Minor CMake Changes

CHANGELOG.md

@@ -2,6 +2,12 @@
 
 Full documentation for hipTensor is available at [rocm.docs.amd.com/projects/hiptensor](https://rocm.docs.amd.com/projects/hipTensor/en/latest/index.html).
 
+## (Unreleased) hipTensor 1.5.0 for ROCm 6.4.0
+
+### Changes
+
+* Renamed the CMake option `AMDGPU_TARGETS` to `GPU_TARGETS`
+
 ## (Unreleased) hipTensor 1.4.0 for ROCm 6.3.0
 
 ### Additions

CMakeLists.txt

@@ -70,15 +70,6 @@ if( CMAKE_PROJECT_NAME STREQUAL "hiptensor" )
   option(BUILD_OFFLOAD_COMPRESS "Build hiptensor with offload compression" ON)
 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( STATUS "WARNING: BUILD_OFFLOAD_COMPRESS=ON but flag not supported by compiler. Ignoring option." )
-  endif()
-endif()
-
 # Setup output paths
 set(CMAKE_RUNTIME_OUTPUT_DIRECTORY  ${CMAKE_BINARY_DIR}/bin)
 set(CMAKE_LIBRARY_OUTPUT_DIRECTORY  ${CMAKE_BINARY_DIR}/lib)
@@ -106,25 +97,41 @@ else()
 endif()
 
 if (ADDRESS_SANITIZER_ENABLED)
-    set(CMAKE_NO_BUILTIN_CHRPATH ON)
-    rocm_check_target_ids(DEFAULT_AMDGPU_TARGETS
+    rocm_check_target_ids(DEFAULT_GPU_TARGETS
         TARGETS "gfx90a:xnack+;gfx942:xnack+" )
 else()
-    rocm_check_target_ids(DEFAULT_AMDGPU_TARGETS
+    rocm_check_target_ids(DEFAULT_GPU_TARGETS
         TARGETS "gfx908;gfx90a;gfx942" )
 endif()
 
-# Variable AMDGPU_TARGET must be a cached variable and must be specified before calling find_package(hip)
-# This is because hip-config.cmake sets --offload-arch via AMDGPU_TARGET cached variable __after__ setting
-# default cached variable AMDGPU_TARGET to "gfx900;gfx906;gfx908;gfx1100;gfx1101;gfx1102", where not all archs are compatible with MFMA instructions
+# 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( STATUS "WARNING: BUILD_OFFLOAD_COMPRESS=ON but flag not supported by compiler. Ignoring option." )
+    if(ADDRESS_SANITIZER_ENABLED)
+      set(CMAKE_NO_BUILTIN_CHRPATH ON)
+    endif()
+  elseif(CXX_COMPILER_SUPPORTS_OFFLOAD_COMPRESS)
+    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)
+# This is because hip-config.cmake sets --offload-arch via GPU_TARGET cached variable __after__ setting
+# default cached variable GPU_TARGET to "gfx900;gfx906;gfx908;gfx1100;gfx1101;gfx1102", where not all archs are compatible with MFMA instructions
 #
 # By rule, once cached variable is set, it cannot be overridden unless we use the FORCE option
-if(AMDGPU_TARGETS)
-  set(AMDGPU_TARGETS "${AMDGPU_TARGETS}" CACHE STRING "List of specific machine types for library to target")
+if(GPU_TARGETS)
+  set(GPU_TARGETS "${GPU_TARGETS}" CACHE STRING "List of specific machine types for library to target")
+elseif(AMDGPU_TARGETS)
+  set(GPU_TARGETS "${AMDGPU_TARGETS}" CACHE STRING "List of specific machine types for library to target")
+  message(STATUS "WARNING: AMDGPU_TARGETS use is deprecated. Use GPU_TARGETS.")
 else()
-  set(AMDGPU_TARGETS "${DEFAULT_AMDGPU_TARGETS}" CACHE STRING "List of specific machine types for library to target")
+  set(GPU_TARGETS "${DEFAULT_GPU_TARGETS}" CACHE STRING "List of specific machine types for library to target")
 endif()
-message( VERBOSE "AMDGPU_TARGETS=${AMDGPU_TARGETS}")
+message( VERBOSE "GPU_TARGETS=${GPU_TARGETS}")
 
 if(HIPTENSOR_DEFAULT_STRIDES_COL_MAJOR)
   add_compile_definitions(HIPTENSOR_DEFAULT_STRIDES_COL_MAJOR=1)

README.md

@@ -34,12 +34,13 @@ For more detailed information, please refer to the [hipTensor installation guide
 
 ### Project options
 
-| Option                          | Description                                       | Default Value                                                  |
-|---------------------------------|---------------------------------------------------|----------------------------------------------------------------|
-| AMDGPU_TARGETS                  | Build code for specific GPU target(s)             | gfx908:xnack-;gfx90a:xnack-;gfx90a:xnack+;gfx940;gfx941;gfx942 |
-| HIPTENSOR_BUILD_TESTS           | Build Tests                                       | ON                                                             |
-| HIPTENSOR_BUILD_SAMPLES         | Build Samples                                     | ON                                                             |
-| HIPTENSOR_BUILD_COMPRESSED_DBG  | Enable compressed debug symbols                   | ON                                                             |
+| Option                          | Description                                        | Default Value                                                  |
+|---------------------------------|----------------------------------------------------|----------------------------------------------------------------|
+| GPU_TARGETS                     | Build code for specific GPU target(s)              | gfx908:xnack-;gfx90a:xnack-;gfx90a:xnack+;gfx940;gfx941;gfx942 |
+| AMDGPU_TARGETS                  | (Deprecated) Build code for specific GPU target(s) | gfx908:xnack-;gfx90a:xnack-;gfx90a:xnack+;gfx940;gfx941;gfx942 |
+| HIPTENSOR_BUILD_TESTS           | Build Tests                                        | ON                                                             |
+| HIPTENSOR_BUILD_SAMPLES         | Build Samples                                      | ON                                                             |
+| HIPTENSOR_BUILD_COMPRESSED_DBG  | Enable compressed debug symbols                    | ON                                                             |
 | HIPTENSOR_DEFAULT_STRIDES_COL_MAJOR | Set hiptensor default data layout to column major | ON                                                             |
 
 ### Example configurations
@@ -49,7 +50,7 @@ By default, the project is configured as Release mode. Here are some of the exam
 | Configuration                    | Command                                                                   |
 |----------------------------------|---------------------------------------------------------------------------|
 | Basic                            | `CC=hipcc CXX=hipcc cmake -B<build_dir> .`                                |
-| Targeting gfx908                 | `CC=hipcc CXX=hipcc cmake -B<build_dir> . -DAMDGPU_TARGETS=gfx908:xnack-` |
+| Targeting gfx908                 | `CC=hipcc CXX=hipcc cmake -B<build_dir> . -DGPU_TARGETS=gfx908:xnack-` |
 | Debug build                      | `CC=hipcc CXX=hipcc cmake -B<build_dir> . -DCMAKE_BUILD_TYPE=Debug`       |
 | Build without tests (default on) | `CC=hipcc CXX=hipcc cmake -B<build_dir> . -DHIPTENSOR_BUILD_TESTS=OFF`    |
 

docs/install/installation.rst

@@ -158,7 +158,7 @@ Below are the project options available to build hipTensor library with or witho
     *   -   **Option**
         -   **Description**
         -   **Default Value**
-    *   -   AMDGPU_TARGETS
+    *   -   GPU_TARGETS
         -   Build code for specific GPU target(s)
         -   ``gfx908:xnack-``; ``gfx90a:xnack-``; ``gfx90a:xnack+``; ``gfx940``; ``gfx941``; ``gfx942``
     *   -   HIPTENSOR_BUILD_TESTS
@@ -184,7 +184,7 @@ Here are some example project configurations:
 +===================================+====================================================================================================================+
 |            Basic                  | :code:`CC=/opt/rocm/bin/amdclang CXX=/opt/rocm/bin/amdclang++ cmake -B<build_dir> .`                               |
 +-----------------------------------+--------------------------------------------------------------------------------------------------------------------+
-|        Targeting gfx908           | :code:`CC=/opt/rocm/bin/amdclang CXX=/opt/rocm/bin/amdclang++ cmake -B<build_dir> . -DAMDGPU_TARGETS=gfx908:xnack-`|
+|        Targeting gfx908           | :code:`CC=/opt/rocm/bin/amdclang CXX=/opt/rocm/bin/amdclang++ cmake -B<build_dir> . -DGPU_TARGETS=gfx908:xnack-`|
 +-----------------------------------+--------------------------------------------------------------------------------------------------------------------+
 |          Debug build              | :code:`CC=/opt/rocm/bin/amdclang CXX=/opt/rocm/bin/amdclang++ cmake -B<build_dir> . -DCMAKE_BUILD_TYPE=Debug`      |
 +-----------------------------------+--------------------------------------------------------------------------------------------------------------------+
@@ -437,7 +437,7 @@ Build performance
 
 Depending on the resources available to the build machine and the build configuration selected, hipTensor build times can be on the order of an hour or more. Here are some things you can do to reduce build times:
 
-* Target a specific GPU (e.g., ``-D AMDGPU_TARGETS=gfx908:xnack-``)
+* Target a specific GPU (e.g., ``-D GPU_TARGETS=gfx908:xnack-``)
 * Use lots of threads (e.g., ``-j32``)
 * If they aren't needed, specify either ``HIPTENSOR_BUILD_TESTS`` or ``HIPTENSOR_BUILD_SAMPLES`` as OFF to disable client builds.
 * During the ``make`` command, build a specific target, e.g: ``logger_test``.

library/src/CMakeLists.txt

@@ -40,7 +40,7 @@ function(add_hiptensor_component COMPONENT_NAME COMPONENT_SOURCES)
 
     # Build components as object files.
     # Make sure they have -fPIC and that they inherit the hip::device environment
-    # so that they build for all archs in AMDGPU_TARGETS
+    # so that they build for all archs in GPU_TARGETS
     add_library(${COMPONENT_NAME} OBJECT ${COMPONENT_SOURCES})
     target_compile_options(${COMPONENT_NAME} PRIVATE ${CLANG_DRIVER_MODE})
     target_link_options(${COMPONENT_NAME} PRIVATE ${CLANG_DRIVER_MODE})