diff --git a/ggml/CMakeLists.txt b/ggml/CMakeLists.txt index 06d371e09..1b3d98967 100644 --- a/ggml/CMakeLists.txt +++ b/ggml/CMakeLists.txt @@ -92,30 +92,33 @@ else() set(INS_ENB ON) endif() -option(GGML_CPU_HBM "ggml: use memkind for CPU HBM" OFF) -option(GGML_CPU_AARCH64 "ggml: use runtime weight conversion of Q4_0 to Q4_X_X" ON) - -option(GGML_AVX "ggml: enable AVX" ${INS_ENB}) -option(GGML_AVX_VNNI "ggml: enable AVX-VNNI" OFF) -option(GGML_AVX2 "ggml: enable AVX2" ${INS_ENB}) -option(GGML_AVX512 "ggml: enable AVX512" OFF) -option(GGML_AVX512_VBMI "ggml: enable AVX512-VBMI" OFF) -option(GGML_AVX512_VNNI "ggml: enable AVX512-VNNI" OFF) -option(GGML_AVX512_BF16 "ggml: enable AVX512-BF16" OFF) -option(GGML_AMX_TILE "ggml: enable AMX-TILE" OFF) -option(GGML_AMX_INT8 "ggml: enable AMX-INT8" OFF) -option(GGML_AMX_BF16 "ggml: enable AMX-BF16" OFF) -option(GGML_FMA "ggml: enable FMA" ${INS_ENB}) +option(GGML_CPU_HBM "ggml: use memkind for CPU HBM" OFF) +option(GGML_CPU_AARCH64 "ggml: use runtime weight conversion of Q4_0 to Q4_X_X" ON) +option(GGML_AVX "ggml: enable AVX" ${INS_ENB}) +option(GGML_AVX_VNNI "ggml: enable AVX-VNNI" OFF) +option(GGML_AVX2 "ggml: enable AVX2" ${INS_ENB}) +option(GGML_AVX512 "ggml: enable AVX512F" OFF) +option(GGML_AVX512_VBMI "ggml: enable AVX512-VBMI" OFF) +option(GGML_AVX512_VNNI "ggml: enable AVX512-VNNI" OFF) +option(GGML_AVX512_BF16 "ggml: enable AVX512-BF16" OFF) if (NOT MSVC) - option(GGML_F16C "ggml: enable F16C" ${INS_ENB}) # in MSVC F16C is implied with AVX2/AVX512 + # in MSVC F16C and FMA is implied with AVX2/AVX512 + option(GGML_FMA "ggml: enable FMA" ${INS_ENB}) + option(GGML_F16C "ggml: enable F16C" ${INS_ENB}) + # MSVC does not seem to support AMX + option(GGML_AMX_TILE "ggml: enable AMX-TILE" OFF) + option(GGML_AMX_INT8 "ggml: enable AMX-INT8" OFF) + option(GGML_AMX_BF16 "ggml: enable AMX-BF16" OFF) endif() -option(GGML_LASX "ggml: enable lasx" ON) -option(GGML_LSX "ggml: enable lsx" ON) -option(GGML_RVV "ggml: enable rvv" ON) -option(GGML_SVE "ggml: enable SVE" OFF) +option(GGML_LASX "ggml: enable lasx" ON) +option(GGML_LSX "ggml: enable lsx" ON) +option(GGML_RVV "ggml: enable rvv" ON) +option(GGML_SVE "ggml: enable SVE" OFF) +option(GGML_CPU_ALL_VARIANTS "ggml: build all variants of the CPU backend (requires GGML_BACKEND_DL)" OFF) + if (WIN32) - set(GGML_WIN_VER "0x602" CACHE STRING "ggml: Windows Version") + set(GGML_WIN_VER "0x602" CACHE STRING "ggml: Windows version") endif() # ggml core @@ -180,11 +183,7 @@ option(GGML_BUILD_EXAMPLES "ggml: build examples" ${GGML_STANDALONE}) set(CMAKE_C_STANDARD 11) set(CMAKE_C_STANDARD_REQUIRED true) -if (GGML_SYCL) - set(CMAKE_CXX_STANDARD 17) -else() - set(CMAKE_CXX_STANDARD 11) -endif() +set(CMAKE_CXX_STANDARD 17) set(CMAKE_CXX_STANDARD_REQUIRED true) set(THREADS_PREFER_PTHREAD_FLAG ON) diff --git a/ggml/src/CMakeLists.txt b/ggml/src/CMakeLists.txt index d4302df91..f07533fdb 100644 --- a/ggml/src/CMakeLists.txt +++ b/ggml/src/CMakeLists.txt @@ -294,12 +294,15 @@ if (GGML_CPU_ALL_VARIANTS) endif() ggml_add_cpu_backend_variant(sandybridge AVX) ggml_add_cpu_backend_variant(haswell AVX F16C AVX2 FMA) - ggml_add_cpu_backend_variant(alderlake AVX F16C AVX2 FMA AVX_VNNI) ggml_add_cpu_backend_variant(skylakex AVX F16C AVX2 FMA AVX512) ggml_add_cpu_backend_variant(icelake AVX F16C AVX2 FMA AVX512 AVX512_VBMI AVX512_VNNI) - ggml_add_cpu_backend_variant(sapphirerapids AVX F16C AVX2 FMA AVX512 AVX512_VBMI AVX512_VNNI AVX512_BF16 AMX_TILE AMX_INT8) + if (NOT MSVC) + # MSVC doesn't support AVX-VNNI or AMX + ggml_add_cpu_backend_variant(alderlake AVX F16C AVX2 FMA AVX_VNNI) + ggml_add_cpu_backend_variant(sapphirerapids AVX F16C AVX2 FMA AVX512 AVX512_VBMI AVX512_VNNI AVX512_BF16 AMX_TILE AMX_INT8) + endif() else () - ggml_add_cpu_backend_variant_impl() + ggml_add_cpu_backend_variant_impl("") endif() ggml_add_backend(BLAS) diff --git a/ggml/src/ggml-cpu/CMakeLists.txt b/ggml/src/ggml-cpu/CMakeLists.txt index 0e7bfe740..bc326c059 100644 --- a/ggml/src/ggml-cpu/CMakeLists.txt +++ b/ggml/src/ggml-cpu/CMakeLists.txt @@ -179,7 +179,7 @@ function(ggml_add_cpu_backend_variant_impl tag_name) if (MSVC) # instruction set detection for MSVC only if (GGML_NATIVE) - include(cmake/FindSIMD.cmake) + include(ggml-cpu/cmake/FindSIMD.cmake) endif () if (GGML_AVX512) list(APPEND ARCH_FLAGS /arch:AVX512) @@ -227,12 +227,10 @@ function(ggml_add_cpu_backend_variant_impl tag_name) list(APPEND ARCH_DEFINITIONS GGML_SSE42) endif() if (GGML_AVX_VNNI) - list(APPEND ARCH_DEFINITIONS __AVXVNNI__ GGML_AVX_VNNI) - if (CMAKE_C_COMPILER_ID STREQUAL "Clang") - list(APPEND ARCH_FLAGS -mavxvnni) - endif() + # MSVC generates AVX512 with AVX-VNNI intrinsics even with /arch:AVX2 + #list(APPEND ARCH_DEFINITIONS __AVXVNNI__ GGML_AVX_VNNI) endif() - else() + else () if (GGML_NATIVE) list(APPEND ARCH_FLAGS -march=native) else () @@ -340,13 +338,14 @@ function(ggml_add_cpu_backend_variant_impl tag_name) # The feature detection code is compiled as a separate target so that # it can be built without the architecture flags # Since multiple variants of the CPU backend may be included in the same - # build, using set_source_files_properties() is not possible - add_library(${GGML_CPU_NAME}_feats OBJECT ggml-cpu/cpu-feats-x86.cpp) - target_include_directories(${GGML_CPU_NAME}_feats PRIVATE . .. ../include) - target_compile_definitions(${GGML_CPU_NAME}_feats PRIVATE ${ARCH_DEFINITIONS}) - target_compile_definitions(${GGML_CPU_NAME}_feats PRIVATE GGML_BACKEND_DL GGML_BACKEND_BUILD GGML_BACKEND_SHARED) - set_target_properties(${GGML_CPU_NAME}_feats PROPERTIES POSITION_INDEPENDENT_CODE ON) - target_link_libraries(${GGML_CPU_NAME} PRIVATE ${GGML_CPU_NAME}_feats) + # build, using set_source_files_properties() to set the arch flags is not possible + set(GGML_CPU_FEATS_NAME ${GGML_CPU_NAME}-feats) + add_library(${GGML_CPU_FEATS_NAME} OBJECT ggml-cpu/cpu-feats-x86.cpp) + target_include_directories(${GGML_CPU_FEATS_NAME} PRIVATE . .. ../include) + target_compile_definitions(${GGML_CPU_FEATS_NAME} PRIVATE ${ARCH_DEFINITIONS}) + target_compile_definitions(${GGML_CPU_FEATS_NAME} PRIVATE GGML_BACKEND_DL GGML_BACKEND_BUILD GGML_BACKEND_SHARED) + set_target_properties(${GGML_CPU_FEATS_NAME} PROPERTIES POSITION_INDEPENDENT_CODE ON) + target_link_libraries(${GGML_CPU_NAME} PRIVATE ${GGML_CPU_FEATS_NAME}) endif() if (EMSCRIPTEN)