onnxruntime/cmake/onnxruntime_mlas.cmake
Jan Scholz ce9acf0c21
iOS crosscompilation under linux (#3298)
* added support for ios crosscompilation under linux

* reverted cmake generator change

* if --ios is added protoc can be compiled for host system

* accidently reverted change to compile protoc for host system for ios if protoc exe is not set

* wdata is now used

* accidentally pasted CMAKE_OSX_ARCHITECTURES into CmakeLists.txt, also made bad merge on build.py previously

* removed print

* fixed typeo, deleted commented statements for earlier debugging

* reverted accidental delete

* added asmmacro.h for aarch64 asm
now MlasSgemmKernel**** gets underscore added if needed
no need anymote to differentiate between iOS arm64 and normal amr64 build
onnxruntime.cmake: added check if iOSCross is set to properly set RPATH

* removed 2 spaces

* fix: logcial error fixed, now protoc gets compiled if not supplied with --path_to_protoc_exe

* removed unecessarily added spaces

* removed some more spaces
2020-03-30 19:39:17 -07:00

265 lines
11 KiB
CMake

# Copyright (c) Microsoft Corporation. All rights reserved.
# Licensed under the MIT License.
set(mlas_common_srcs
${ONNXRUNTIME_ROOT}/core/mlas/lib/platform.cpp
${ONNXRUNTIME_ROOT}/core/mlas/lib/threading.cpp
${ONNXRUNTIME_ROOT}/core/mlas/lib/dgemm.cpp
${ONNXRUNTIME_ROOT}/core/mlas/lib/sgemm.cpp
${ONNXRUNTIME_ROOT}/core/mlas/lib/qgemm.cpp
${ONNXRUNTIME_ROOT}/core/mlas/lib/convolve.cpp
${ONNXRUNTIME_ROOT}/core/mlas/lib/pooling.cpp
${ONNXRUNTIME_ROOT}/core/mlas/lib/reorder.cpp
${ONNXRUNTIME_ROOT}/core/mlas/lib/snchwc.cpp
${ONNXRUNTIME_ROOT}/core/mlas/lib/activate.cpp
${ONNXRUNTIME_ROOT}/core/mlas/lib/logistic.cpp
${ONNXRUNTIME_ROOT}/core/mlas/lib/tanh.cpp
${ONNXRUNTIME_ROOT}/core/mlas/lib/erf.cpp
${ONNXRUNTIME_ROOT}/core/mlas/lib/quantize.cpp
)
if(MSVC)
if(onnxruntime_target_platform STREQUAL "ARM64")
set(asm_filename ${ONNXRUNTIME_ROOT}/core/mlas/lib/arm64/SgemmKernelNeon.asm)
set(pre_filename ${CMAKE_CURRENT_BINARY_DIR}/SgemmKernelNeon.i)
set(obj_filename ${CMAKE_CURRENT_BINARY_DIR}/SgemmKernelNeon.obj)
if(CMAKE_BUILD_TYPE STREQUAL "Debug")
set(ARMASM_FLAGS "-g")
else()
set(ARMASM_FLAGS "")
endif()
add_custom_command(
OUTPUT ${obj_filename}
COMMAND
cl.exe /P ${asm_filename}
COMMAND
armasm64.exe ${ARMASM_FLAGS} ${pre_filename} ${obj_filename}
)
set(mlas_platform_srcs ${obj_filename})
elseif(onnxruntime_target_platform STREQUAL "ARM")
set(mlas_platform_srcs
${ONNXRUNTIME_ROOT}/core/mlas/lib/arm/sgemmc.cpp
)
elseif(onnxruntime_target_platform STREQUAL "x64")
enable_language(ASM_MASM)
set(mlas_platform_srcs
${ONNXRUNTIME_ROOT}/core/mlas/lib/amd64/QgemmU8S8KernelAvx2.asm
${ONNXRUNTIME_ROOT}/core/mlas/lib/amd64/QgemvU8S8KernelAvx2.asm
${ONNXRUNTIME_ROOT}/core/mlas/lib/amd64/QgemmU8S8KernelAvx512Core.asm
${ONNXRUNTIME_ROOT}/core/mlas/lib/amd64/QgemvU8S8KernelAvx512Core.asm
${ONNXRUNTIME_ROOT}/core/mlas/lib/amd64/QgemmU8S8KernelAvx512Vnni.asm
${ONNXRUNTIME_ROOT}/core/mlas/lib/amd64/QgemvU8S8KernelAvx512Vnni.asm
${ONNXRUNTIME_ROOT}/core/mlas/lib/amd64/QgemmU8U8KernelAvx2.asm
${ONNXRUNTIME_ROOT}/core/mlas/lib/amd64/QgemmU8U8KernelAvx512Core.asm
${ONNXRUNTIME_ROOT}/core/mlas/lib/amd64/QgemmU8U8KernelAvx512Vnni.asm
${ONNXRUNTIME_ROOT}/core/mlas/lib/amd64/DgemmKernelSse2.asm
${ONNXRUNTIME_ROOT}/core/mlas/lib/amd64/DgemmKernelAvx.asm
${ONNXRUNTIME_ROOT}/core/mlas/lib/amd64/DgemmKernelFma3.asm
${ONNXRUNTIME_ROOT}/core/mlas/lib/amd64/DgemmKernelAvx512F.asm
${ONNXRUNTIME_ROOT}/core/mlas/lib/amd64/SgemmKernelSse2.asm
${ONNXRUNTIME_ROOT}/core/mlas/lib/amd64/SgemmKernelAvx.asm
${ONNXRUNTIME_ROOT}/core/mlas/lib/amd64/SgemmKernelM1Avx.asm
${ONNXRUNTIME_ROOT}/core/mlas/lib/amd64/SgemmKernelFma3.asm
${ONNXRUNTIME_ROOT}/core/mlas/lib/amd64/SgemmKernelAvx512F.asm
${ONNXRUNTIME_ROOT}/core/mlas/lib/amd64/SconvKernelSse2.asm
${ONNXRUNTIME_ROOT}/core/mlas/lib/amd64/SconvKernelAvx.asm
${ONNXRUNTIME_ROOT}/core/mlas/lib/amd64/SconvKernelFma3.asm
${ONNXRUNTIME_ROOT}/core/mlas/lib/amd64/SconvKernelAvx512F.asm
${ONNXRUNTIME_ROOT}/core/mlas/lib/amd64/SpoolKernelSse2.asm
${ONNXRUNTIME_ROOT}/core/mlas/lib/amd64/SpoolKernelAvx.asm
${ONNXRUNTIME_ROOT}/core/mlas/lib/amd64/SpoolKernelAvx512F.asm
${ONNXRUNTIME_ROOT}/core/mlas/lib/amd64/sgemma.asm
${ONNXRUNTIME_ROOT}/core/mlas/lib/amd64/cvtfp16a.asm
${ONNXRUNTIME_ROOT}/core/mlas/lib/amd64/LogisticKernelFma3.asm
${ONNXRUNTIME_ROOT}/core/mlas/lib/amd64/TanhKernelFma3.asm
${ONNXRUNTIME_ROOT}/core/mlas/lib/amd64/ErfKernelFma3.asm
)
else()
enable_language(ASM_MASM)
set(CMAKE_ASM_MASM_FLAGS "${CMAKE_ASM_MASM_FLAGS} /safeseh")
set(mlas_platform_srcs
${ONNXRUNTIME_ROOT}/core/mlas/lib/i386/SgemmKernelSse2.asm
${ONNXRUNTIME_ROOT}/core/mlas/lib/i386/SgemmKernelAvx.asm
)
endif()
else()
if (CMAKE_SYSTEM_NAME STREQUAL "Android")
if (CMAKE_ANDROID_ARCH_ABI STREQUAL "armeabi-v7a")
set(ARM TRUE)
elseif (CMAKE_ANDROID_ARCH_ABI STREQUAL "arm64-v8a")
set(ARM64 TRUE)
elseif (CMAKE_ANDROID_ARCH_ABI STREQUAL "x86_64")
set(X86_64 TRUE)
elseif (CMAKE_ANDROID_ARCH_ABI STREQUAL "x86")
set(X86 TRUE)
endif()
elseif(CMAKE_SYSTEM_NAME STREQUAL "iOSCross")
set(IOS TRUE)
if (CMAKE_OSX_ARCHITECTURES STREQUAL "arm64")
set(ARM64 TRUE)
elseif (CMAKE_OSX_ARCHITECTURES STREQUAL "arm")
set(ARM TRUE)
endif()
else()
execute_process(
COMMAND ${CMAKE_C_COMPILER} -dumpmachine
OUTPUT_VARIABLE dumpmachine_output
ERROR_QUIET
)
if(dumpmachine_output MATCHES "^arm.*")
set(ARM TRUE)
elseif(dumpmachine_output MATCHES "^aarch64.*")
set(ARM64 TRUE)
elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(i.86|x86?)$")
set(X86 TRUE)
elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(x86_64|amd64)$")
set(X86_64 TRUE)
endif()
endif()
if(ARM)
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mfpu=neon")
set(mlas_platform_srcs
${ONNXRUNTIME_ROOT}/core/mlas/lib/arm/sgemmc.cpp
)
elseif(ARM64)
enable_language(ASM)
set(mlas_platform_srcs
${ONNXRUNTIME_ROOT}/core/mlas/lib/aarch64/SgemmKernelNeon.S
)
elseif(X86)
enable_language(ASM)
set(mlas_platform_srcs_sse2
${ONNXRUNTIME_ROOT}/core/mlas/lib/x86/SgemmKernelSse2.S
)
set_source_files_properties(${mlas_platform_srcs_sse2} PROPERTIES COMPILE_FLAGS "-msse2")
set(mlas_platform_srcs_avx
${ONNXRUNTIME_ROOT}/core/mlas/lib/x86/SgemmKernelAvx.S
)
set_source_files_properties(${mlas_platform_srcs_avx} PROPERTIES COMPILE_FLAGS "-mavx")
set(mlas_platform_srcs
${mlas_platform_srcs_sse2}
${mlas_platform_srcs_avx}
)
elseif(X86_64)
enable_language(ASM)
# The LLVM assembler does not support the .arch directive to enable instruction
# set extensions and also doesn't support AVX-512F instructions without
# turning on support via command-line option. Group the sources by the
# instruction set extension and explicitly set the compiler flag as appropriate.
set(mlas_platform_srcs_sse2
${ONNXRUNTIME_ROOT}/core/mlas/lib/x86_64/DgemmKernelSse2.S
${ONNXRUNTIME_ROOT}/core/mlas/lib/x86_64/SgemmKernelSse2.S
${ONNXRUNTIME_ROOT}/core/mlas/lib/x86_64/SgemmTransposePackB16x4Sse2.S
${ONNXRUNTIME_ROOT}/core/mlas/lib/x86_64/SconvKernelSse2.S
${ONNXRUNTIME_ROOT}/core/mlas/lib/x86_64/SpoolKernelSse2.S
)
set_source_files_properties(${mlas_platform_srcs_sse2} PROPERTIES COMPILE_FLAGS "-msse2")
set(mlas_platform_srcs_avx
${ONNXRUNTIME_ROOT}/core/mlas/lib/x86_64/DgemmKernelAvx.S
${ONNXRUNTIME_ROOT}/core/mlas/lib/x86_64/SgemmKernelAvx.S
${ONNXRUNTIME_ROOT}/core/mlas/lib/x86_64/SgemmKernelM1Avx.S
${ONNXRUNTIME_ROOT}/core/mlas/lib/x86_64/SgemmKernelM1TransposeBAvx.S
${ONNXRUNTIME_ROOT}/core/mlas/lib/x86_64/SgemmTransposePackB16x4Avx.S
${ONNXRUNTIME_ROOT}/core/mlas/lib/x86_64/SconvKernelAvx.S
${ONNXRUNTIME_ROOT}/core/mlas/lib/x86_64/SpoolKernelAvx.S
)
set_source_files_properties(${mlas_platform_srcs_avx} PROPERTIES COMPILE_FLAGS "-mavx")
set(mlas_platform_srcs_avx2
${ONNXRUNTIME_ROOT}/core/mlas/lib/x86_64/QgemmU8S8KernelAvx2.S
${ONNXRUNTIME_ROOT}/core/mlas/lib/x86_64/QgemvU8S8KernelAvx2.S
${ONNXRUNTIME_ROOT}/core/mlas/lib/x86_64/QgemmU8U8KernelAvx2.S
${ONNXRUNTIME_ROOT}/core/mlas/lib/x86_64/DgemmKernelFma3.S
${ONNXRUNTIME_ROOT}/core/mlas/lib/x86_64/SgemmKernelFma3.S
${ONNXRUNTIME_ROOT}/core/mlas/lib/x86_64/SconvKernelFma3.S
${ONNXRUNTIME_ROOT}/core/mlas/lib/x86_64/LogisticKernelFma3.S
${ONNXRUNTIME_ROOT}/core/mlas/lib/x86_64/TanhKernelFma3.S
${ONNXRUNTIME_ROOT}/core/mlas/lib/x86_64/ErfKernelFma3.S
)
set_source_files_properties(${mlas_platform_srcs_avx2} PROPERTIES COMPILE_FLAGS "-mavx2 -mfma")
# Some toolchains do not support AVX512 compiler flags but are still able
# to build the sources. Other toolchains require the AVX512 compiler flags
# to be specified.
check_cxx_compiler_flag("-mavx512f" HAS_AVX512F)
if(HAS_AVX512F)
set(CMAKE_REQUIRED_FLAGS "-mavx512f")
else()
set(CMAKE_REQUIRED_FLAGS "")
endif()
check_cxx_source_compiles("
int main() {
asm(\"vpxord %zmm0,%zmm0,%zmm0\");
return 0;
}"
COMPILES_AVX512F
)
if(COMPILES_AVX512F)
set(mlas_platform_srcs_avx512f
${ONNXRUNTIME_ROOT}/core/mlas/lib/x86_64/DgemmKernelAvx512F.S
${ONNXRUNTIME_ROOT}/core/mlas/lib/x86_64/SgemmKernelAvx512F.S
${ONNXRUNTIME_ROOT}/core/mlas/lib/x86_64/SconvKernelAvx512F.S
${ONNXRUNTIME_ROOT}/core/mlas/lib/x86_64/SpoolKernelAvx512F.S
)
if(HAS_AVX512F)
set_source_files_properties(${mlas_platform_srcs_avx512f} PROPERTIES COMPILE_FLAGS "-mavx512f")
endif()
check_cxx_compiler_flag("-mavx512bw -mavx512dq -mavx512vl" HAS_AVX512CORE)
if(HAS_AVX512CORE)
set(CMAKE_REQUIRED_FLAGS "-mavx512bw -mavx512dq -mavx512vl")
endif()
check_cxx_source_compiles("
int main() {
asm(\"vpmaddwd %zmm0,%zmm0,%zmm0\"); // AVX512BW feature
asm(\"vandnps %xmm31,%xmm31,%xmm31\"); // AVX512DQ/AVX512VL feature
return 0;
}"
COMPILES_AVX512CORE
)
if(COMPILES_AVX512CORE)
set(mlas_platform_srcs_avx512core
${ONNXRUNTIME_ROOT}/core/mlas/lib/x86_64/QgemmU8S8KernelAvx512Core.S
${ONNXRUNTIME_ROOT}/core/mlas/lib/x86_64/QgemvU8S8KernelAvx512Core.S
${ONNXRUNTIME_ROOT}/core/mlas/lib/x86_64/QgemmU8S8KernelAvx512Vnni.S
${ONNXRUNTIME_ROOT}/core/mlas/lib/x86_64/QgemvU8S8KernelAvx512Vnni.S
${ONNXRUNTIME_ROOT}/core/mlas/lib/x86_64/QgemmU8U8KernelAvx512Core.S
${ONNXRUNTIME_ROOT}/core/mlas/lib/x86_64/QgemmU8U8KernelAvx512Vnni.S
)
if(HAS_AVX512CORE)
set_source_files_properties(${mlas_platform_srcs_avx512core} PROPERTIES COMPILE_FLAGS "-mavx512bw -mavx512dq -mavx512vl")
endif()
else()
set_source_files_properties(${mlas_common_srcs} PROPERTIES COMPILE_FLAGS "-DMLAS_AVX512CORE_UNSUPPORTED")
endif()
else()
set_source_files_properties(${mlas_common_srcs} PROPERTIES COMPILE_FLAGS "-DMLAS_AVX512F_UNSUPPORTED")
endif()
set(mlas_platform_srcs
${mlas_platform_srcs_sse2}
${mlas_platform_srcs_avx}
${mlas_platform_srcs_avx2}
${mlas_platform_srcs_avx512f}
${mlas_platform_srcs_avx512core}
)
endif()
endif()
add_library(onnxruntime_mlas STATIC ${mlas_common_srcs} ${mlas_platform_srcs})
target_include_directories(onnxruntime_mlas PRIVATE ${ONNXRUNTIME_ROOT}/core/mlas/inc ${ONNXRUNTIME_ROOT}/core/mlas/lib ${ONNXRUNTIME_ROOT}/core/mlas/lib/amd64)
set_target_properties(onnxruntime_mlas PROPERTIES FOLDER "ONNXRuntime")