mirror of
https://github.com/zebrajr/pytorch.git
synced 2025-12-07 00:21:07 +01:00
e1e6417d4c
Adds an accelerated version of the embedding_lookup_idx perfkernels. This is done via a python codegen file similarly to `caffe2/perfkernels/hp_emblookup_codegen.py` Pull Request resolved: https://github.com/pytorch/pytorch/pull/133995 Approved by: https://github.com/malfet, https://github.com/huydhn
75 lines
2.5 KiB
CMake
75 lines
2.5 KiB
CMake
if(INTERN_BUILD_MOBILE)
|
|
list(APPEND Caffe2_CPU_SRCS
|
|
"${CMAKE_CURRENT_SOURCE_DIR}/embedding_lookup_idx.cc"
|
|
)
|
|
set(Caffe2_CPU_SRCS ${Caffe2_CPU_SRCS} PARENT_SCOPE)
|
|
return()
|
|
endif()
|
|
|
|
# ---[ CPU files.
|
|
file(GLOB common_srcs *.cc)
|
|
file(GLOB avx_srcs *_avx.cc)
|
|
file(GLOB avx2_srcs *_avx2.cc)
|
|
file(GLOB avx512_srcs *_avx512.cc)
|
|
file(GLOB sve_srcs *_sve.cc)
|
|
# exclude avx, avx2, avx512, and sve srcs from common_srcs
|
|
exclude(common_srcs "${common_srcs}" ${avx_srcs})
|
|
exclude(common_srcs "${common_srcs}" ${avx2_srcs})
|
|
exclude(common_srcs "${common_srcs}" ${avx512_srcs})
|
|
exclude(common_srcs "${common_srcs}" ${sve_srcs})
|
|
|
|
# We will always build common srcs.
|
|
set(Caffe2_CPU_SRCS ${Caffe2_CPU_SRCS} ${common_srcs})
|
|
|
|
# We will only build the perf kernel files if the compiler supports avx2
|
|
# extensions.
|
|
if(CXX_AVX2_FOUND)
|
|
add_library(Caffe2_perfkernels_avx2 STATIC ${avx2_srcs})
|
|
target_link_libraries(Caffe2_perfkernels_avx2 PRIVATE c10)
|
|
|
|
if(MSVC AND NOT "${CMAKE_CXX_COMPILER_ID}" STREQUAL "Clang")
|
|
target_compile_options(Caffe2_perfkernels_avx2
|
|
PRIVATE "/arch:AVX2"
|
|
PRIVATE "/D__FMA__"
|
|
PRIVATE "/D__F16C__")
|
|
else()
|
|
target_compile_options(Caffe2_perfkernels_avx2
|
|
PRIVATE "-mavx2"
|
|
PRIVATE "-mfma"
|
|
PRIVATE "-mavx"
|
|
PRIVATE "-mf16c")
|
|
endif()
|
|
caffe2_interface_library(
|
|
Caffe2_perfkernels_avx2 Caffe2_perfkernels_avx2_interface)
|
|
list(APPEND
|
|
Caffe2_DEPENDENCY_WHOLE_LINK_LIBS
|
|
"Caffe2_perfkernels_avx2_interface")
|
|
endif()
|
|
|
|
# We will only build the SVE perfkernel files if the compiler supports SVE
|
|
# extensions.
|
|
if(CXX_SVE_FOUND)
|
|
add_library(Caffe2_perfkernels_sve STATIC ${sve_srcs})
|
|
target_link_libraries(Caffe2_perfkernels_sve PRIVATE c10)
|
|
install(TARGETS Caffe2_perfkernels_sve
|
|
ARCHIVE DESTINATION "${CMAKE_INSTALL_LIBDIR}")
|
|
|
|
target_compile_options(Caffe2_perfkernels_sve PRIVATE "-march=armv8-a+sve")
|
|
|
|
caffe2_interface_library(
|
|
Caffe2_perfkernels_sve Caffe2_perfkernels_sve_interface)
|
|
list(APPEND
|
|
Caffe2_DEPENDENCY_WHOLE_LINK_LIBS "Caffe2_perfkernels_sve_interface")
|
|
endif()
|
|
|
|
# TODO(jiayq): currently, we only implement the very base files for the
|
|
# perfkernels. This is because to implement avx and avx2 files, we actually
|
|
# need to set up different compilation units and this is a bit more involving
|
|
# in terms of CMakefile changes. This is a stop-gap solution until we get a
|
|
# more proper implementation.
|
|
|
|
set(Caffe2_CPU_SRCS ${Caffe2_CPU_SRCS} PARENT_SCOPE)
|
|
set(Caffe2_DEPENDENCY_WHOLE_LINK_LIBS
|
|
${Caffe2_DEPENDENCY_WHOLE_LINK_LIBS}
|
|
PARENT_SCOPE)
|