[hipTensor] hipTensor buildable on Windows

projects/hiptensor/CMakeLists.txt

@@ -29,25 +29,29 @@ message(STATUS "CMake version: ${CMAKE_VERSION}")
 
 # NOTE: This has to be initialized before the project() command appears
 # Set the default of CMAKE_BUILD_TYPE to be release, unless user specifies with -D.  MSVC_IDE does not use CMAKE_BUILD_TYPE
-if( NOT DEFINED CMAKE_CONFIGURATION_TYPES AND NOT DEFINED CMAKE_BUILD_TYPE )
-  set( CMAKE_BUILD_TYPE Release CACHE STRING "Choose the type of build, options are: Debug Release RelWithDebInfo MinSizeRel." )
+if(NOT DEFINED CMAKE_CONFIGURATION_TYPES AND NOT DEFINED CMAKE_BUILD_TYPE)
+    set(CMAKE_BUILD_TYPE Release CACHE STRING "Choose the type of build, options are: Debug Release RelWithDebInfo MinSizeRel.")
 endif()
 message(STATUS "CMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE}")
-set(CMAKE_EXPORT_COMPILE_COMMANDS ON CACHE BOOL "" FORCE )
+set(CMAKE_EXPORT_COMPILE_COMMANDS ON CACHE BOOL "" FORCE)
 
 if(NOT DEFINED CMAKE_INSTALL_PREFIX)
-  set(CMAKE_INSTALL_PREFIX "/opt/rocm" CACHE PATH "Install path prefix, prepended onto install directories")
+    if(NOT WIN32)
+        set(CMAKE_INSTALL_PREFIX "/opt/rocm" CACHE PATH "Install path prefix, prepended onto install directories")
+    else()
+        set(CMAKE_INSTALL_PREFIX "C:/dist/TheRock" CACHE PATH "Install path prefix, prepended onto install directories")
+    endif()
 endif()
-if( NOT DEFINED CPACK_PACKAGING_INSTALL_PREFIX )
-  set(CPACK_PACKAGING_INSTALL_PREFIX "${CMAKE_INSTALL_PREFIX}" CACHE PATH "Install path prefix for packages.")
-  set(CPACK_SET_DESTDIR OFF)
+if(NOT DEFINED CPACK_PACKAGING_INSTALL_PREFIX)
+    set(CPACK_PACKAGING_INSTALL_PREFIX "${CMAKE_INSTALL_PREFIX}" CACHE PATH "Install path prefix for packages.")
+    set(CPACK_SET_DESTDIR OFF)
 endif()
 
 # This helps cmake properly find hip-config.cmake
-list( APPEND CMAKE_PREFIX_PATH $ENV{ROCM_PATH} ${ROCM_PATH} /opt/rocm )
+list(APPEND CMAKE_PREFIX_PATH $ENV{ROCM_PATH} ${ROCM_PATH} /opt/rocm)
 # Append our library helper cmake path and the cmake path for hip (for convenience).
 # Users may override HIP path by specifying their own in CMAKE_MODULE_PATH
-list( APPEND CMAKE_MODULE_PATH ${CMAKE_CURRENT_SOURCE_DIR}/cmake )
+list(APPEND CMAKE_MODULE_PATH ${CMAKE_CURRENT_SOURCE_DIR}/cmake)
 
 # Build shared libs
 set(BUILD_SHARED_LIBS ON)
@@ -63,13 +67,13 @@ set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -g -O1 -ggdb")           # c
 set(CMAKE_CXX_FLAGS_MINSIZEREL "${CMAKE_CXX_FLAGS_MINSIZEREL} -O2 -DNDEBUG") # clang++ failed to build the project with the default -Os
 
 # Project options
-if( CMAKE_PROJECT_NAME STREQUAL "hiptensor" )
-  option( HIPTENSOR_BUILD_TESTS "Build hiptensor tests" ON )
-  option( HIPTENSOR_BUILD_SAMPLES "Build hiptensor samples" ON )
-  option( HIPTENSOR_BUILD_COMPRESSED_DBG "Enable compressed debug symbols" ON )
-  option( HIPTENSOR_DEFAULT_STRIDES_COL_MAJOR "Set hiptensor default strides to column major" ON )
-  option( BUILD_OFFLOAD_COMPRESS "Build hiptensor with offload compression" ON )
-  option( HIPTENSOR_CODE_COVERAGE "Build with code coverage flags (clang only)" OFF )
+if(CMAKE_PROJECT_NAME STREQUAL "hiptensor")
+    option(HIPTENSOR_BUILD_TESTS "Build hiptensor tests" ON)
+    option(HIPTENSOR_BUILD_SAMPLES "Build hiptensor samples" ON)
+    option(HIPTENSOR_BUILD_COMPRESSED_DBG "Enable compressed debug symbols" ON)
+    option(HIPTENSOR_DEFAULT_STRIDES_COL_MAJOR "Set hiptensor default strides to column major" ON)
+    option(BUILD_OFFLOAD_COMPRESS "Build hiptensor with offload compression" ON)
+    option(HIPTENSOR_CODE_COVERAGE "Build with code coverage flags (clang only)" OFF)
 endif()
 
 # Setup output paths
@@ -93,9 +97,10 @@ include(CheckCXXCompilerFlag)
 include(cmake/Macros/CheckF8.cmake)
 check_f8(F8_EXISTS)
 if(F8_EXISTS)
-    message( STATUS "Performing Test `__hip_fp8_e5m2` and `__hip_fp8_e4m3` - Success" )
+    message(STATUS "Performing Test `__hip_fp8_e5m2` and `__hip_fp8_e4m3` - Success")
 else()
-    message(FATAL_ERROR "The detected ROCm does not support data type `__hip_fp8_e5m2` or `__hip_fp8_e4m3`.")
+    # Inform but do not fail the build, as these data types are not required for the library to function.
+    message(STATUS "The detected ROCm does not support data type `__hip_fp8_e5m2` or `__hip_fp8_e4m3`.")
 endif()
 
 # check if asan is enabled
@@ -111,23 +116,23 @@ endif()
 include(cmake/Functions/hiptensorSupportedArchitectures.cmake)
 if (ADDRESS_SANITIZER_ENABLED)
     rocm_check_target_ids(DEFAULT_GPU_TARGETS
-        TARGETS "gfx90a:xnack+;gfx942:xnack+;gfx950:xnack+" )
+        TARGETS "gfx90a:xnack+;gfx942:xnack+;gfx950:xnack+")
 else()
-    hiptensor_get_supported_architectures(SUPPORTED_ARCHS)
+    hiptensor_get_default_architectures(DEFAULT_ARCHS)
     rocm_check_target_ids(DEFAULT_GPU_TARGETS
-        TARGETS ${SUPPORTED_ARCHS} )
+        TARGETS ${DEFAULT_ARCHS})
 endif()
 
 # Check if offload compression is supported
 include(CheckCXXCompilerFlag)
 if (BUILD_OFFLOAD_COMPRESS)
-  check_cxx_compiler_flag("--offload-compress" CXX_COMPILER_SUPPORTS_OFFLOAD_COMPRESS)
-  if (NOT CXX_COMPILER_SUPPORTS_OFFLOAD_COMPRESS)
-    message(WARNING "BUILD_OFFLOAD_COMPRESS=ON but flag not supported by compiler. Ignoring option." )
-    set(CMAKE_NO_BUILTIN_CHRPATH ON)
-  else()
-    set(CMAKE_NO_BUILTIN_CHRPATH OFF)
-  endif()
+    check_cxx_compiler_flag("--offload-compress" CXX_COMPILER_SUPPORTS_OFFLOAD_COMPRESS)
+    if (NOT CXX_COMPILER_SUPPORTS_OFFLOAD_COMPRESS)
+        message(WARNING "BUILD_OFFLOAD_COMPRESS=ON but flag not supported by compiler. Ignoring option.")
+        set(CMAKE_NO_BUILTIN_CHRPATH ON)
+    else()
+        set(CMAKE_NO_BUILTIN_CHRPATH OFF)
+    endif()
 endif()
 
 # Variable GPU_TARGET must be a cached variable and must be specified before calling find_package(hip)
@@ -136,42 +141,42 @@ endif()
 #
 # By rule, once cached variable is set, it cannot be overridden unless we use the FORCE option
 if(GPU_TARGETS)
-  set(GPU_TARGETS "${GPU_TARGETS}" CACHE STRING "List of specific machine types for library to target")
+    set(GPU_TARGETS "${GPU_TARGETS}" CACHE STRING "List of specific machine types for library to target")
 elseif(AMDGPU_TARGETS)
-  set(GPU_TARGETS "${AMDGPU_TARGETS}" CACHE STRING "List of specific machine types for library to target")
-  message(DEPRECATION "AMDGPU_TARGETS use is deprecated. Use GPU_TARGETS.")
+    set(GPU_TARGETS "${AMDGPU_TARGETS}" CACHE STRING "List of specific machine types for library to target")
+    message(DEPRECATION "AMDGPU_TARGETS use is deprecated. Use GPU_TARGETS.")
 else()
-  set(GPU_TARGETS "${DEFAULT_GPU_TARGETS}" CACHE STRING "List of specific machine types for library to target")
+    set(GPU_TARGETS "${DEFAULT_GPU_TARGETS}" CACHE STRING "List of specific machine types for library to target")
 endif()
 message(STATUS "GPU_TARGETS=${GPU_TARGETS}")
 
 if(HIPTENSOR_DEFAULT_STRIDES_COL_MAJOR)
-  add_compile_definitions(HIPTENSOR_DEFAULT_STRIDES_COL_MAJOR=1)
+    add_compile_definitions(HIPTENSOR_DEFAULT_STRIDES_COL_MAJOR=1)
 else()
-  add_compile_definitions(HIPTENSOR_DEFAULT_STRIDES_COL_MAJOR=0)
+    add_compile_definitions(HIPTENSOR_DEFAULT_STRIDES_COL_MAJOR=0)
 endif()
 message(STATUS "HIPTENSOR_DEFAULT_STRIDES_COL_MAJOR=${HIPTENSOR_DEFAULT_STRIDES_COL_MAJOR}")
 
 # Setup HIP
-find_package(hip REQUIRED )
+find_package(hip REQUIRED)
 math(EXPR hip_VERSION_FLAT "(${hip_VERSION_MAJOR} * 1000 + ${hip_VERSION_MINOR}) * 100000 + ${hip_VERSION_PATCH}")
 message(STATUS "hip_version_flat=${hip_VERSION_FLAT}")
 
 # No assumption that HIP kernels are launched with uniform block size for backward compatibility
 # SWDEV-413293 and https://reviews.llvm.org/D155213
 if(NOT WIN32 AND ${hip_VERSION_FLAT} GREATER 500723302)
-  message(STATUS "Adding the fno-offload-uniform-block compiler flag")
-  add_compile_options(-fno-offload-uniform-block)
+    message(STATUS "Adding the fno-offload-uniform-block compiler flag")
+    add_compile_options(-fno-offload-uniform-block)
 endif()
 # Add optimization flags needed by backend
 if(NOT WIN32 AND ${hip_VERSION_FLAT} GREATER 600140090)
-  message(STATUS "Adding the enable-post-misched=0 compiler flag")
-  add_compile_options("SHELL: -mllvm -enable-post-misched=0")
+    message(STATUS "Adding the enable-post-misched=0 compiler flag")
+    add_compile_options("SHELL: -mllvm -enable-post-misched=0")
 endif()
 if(NOT WIN32 AND ${hip_VERSION_FLAT} GREATER 600241132)
-  message(STATUS "Adding -amdgpu-early-inline-all=true and -amdgpu-function-calls=false")
-  add_compile_options("SHELL: -mllvm -amdgpu-early-inline-all=true")
-  add_compile_options("SHELL: -mllvm -amdgpu-function-calls=false")
+    message(STATUS "Adding -amdgpu-early-inline-all=true and -amdgpu-function-calls=false")
+    add_compile_options("SHELL: -mllvm -amdgpu-early-inline-all=true")
+    add_compile_options("SHELL: -mllvm -amdgpu-function-calls=false")
 endif()
 
 if(HIPTENSOR_CODE_COVERAGE)
@@ -190,29 +195,29 @@ add_subdirectory(library/src)
 
 # Configure testing setup
 if(HIPTENSOR_BUILD_TESTS OR HIPTENSOR_BUILD_SAMPLES)
-  enable_testing()
-  rocm_package_setup_component(clients)
+    enable_testing()
+    rocm_package_setup_component(clients)
 endif()
 
 # Configure tests build
 if(HIPTENSOR_BUILD_TESTS)
-  rocm_package_setup_client_component(tests)
-  add_subdirectory(test)
+    rocm_package_setup_client_component(tests)
+    add_subdirectory(test)
 endif()
 
 # Configure clients build
 if(HIPTENSOR_BUILD_SAMPLES)
-  rocm_package_setup_client_component(samples)
-  add_subdirectory(samples)
+    rocm_package_setup_client_component(samples)
+    add_subdirectory(samples)
 endif()
 
 # Versioning via rocm-cmake
-set ( VERSION_STRING "2.2.0" )
-rocm_setup_version( VERSION ${VERSION_STRING} )
+set (VERSION_STRING "2.2.0")
+rocm_setup_version(VERSION ${VERSION_STRING})
 
 # Configure a header file to pass the CMake version settings to the source
 configure_file("${CMAKE_CURRENT_SOURCE_DIR}/library/include/hiptensor/internal/hiptensor-version.h.in"
-               "${CMAKE_CURRENT_SOURCE_DIR}/library/include/hiptensor/internal/hiptensor-version.h" )
+               "${CMAKE_CURRENT_SOURCE_DIR}/library/include/hiptensor/internal/hiptensor-version.h")
 
 # Define package dependencies for export
 set(PACKAGE_DEPENDS

projects/hiptensor/cmake/Functions/hiptensorSupportedArchitectures.cmake

@@ -17,6 +17,15 @@ set(SUPPORTED_ARCHITECTURES
     gfx12-generic
 )
 
+set(DEFAULT_ARCHITECTURES
+    gfx908
+    gfx90a
+    gfx942
+    gfx950
+    gfx11-generic
+    gfx12-generic
+)
+
 function(hiptensor_is_supported_architecture ARCHITECTURE_NAME RESULT_VAR)
     list(FIND SUPPORTED_ARCHITECTURES "${ARCHITECTURE_NAME}" INDEX)
     if (INDEX EQUAL -1)
@@ -26,7 +35,7 @@ function(hiptensor_is_supported_architecture ARCHITECTURE_NAME RESULT_VAR)
     endif()
 endfunction()
 
-function(hiptensor_get_supported_architectures RESULT_VAR)
-    string(JOIN ";" SUPPORTED_ARCHITECTURES_STRING ${SUPPORTED_ARCHITECTURES})
-    set(${RESULT_VAR} "${SUPPORTED_ARCHITECTURES_STRING}" PARENT_SCOPE)
+function(hiptensor_get_default_architectures RESULT_VAR)
+    string(JOIN ";" DEFAULT_ARCHITECTURES_STRING ${DEFAULT_ARCHITECTURES})
+    set(${RESULT_VAR} "${DEFAULT_ARCHITECTURES_STRING}" PARENT_SCOPE)
 endfunction()