未验证 提交 3fc698fb 编写于 作者: Z Zhanlue Yang 提交者: GitHub

Adjust GPU Arches for next level Whl release strategy (#39910)

* Adjust GPU Arches for Whl releases

* Adjusted CUDA arches

* fixed minor issue

* adjusted gpu arches
上级 ebc6959c
...@@ -238,7 +238,8 @@ option(WITH_MIPS "Compile PaddlePaddle with mips support" OFF) ...@@ -238,7 +238,8 @@ option(WITH_MIPS "Compile PaddlePaddle with mips support" OFF)
option(WITH_MUSL "Compile with musl libc instead of gblic" OFF) option(WITH_MUSL "Compile with musl libc instead of gblic" OFF)
option(WITH_UNITY_BUILD "Compile with UnityBuild mode" OFF) option(WITH_UNITY_BUILD "Compile with UnityBuild mode" OFF)
option(WITH_STRIP "Strip so files of Whl packages" OFF) option(WITH_STRIP "Strip so files of Whl packages" OFF)
option(NEW_RELEASE_CUBIN "PaddlePaddle next-level release strategy for pypi cubin package" OFF) option(NEW_RELEASE_PYPI "PaddlePaddle next-level release strategy for pypi cubin package" OFF)
option(NEW_RELEASE_ALL "PaddlePaddle next-level release strategy for all arches cubin package" OFF)
option(NEW_RELEASE_JIT "PaddlePaddle next-level release strategy for backup jit package" OFF) option(NEW_RELEASE_JIT "PaddlePaddle next-level release strategy for backup jit package" OFF)
option(WITH_ASCEND_INT64 "Compile with int64 kernel for ascend NPU" OFF) option(WITH_ASCEND_INT64 "Compile with int64 kernel for ascend NPU" OFF)
option(WITH_POCKETFFT "Compile with pocketfft support" ON) option(WITH_POCKETFFT "Compile with pocketfft support" ON)
......
...@@ -6,16 +6,22 @@ if(WITH_NV_JETSON) ...@@ -6,16 +6,22 @@ if(WITH_NV_JETSON)
add_definitions(-DWITH_NV_JETSON) add_definitions(-DWITH_NV_JETSON)
set(paddle_known_gpu_archs "53 62 72") set(paddle_known_gpu_archs "53 62 72")
set(paddle_known_gpu_archs10 "53 62 72") set(paddle_known_gpu_archs10 "53 62 72")
elseif(NEW_RELEASE_CUBIN) elseif(NEW_RELEASE_ALL)
message("Using New Release Strategy - All Arches Packge")
add_definitions(-DNEW_RELEASE_ALL)
set(paddle_known_gpu_archs "35 50 52 60 61 70 75 80 86")
set(paddle_known_gpu_archs10 "35 50 52 60 61 70 75")
set(paddle_known_gpu_archs11 "35 50 52 60 61 70 75 80")
elseif(NEW_RELEASE_PYPI)
message("Using New Release Strategy - Cubin Packge") message("Using New Release Strategy - Cubin Packge")
add_definitions(-DNEW_RELEASE_CUBIN) add_definitions(-DNEW_RELEASE_PYPI)
set(paddle_known_gpu_archs "35 37 50 52 60 61 70 75 80 86") set(paddle_known_gpu_archs "35 50 52 60 61 70 75 80 86")
set(paddle_known_gpu_archs10 "50 60 70 75") set(paddle_known_gpu_archs10 "")
set(paddle_known_gpu_archs11 "60 70 75 80") set(paddle_known_gpu_archs11 "60 61 70 75 80")
elseif(NEW_RELEASE_JIT) elseif(NEW_RELEASE_JIT)
message("Using New Release Strategy - JIT Packge") message("Using New Release Strategy - JIT Packge")
add_definitions(-DNEW_RELEASE_JIT) add_definitions(-DNEW_RELEASE_JIT)
set(paddle_known_gpu_archs "35 37 50 52 60 61 70 75 80 86") set(paddle_known_gpu_archs "35 50 52 60 61 70 75 80 86")
set(paddle_known_gpu_archs10 "35 50 60 70 75") set(paddle_known_gpu_archs10 "35 50 60 70 75")
set(paddle_known_gpu_archs11 "35 50 60 70 75 80") set(paddle_known_gpu_archs11 "35 50 60 70 75 80")
else() else()
...@@ -148,7 +154,7 @@ function(select_nvcc_arch_flags out_variable) ...@@ -148,7 +154,7 @@ function(select_nvcc_arch_flags out_variable)
# remove dots and convert to lists # remove dots and convert to lists
string(REGEX REPLACE "\\." "" cuda_arch_bin "${cuda_arch_bin}") string(REGEX REPLACE "\\." "" cuda_arch_bin "${cuda_arch_bin}")
string(REGEX REPLACE "\\." "" cuda_arch_ptx "${CUDA_ARCH_PTX}") string(REGEX REPLACE "\\." "" cuda_arch_ptx "${cuda_arch_ptx}")
string(REGEX MATCHALL "[0-9()]+" cuda_arch_bin "${cuda_arch_bin}") string(REGEX MATCHALL "[0-9()]+" cuda_arch_bin "${cuda_arch_bin}")
string(REGEX MATCHALL "[0-9]+" cuda_arch_ptx "${cuda_arch_ptx}") string(REGEX MATCHALL "[0-9]+" cuda_arch_ptx "${cuda_arch_ptx}")
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册