mindspore-Ecosystem
/
mindspore
21
0
Fork 27
Code Issues Pull Requests 6 Projects Releases Wiki Activity

update gpu build

This commit is contained in:
VectorSL 2022-08-22 15:56:35 +08:00
parent 4593e23a06
commit 087220da91
1 changed files with 22 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

34
mindspore/ccsrc/CMakeLists.txt
View File

@ -57,20 +57,26 @@ function(set_nvcc_flag CUDA_NVCC_FLAGS)
# Get build flag from env to choose common/auto build.
set(NVCC_ARCH_FLAG_FROM_ENV $ENV{CUDA_ARCH})
if(NVCC_ARCH_FLAG_FROM_ENV STREQUAL "common")
foreach(arch ${cuda_archs_bin})
if(arch VERSION_LESS "7.0") # For common build, we need a Volta arch at least.
message(FATAL_ERROR "The device arch must >= 7.0 to build common archs. But got " ${arch}
" Using -G auto is recommended to detect device arch automatically.")
endif()
endforeach()
message("Build common archs for release.")
list(APPEND CUDA_NVCC_FLAGS -gencode=arch=compute_53,code=sm_53
-gencode=arch=compute_60,code=sm_60
-gencode=arch=compute_62,code=sm_62
-gencode=arch=compute_70,code=sm_70
-gencode=arch=compute_72,code=sm_72
-gencode=arch=compute_75,code=compute_75
--expt-relaxed-constexpr)
-gencode=arch=compute_72,code=sm_72)
if(${CUDA_VERSION} VERSION_GREATER "9.5")
list(APPEND CUDA_NVCC_FLAGS -gencode=arch=compute_75,code=sm_75)
if(${CUDA_VERSION} VERSION_LESS "11.0")
list(APPEND CUDA_NVCC_FLAGS -gencode=arch=compute_75,code=compute_75)
endif()
endif()
if(${CUDA_VERSION} VERSION_GREATER "10.5")
list(APPEND CUDA_NVCC_FLAGS -gencode=arch=compute_80,code=sm_80)
if(${CUDA_VERSION} VERSION_LESS "11.1")
list(APPEND CUDA_NVCC_FLAGS -gencode=arch=compute_80,code=compute_80)
endif()
endif()
if(NOT ${CUDA_VERSION} VERSION_LESS "11.1")
list(APPEND CUDA_NVCC_FLAGS -gencode=arch=compute_86,code=compute_86)
endif()
else()
message("Auto build for arch(s) " ${cuda_archs_bin})
string(REGEX REPLACE "\\." "" cuda_archs_bin "${cuda_archs_bin}")
@ -78,12 +84,16 @@ function(set_nvcc_flag CUDA_NVCC_FLAGS)
foreach(arch ${cuda_archs_bin})
list(APPEND CUDA_NVCC_FLAGS -gencode=arch=compute_${arch},code=sm_${arch})
endforeach()
list(APPEND CUDA_NVCC_FLAGS --expt-relaxed-constexpr)
# For auto build, it only generate the code for indeed arch, so add sm_53 as a default arch
# to avoid error in different archs. It may increase the compilation time.
list(APPEND CUDA_NVCC_FLAGS -arch=sm_53)
endif()
message("Final CUDA_NVCC_FLASG " ${CUDA_NVCC_FLAGS})
else()
message("Failed to detect gpu arch automatically, build a base arch 5.3.")
list(APPEND CUDA_NVCC_FLAGS -arch=sm_53 --expt-relaxed-constexpr)
list(APPEND CUDA_NVCC_FLAGS -arch=sm_53)
endif()
list(APPEND CUDA_NVCC_FLAGS --expt-relaxed-constexpr)
set(${CUDA_NVCC_FLAGS} ${${CUDA_NVCC_FLAGS}} PARENT_SCOPE)
endfunction()