From 94b1fc8c656c5e084d18c349bf002b1aa7026681 Mon Sep 17 00:00:00 2001 From: nihuini Date: Thu, 23 May 2024 12:04:21 +0800 Subject: [PATCH] build --- CMakeLists.txt | 31 ++++++++++++++++++++++++++----- 1 file changed, 26 insertions(+), 5 deletions(-) diff --git a/CMakeLists.txt b/CMakeLists.txt index e4d4877e056..586ce136da3 100644 --- a/CMakeLists.txt +++ b/CMakeLists.txt @@ -447,12 +447,9 @@ else() option(NCNN_SSE2 "optimize x86 platform with sse2 extension" ON) - if(CMAKE_CXX_COMPILER_ID MATCHES "MSVC" OR (CMAKE_CXX_COMPILER_ID MATCHES "Clang" AND CMAKE_CXX_SIMULATE_ID MATCHES "MSVC" AND CMAKE_CXX_COMPILER_FRONTEND_VARIANT MATCHES "MSVC")) + if(CMAKE_CXX_COMPILER_ID MATCHES "MSVC") check_cxx_compiler_flag("/arch:AVX" NCNN_COMPILER_SUPPORT_X86_AVX) - - set(CMAKE_REQUIRED_FLAGS "/arch:AVX -mfma") - check_cxx_source_compiles("#include \nint main() { __m256 _s, _a, _b; _s = _mm256_fmadd_ps(_a, _b, _s); return 0; }" NCNN_COMPILER_SUPPORT_X86_FMA) - + check_cxx_compiler_flag("/arch:AVX" NCNN_COMPILER_SUPPORT_X86_FMA) check_cxx_compiler_flag("/arch:AVX" NCNN_COMPILER_SUPPORT_X86_XOP) check_cxx_compiler_flag("/arch:AVX" NCNN_COMPILER_SUPPORT_X86_F16C) check_cxx_compiler_flag("/arch:AVX2" NCNN_COMPILER_SUPPORT_X86_AVX2) @@ -470,6 +467,30 @@ else() set(CMAKE_REQUIRED_FLAGS "/arch:AVX512") check_cxx_source_compiles("#include \nint main() { __m512h _s, _a, _b; _s = _mm512_fmadd_ph(_s, _a, _b); __m512 _s2; _s2 = _mm512_cvtxph_ps(_mm512_cvtxps_ph(_s2)); return 0; }" NCNN_COMPILER_SUPPORT_X86_AVX512_FP16) + unset(CMAKE_REQUIRED_FLAGS) + elseif(CMAKE_CXX_COMPILER_ID MATCHES "Clang" AND CMAKE_CXX_SIMULATE_ID MATCHES "MSVC" AND CMAKE_CXX_COMPILER_FRONTEND_VARIANT MATCHES "MSVC") + check_cxx_compiler_flag("/arch:AVX" NCNN_COMPILER_SUPPORT_X86_AVX) + + set(CMAKE_REQUIRED_FLAGS "/arch:AVX -mfma -mf16c") + check_cxx_source_compiles("#include \nint main() { __m256 _s, _a, _b; _s = _mm256_fmadd_ps(_a, _b, _s); return 0; }" NCNN_COMPILER_SUPPORT_X86_FMA) + + check_cxx_compiler_flag("/arch:AVX -mxop" NCNN_COMPILER_SUPPORT_X86_XOP) + check_cxx_compiler_flag("/arch:AVX -mf16c" NCNN_COMPILER_SUPPORT_X86_F16C) + check_cxx_compiler_flag("/arch:AVX2" NCNN_COMPILER_SUPPORT_X86_AVX2) + check_cxx_compiler_flag("/arch:AVX512 -mavx512cd -mavx512bw -mavx512dq -mavx512vl" NCNN_COMPILER_SUPPORT_X86_AVX512) + + set(CMAKE_REQUIRED_FLAGS "/arch:AVX2 -mavxvnni") + check_cxx_source_compiles("#include \nint main() { __m256i _s, _a, _b; _s = _mm256_dpwssd_epi32(_s, _a, _b); return 0; }" NCNN_COMPILER_SUPPORT_X86_AVX_VNNI) + + set(CMAKE_REQUIRED_FLAGS "/arch:AVX512 -mavx512cd -mavx512bw -mavx512dq -mavx512vl -mavx512vnni") + check_cxx_source_compiles("#include \nint main() { __m512i _s, _a, _b; _s = _mm512_dpwssd_epi32(_s, _a, _b); return 0; }" NCNN_COMPILER_SUPPORT_X86_AVX512_VNNI) + + set(CMAKE_REQUIRED_FLAGS "/arch:AVX512 -mavx512cd -mavx512bw -mavx512dq -mavx512vl -mavx512bf16") + check_cxx_source_compiles("#include \nint main() { __m256bh _s; __m512bh _a, _b; _s = _mm512_cvtneps_pbh(_mm512_dpbf16_ps(_mm512_cvtpbh_ps(_s), _a, _b)); return 0; }" NCNN_COMPILER_SUPPORT_X86_AVX512_BF16) + + set(CMAKE_REQUIRED_FLAGS "/arch:AVX512 -mavx512cd -mavx512bw -mavx512dq -mavx512vl -mavx512fp16") + check_cxx_source_compiles("#include \nint main() { __m512h _s, _a, _b; _s = _mm512_fmadd_ph(_s, _a, _b); __m512 _s2; _s2 = _mm512_cvtxph_ps(_mm512_cvtxps_ph(_s2)); return 0; }" NCNN_COMPILER_SUPPORT_X86_AVX512_FP16) + unset(CMAKE_REQUIRED_FLAGS) else() check_cxx_compiler_flag("-mavx" NCNN_COMPILER_SUPPORT_X86_AVX)