frozenleaves/pytorch
1
0
Fork 0
mirror of https://github.com/pytorch/pytorch.git synced 2025-10-20 12:54:11 +08:00
Code Issues Packages Projects Releases Wiki Activity

[CMake] Fix USE_FBGEMM_GENAI option (#164165)

Browse Source 
----

- `cmake_dependent_option` condition should be `USE_ROCM OR (USE_CUDA AND NOT MSVC)` (similar to the one for flash attention)
- Default settings should be user overridable, i.e. even if one builds for SM_10, they should be able to pass `USE_FBGEMM_GENAI=0` and skip the build

Pull Request resolved: https://github.com/pytorch/pytorch/pull/164165
Approved by: https://github.com/Skylion007
This commit is contained in:
Nikita Shulga
2025-09-29 11:51:52 -07:00
committed by PyTorch MergeBot
parent b7419b920d
commit 55840fb4bb
2 changed files with 14 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
CMakeLists.txt
View File

@ -888,23 +888,28 @@ cmake_dependent_option(
"(USE_CUDA AND NOT MSVC) OR USE_ROCM"
OFF)
IF(USE_ROCM AND "gfx942" IN_LIST PYTORCH_ROCM_ARCH)
message(WARNING "Setting USE_FBGEMM_GENAI for gfx942 to ON by default, doing ROCM build")
set(USE_FBGEMM_GENAI_DEFAULT ON)
elseif(USE_CUDA AND "$ENV{TORCH_CUDA_ARCH_LIST}" MATCHES "10.0" AND CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL 12.8 AND NOT WIN32)
message(STATUS "Setting USE_FBGEMM_GENAI to ON by default , doing CUDA build for SM100a")
set(USE_FBGEMM_GENAI_DEFAULT ON)
else()
set(USE_FBGEMM_GENAI_DEFAULT OFF)
endif()
cmake_dependent_option(
USE_FBGEMM_GENAI
"Whether to build FBGEMM GenAI quantized GEMM kernels.\
Will be disabled if not supported by the platform"
ON
"USE_ROCM"
${USE_FBGEMM_GENAI_DEFAULT}
"(USE_CUDA AND NOT MSVC) OR USE_ROCM"
OFF)
IF(USE_FBGEMM_GENAI AND USE_ROCM AND NOT "gfx942" IN_LIST PYTORCH_ROCM_ARCH)
message(WARNING "Unsupported ROCM arch for FBGEMM GenAI, will set USE_FBGEMM_GENAI to OFF")
set(USE_FBGEMM_GENAI off)
endif()
# Set USE_FBGEMM_GENAI to ON for CUDA build on SM100.
if(USE_CUDA AND "$ENV{TORCH_CUDA_ARCH_LIST}" MATCHES "10.0" AND CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL 12.8 AND NOT WIN32)
message(STATUS "Setting USE_FBGEMM_GENAI to ON, doing CUDA build for SM100a")
set(USE_FBGEMM_GENAI ON)
endif()
# CAVEAT: Again, Flash Attention2 will error while building for sm52 while Mem

1
cmake/Summary.cmake
View File

@ -138,6 +138,7 @@ function(caffe2_print_configuration_summary)
message(STATUS " USE_EIGEN_FOR_BLAS : ${CAFFE2_USE_EIGEN_FOR_BLAS}")
message(STATUS " USE_EIGEN_FOR_SPARSE : ${USE_EIGEN_SPARSE}")
message(STATUS " USE_FBGEMM : ${USE_FBGEMM}")
message(STATUS " USE_FBGEMM_GENAI : ${USE_FBGEMM_GENAI}")
message(STATUS " USE_KINETO : ${USE_KINETO}")
message(STATUS " USE_GFLAGS : ${USE_GFLAGS}")
message(STATUS " USE_GLOG : ${USE_GLOG}")