivers/download-the-wdk) - same version as the SDK
-
-After cloning the project, you can set up a build environment shortcut by double-clicking the `utils\hct\hctshortcut.js` file. This will create a shortcut on your desktop with a default configuration. If your system doesn't have the requisite association for .js files, this may not work. If so, open a cmd window and invoke: `wscript.exe utils\hct\hctshortcut.js`.
-
-Tests are built using the TAEF framework which is included in the Windows Driver Kit.
-
-To build, run this command on the HLSL Console.
-
-    hctbuild
-
-You can also run tests with this command.
-
-    hcttest
-
-
-To see a list of additional commands available, run `hcthelp`
-
-## Running Tests
-
-To run tests, open the HLSL Console and run this command after a successful build.
-
-    hcttest
-
-Some tests will run shaders and verify their behavior. These tests also involve a driver that can execute these shaders. See the next section on how this should be currently set up.
+See the full documentation for [Building and testing DXC](docs/BuildingAndTestingDXC.rst) for detailed instructions.
 
 ## Running Shaders
 

azure-pipelines.yml

@@ -53,7 +53,7 @@ stages:
           configuration: Release
           CC: clang
           CXX: clang++
-          CMAKE_OPTS: -DLLVM_ENABLE_WERROR=On -DLLVM_USE_SANITIZER=Address,Undefined -DLLVM_ENABLE_LIBCXX=On -DLLVM_USE_LINKER=lld
+          CMAKE_OPTS: -DLLVM_ENABLE_WERROR=On -DLLVM_USE_SANITIZER='Address;Undefined' -DLLVM_ENABLE_LIBCXX=On -DLLVM_USE_LINKER=lld
           CHECK_ALL_ENV: ASAN_OPTIONS=alloc_dealloc_mismatch=0
           OS: Linux
         Linux_Clang_Debug:

cmake/modules/HandleLLVMOptions.cmake

@@ -532,16 +532,16 @@ if(LLVM_USE_SANITIZER)
     elseif (LLVM_USE_SANITIZER STREQUAL "Address;Undefined" OR
             LLVM_USE_SANITIZER STREQUAL "Undefined;Address")
       append_common_sanitizer_flags()
-      append("-fsanitize=address,undefined -fno-sanitize=vptr,function -fno-sanitize-recover=all"
+      append("-fsanitize=address,undefined -fno-sanitize=vptr,function,alignment -fno-sanitize-recover=all"
               CMAKE_C_FLAGS CMAKE_CXX_FLAGS)
     else()
-      message(WARNING "Unsupported value of LLVM_USE_SANITIZER: ${LLVM_USE_SANITIZER}")
+      message(FATAL_ERROR "Unsupported value of LLVM_USE_SANITIZER: ${LLVM_USE_SANITIZER}")
     endif()
   else()
     if (LLVM_USE_SANITIZER STREQUAL "Address")
       append("-fsanitize=address" CMAKE_C_FLAGS CMAKE_CXX_FLAGS)
     else()
-      message(WARNING "Unsupported value of LLVM_USE_SANITIZER: ${LLVM_USE_SANITIZER}")
+      message(FATAL_ERROR "Unsupported value of LLVM_USE_SANITIZER: ${LLVM_USE_SANITIZER}")
     endif()
   endif()
   if (LLVM_USE_SANITIZE_COVERAGE)

docs/BuildingAndTestingDXC.rst

@@ -21,6 +21,30 @@ generator. In subsequent sections we will describe workflows using Ninja and
 Visual Studio. The Ninja workflow should also apply to makefile generators with
 minimal adaption.
 
+Prerequisites
+-------------
+
+* [Git](http://git-scm.com/downloads).
+* [Python](https://www.python.org/downloads/) - version 3.x is required
+* [CMake](https://cmake.org/download/) - version >= 3.17.2
+    * The bundled version with Visual Studio works for Windows.
+* The C++ 14 compiler and runtime of your choosing.
+    * DXC is known to compile with recent versions of GCC, Clang and MSVC.
+
+Building on windows additionally requires:
+
+* [Visual Studio 2019 or later](https://www.visualstudio.com/downloads) - select the following workloads: 
+    * Universal Windows Platform Development
+    * Desktop Development with C++
+* [Windows SDK](https://developer.microsoft.com/en-US/windows/downloads/windows-10-sdk) - version 10.0.18362.0 or newer
+* [Windows Driver Kit](https://docs.microsoft.com/en-us/windows-hardware/drivers/download-the-wdk) - same version as the SDK
+
+Note: DXC uses submodules for some external dependencies. You must initialize
+the submodules in order to build DXC:
+
+.. code-block:: sh
+  git submodule update --init --recursive
+
 Basic CMake Usage
 -----------------
 
@@ -149,3 +173,33 @@ four commonly used option prefixes:
   code coverage reports. With this setting enabled the
   ``generate-coverage-report`` target is added to the build which produces a
   static HTML page with code coverage analysis results.
+
+Legacy Windows Build Tooling
+----------------------------
+
+After cloning the project, you can set up a build environment shortcut by
+double-clicking the `utils\hct\hctshortcut.js` file. This will create a shortcut
+on your desktop with a default configuration. If your system doesn't have the
+requisite association for .js files, this may not work. If so, open a cmd window
+and invoke: `wscript.exe utils\hct\hctshortcut.js`.
+
+Tests are built using the TAEF framework which is included in the Windows Driver
+Kit.
+
+To build, run this command on the HLSL Console.
+
+.. code-block:: sh
+
+  hctbuild
+
+You can also run tests with this command.
+
+.. code-block:: sh
+
+  hcttest
+
+Some tests will run shaders and verify their behavior. These tests also involve
+a driver that can execute these shaders. See the next section on how this should
+be currently set up.
+
+To see a list of additional commands available, run `hcthelp`

docs/DXIL.rst

@@ -3216,6 +3216,14 @@ SM.GSVALIDINPUTPRIMITIVE                  GS input primitive unrecognized.
 SM.GSVALIDOUTPUTPRIMITIVETOPOLOGY         GS output primitive topology unrecognized.
 SM.HSINPUTCONTROLPOINTCOUNTRANGE          HS input control point count must be [0..%0].  %1 specified.
 SM.HULLPASSTHRUCONTROLPOINTCOUNTMATCH     For pass thru hull shader, input control point count must match output control point count
+SM.INCOMPATIBLECALLINENTRY                Features used in internal function calls must be compatible with entry
+SM.INCOMPATIBLEDERIVINCOMPUTESHADERMODEL  Derivatives in compute-model shaders require shader model 6.6 and above
+SM.INCOMPATIBLEDERIVLAUNCH                Node shaders only support derivatives in broadcasting launch mode
+SM.INCOMPATIBLEOPERATION                  Operations used in entry function must be compatible with shader stage and other properties
+SM.INCOMPATIBLEREQUIRESGROUP              Functions requiring groupshared memory must be called from shaders with a visible group
+SM.INCOMPATIBLESHADERMODEL                Functions may only use features available in the current shader model
+SM.INCOMPATIBLESTAGE                      Functions may only use features available in the entry function's stage
+SM.INCOMPATIBLETHREADGROUPDIM             When derivatives are used in compute-model shaders, the thread group dimensions must be compatible
 SM.INSIDETESSFACTORSIZEMATCHDOMAIN        InsideTessFactor rows, columns (%0, %1) invalid for domain %2.  Expected %3 rows and 1 column.
 SM.INVALIDRESOURCECOMPTYPE                Invalid resource return type.
 SM.INVALIDRESOURCEKIND                    Invalid resources kind.
@@ -3241,7 +3249,7 @@ SM.NOPSOUTPUTIDX                          Pixel shader output registers are not
 SM.OPCODE                                 Opcode must be defined in target shader model
 SM.OPCODEININVALIDFUNCTION                Invalid DXIL opcode usage like StorePatchConstant in patch constant function
 SM.OPERAND                                Operand must be defined in target shader model.
-SM.OUTPUTCONTROLPOINTCOUNTRANGE           output control point count must be [0..%0].  %1 specified.
+SM.OUTPUTCONTROLPOINTCOUNTRANGE           output control point count must be [%0..%1].  %2 specified.
 SM.OUTPUTCONTROLPOINTSTOTALSCALARS        Total number of scalars across all HS output control points must not exceed .
 SM.PATCHCONSTANTONLYFORHSDS               patch constant signature only valid in HS and DS.
 SM.PSCONSISTENTINTERP                     Interpolation mode for PS input position must be linear_noperspective_centroid or linear_noperspective_sample when outputting oDepthGE or oDepthLE and not running at sample frequency (which is forced by inputting SV_SampleIndex or declaring an input linear_sample or linear_noperspective_sample).
@@ -3382,4 +3390,3 @@ The following work on this specification is still pending:
 
 * Consider moving some additional tables and lists into hctdb and cross-reference.
 * Complete the extended documentation for instructions.
-

include/dxc/DXIL/DxilConstants.h

@@ -112,6 +112,7 @@ const unsigned kMaxClipOrCullDistanceElementCount = 2;
 const unsigned kMaxClipOrCullDistanceCount = 2 * 4;
 const unsigned kMaxGSOutputVertexCount = 1024;
 const unsigned kMaxGSInstanceCount = 32;
+const unsigned kMinIAPatchControlPointCount = 1;
 const unsigned kMaxIAPatchControlPointCount = 32;
 const float kHSMaxTessFactorLowerBound = 1.0f;
 const float kHSMaxTessFactorUpperBound = 64.0f;

include/dxc/DXIL/DxilMetadataHelper.h

@@ -332,6 +332,7 @@ class DxilMDHelper {
   // Node Record Type
   static const unsigned kDxilNodeRecordSizeTag = 0;
   static const unsigned kDxilNodeSVDispatchGridTag = 1;
+  static const unsigned kDxilNodeRecordAlignmentTag = 2;
 
   // GSState.
   static const unsigned kDxilGSStateNumFields = 5;
@@ -624,6 +625,7 @@ class DxilMDHelper {
                        unsigned &payloadSizeInBytes);
 
   llvm::MDTuple *EmitDxilNodeIOState(const NodeIOProperties &Node);
+  llvm::MDTuple *EmitDxilNodeRecordType(const NodeRecordType &RecordType);
   hlsl::NodeIOProperties LoadDxilNodeIOState(const llvm::MDOperand &MDO);
   hlsl::NodeRecordType LoadDxilNodeRecordType(const llvm::MDOperand &MDO);
 

include/dxc/DXIL/DxilNodeProps.h

@@ -45,6 +45,7 @@ struct SVDispatchGrid {
 //
 struct NodeRecordType {
   unsigned size;
+  unsigned alignment;
   SVDispatchGrid SV_DispatchGrid;
 };