ggml-cpu: Add CPU backend support for KleidiAI library (#11390)
* ggml-cpu: Add CPU backend support for KleidiAI library * Add environmental variable GGML_KLEIDIAI_SME * Add support for multithread LHS conversion * Switch kernel selection order to dotprod and i8mm * updates for review comments * More updates for review comments * Reorganize and rename KleidiAI files * Move ggml-cpu-traits.h to source file * Update cmake for SME build and add alignment for SME * Remove append GGML_USE_CPU_KLEIDIAI to the GGML_CDEF_PUBLIC list
This commit is contained in:
parent
4806498bf1
commit
c5d91a7400
9 changed files with 767 additions and 10 deletions
|
@ -111,14 +111,15 @@ function(ggml_add_cpu_backend_variant_impl tag_name)
|
|||
function(check_arm_feature tag code)
|
||||
set(CMAKE_REQUIRED_FLAGS_SAVE ${CMAKE_REQUIRED_FLAGS})
|
||||
set(CMAKE_REQUIRED_FLAGS "${ARM_MCPU_FLAG}+${tag}")
|
||||
check_cxx_source_runs(
|
||||
"${code}"
|
||||
GGML_MACHINE_SUPPORTS_${tag}
|
||||
)
|
||||
check_cxx_source_runs("${code}" GGML_MACHINE_SUPPORTS_${tag})
|
||||
if (GGML_MACHINE_SUPPORTS_${tag})
|
||||
set(ARM_MCPU_FLAG_FIX "${ARM_MCPU_FLAG_FIX}+${tag}" PARENT_SCOPE)
|
||||
else()
|
||||
set(ARM_MCPU_FLAG_FIX "${ARM_MCPU_FLAG_FIX}+no${tag}" PARENT_SCOPE)
|
||||
set(CMAKE_REQUIRED_FLAGS "${ARM_MCPU_FLAG}+no${tag}")
|
||||
check_cxx_source_compiles("int main() { return 0; }" GGML_MACHINE_SUPPORTS_no${tag})
|
||||
if (GGML_MACHINE_SUPPORTS_no${tag})
|
||||
set(ARM_MCPU_FLAG_FIX "${ARM_MCPU_FLAG_FIX}+no${tag}" PARENT_SCOPE)
|
||||
endif()
|
||||
endif()
|
||||
set(CMAKE_REQUIRED_FLAGS ${CMAKE_REQUIRED_FLAGS_SAVE})
|
||||
endfunction()
|
||||
|
@ -126,6 +127,7 @@ function(ggml_add_cpu_backend_variant_impl tag_name)
|
|||
check_arm_feature(dotprod "#include <arm_neon.h>\nint main() { int8x16_t _a, _b; volatile int32x4_t _s = vdotq_s32(_s, _a, _b); return 0; }")
|
||||
check_arm_feature(i8mm "#include <arm_neon.h>\nint main() { int8x16_t _a, _b; volatile int32x4_t _s = vmmlaq_s32(_s, _a, _b); return 0; }")
|
||||
check_arm_feature(sve "#include <arm_sve.h>\nint main() { svfloat32_t _a, _b; volatile svfloat32_t _c = svadd_f32_z(svptrue_b8(), _a, _b); return 0; }")
|
||||
check_arm_feature(sme "#include <arm_sme.h>\n__arm_locally_streaming int main() { __asm__ volatile(\"smstart; smstop;\"); return 0; }")
|
||||
|
||||
list(APPEND ARCH_FLAGS "${ARM_MCPU_FLAG}${ARM_MCPU_FLAG_FIX}")
|
||||
else()
|
||||
|
@ -150,7 +152,7 @@ function(ggml_add_cpu_backend_variant_impl tag_name)
|
|||
if (ARM_FEATURE_RESULT)
|
||||
message(WARNING "Failed to get ARM features")
|
||||
else()
|
||||
foreach(feature DOTPROD SVE MATMUL_INT8 FMA FP16_VECTOR_ARITHMETIC)
|
||||
foreach(feature DOTPROD SVE MATMUL_INT8 FMA FP16_VECTOR_ARITHMETIC SME)
|
||||
string(FIND "${ARM_FEATURE}" "__ARM_FEATURE_${feature} 1" feature_pos)
|
||||
if (NOT ${feature_pos} EQUAL -1)
|
||||
message(STATUS "ARM feature ${feature} enabled")
|
||||
|
@ -316,6 +318,94 @@ function(ggml_add_cpu_backend_variant_impl tag_name)
|
|||
target_compile_definitions(${GGML_CPU_NAME} PRIVATE GGML_USE_CPU_AARCH64)
|
||||
endif()
|
||||
|
||||
if (GGML_CPU_KLEIDIAI)
|
||||
message(STATUS "Using KleidiAI optimized kernels if applicable")
|
||||
|
||||
# Disable the KleidiAI tests
|
||||
set(KLEIDIAI_BUILD_TESTS OFF)
|
||||
|
||||
# Fetch KleidiAI sources:
|
||||
include(FetchContent)
|
||||
set(KLEIDIAI_COMMIT_TAG "v1.3.0")
|
||||
set(KLEIDIAI_DOWNLOAD_URL "https://github.com/ARM-software/kleidiai/archive/refs/tags/${KLEIDIAI_COMMIT_TAG}.tar.gz")
|
||||
set(KLEIDIAI_ARCHIVE_MD5 "060bd2dc64642b091f461cc8dd7426d9")
|
||||
|
||||
if (POLICY CMP0135)
|
||||
cmake_policy(SET CMP0135 NEW)
|
||||
endif()
|
||||
|
||||
FetchContent_Declare(KleidiAI_Download
|
||||
URL ${KLEIDIAI_DOWNLOAD_URL}
|
||||
DOWNLOAD_EXTRACT_TIMESTAMP NEW
|
||||
URL_HASH MD5=${KLEIDIAI_ARCHIVE_MD5})
|
||||
|
||||
FetchContent_MakeAvailable(KleidiAI_Download)
|
||||
FetchContent_GetProperties(KleidiAI_Download
|
||||
SOURCE_DIR KLEIDIAI_SRC
|
||||
POPULATED KLEIDIAI_POPULATED)
|
||||
|
||||
if (NOT KLEIDIAI_POPULATED)
|
||||
message(FATAL_ERROR "KleidiAI source downloaded failed.")
|
||||
endif()
|
||||
|
||||
add_compile_definitions(GGML_USE_CPU_KLEIDIAI)
|
||||
|
||||
# Remove kleidiai target after fetching it
|
||||
if (TARGET kleidiai)
|
||||
set_target_properties(kleidiai PROPERTIES EXCLUDE_FROM_ALL TRUE)
|
||||
endif()
|
||||
|
||||
list(APPEND GGML_CPU_SOURCES
|
||||
ggml-cpu/kleidiai/kleidiai.cpp
|
||||
ggml-cpu/kleidiai/kernels.cpp
|
||||
ggml-cpu/kleidiai/kleidiai.h
|
||||
ggml-cpu/kleidiai/kernels.h
|
||||
)
|
||||
|
||||
# KleidiAI
|
||||
include_directories(
|
||||
${KLEIDIAI_SRC}/
|
||||
${KLEIDIAI_SRC}/kai/
|
||||
${KLEIDIAI_SRC}/kai/ukernels/
|
||||
${KLEIDIAI_SRC}/kai/ukernels/matmul/
|
||||
${KLEIDIAI_SRC}/kai/ukernels/matmul/matmul_clamp_f32_qsi8d32p_qsi4c32p/
|
||||
${KLEIDIAI_SRC}/kai/ukernels/matmul/pack/)
|
||||
|
||||
set(ARCH_FLAGS_TEMP "${ARCH_FLAGS}")
|
||||
if (NOT ARCH_FLAGS_TEMP)
|
||||
string(REGEX MATCH "-march=[^ ]+" ARCH_FLAGS_TEMP "${CMAKE_C_FLAGS}")
|
||||
endif()
|
||||
string(FIND "${ARCH_FLAGS_TEMP}" "+dotprod" DOTPROD_ENABLED)
|
||||
string(FIND "${ARCH_FLAGS_TEMP}" "+i8mm" I8MM_ENABLED)
|
||||
string(FIND "${ARCH_FLAGS_TEMP}" "+sme" SME_ENABLED)
|
||||
|
||||
set(PRIVATE_ARCH_FLAGS ${ARCH_FLAGS})
|
||||
|
||||
list(APPEND GGML_KLEIDIAI_SOURCES ${KLEIDIAI_SRC}/kai/ukernels/matmul/pack/kai_lhs_quant_pack_qsi8d32p_f32.c)
|
||||
list(APPEND GGML_KLEIDIAI_SOURCES ${KLEIDIAI_SRC}/kai/ukernels/matmul/pack/kai_rhs_pack_nxk_qsi4c32ps1s0scalef16_qsu4c32s16s0_neon.c)
|
||||
list(APPEND GGML_KLEIDIAI_SOURCES ${KLEIDIAI_SRC}/kai/ukernels/matmul/pack/kai_lhs_quant_pack_qsi8d32p_f32_neon.c)
|
||||
list(APPEND GGML_KLEIDIAI_SOURCES ${KLEIDIAI_SRC}/kai/ukernels/matmul/pack/kai_rhs_pack_nxk_qsi4c32pscalef16_qsu4c32s16s0.c)
|
||||
|
||||
if (NOT DOTPROD_ENABLED MATCHES -1)
|
||||
list(APPEND GGML_KLEIDIAI_SOURCES ${KLEIDIAI_SRC}/kai/ukernels/matmul/matmul_clamp_f32_qsi8d32p_qsi4c32p/kai_matmul_clamp_f32_qsi8d32p1x8_qsi4c32p4x8_1x4x32_neon_dotprod.c)
|
||||
list(APPEND GGML_KLEIDIAI_SOURCES ${KLEIDIAI_SRC}/kai/ukernels/matmul/matmul_clamp_f32_qsi8d32p_qsi4c32p/kai_matmul_clamp_f32_qsi8d32p1x4_qsi4c32p4x4_1x4_neon_dotprod.c)
|
||||
list(APPEND GGML_KLEIDIAI_SOURCES ${KLEIDIAI_SRC}/kai/ukernels/matmul/matmul_clamp_f32_qsi8d32p_qsi4c32p/kai_matmul_clamp_f32_qsi8d32p4x4_qsi4c32p4x4_16x4_neon_dotprod.c)
|
||||
endif()
|
||||
|
||||
if (NOT I8MM_ENABLED MATCHES -1)
|
||||
list(APPEND GGML_KLEIDIAI_SOURCES ${KLEIDIAI_SRC}/kai/ukernels/matmul/matmul_clamp_f32_qsi8d32p_qsi4c32p/kai_matmul_clamp_f32_qsi8d32p4x8_qsi4c32p4x8_16x4_neon_i8mm.c)
|
||||
endif()
|
||||
|
||||
if (NOT SME_ENABLED MATCHES -1)
|
||||
list(APPEND GGML_KLEIDIAI_SOURCES ${KLEIDIAI_SRC}/kai/ukernels/matmul/matmul_clamp_f32_qsi8d32p_qsi4c32p/kai_matmul_clamp_f32_qsi8d32p1vlx4_qsi4c32p4vlx4_1vlx4vl_sme2_mopa.c)
|
||||
list(APPEND GGML_KLEIDIAI_SOURCES ${KLEIDIAI_SRC}/kai/ukernels/matmul/matmul_clamp_f32_qsi8d32p_qsi4c32p/kai_matmul_clamp_f32_qsi8d32p1x4_qsi4c32p4vlx4_1x4vl_sme2_sdot.c)
|
||||
set(PRIVATE_ARCH_FLAGS "${PRIVATE_ARCH_FLAGS}+sve+sve2")
|
||||
endif()
|
||||
|
||||
set_source_files_properties(${GGML_KLEIDIAI_SOURCES} PROPERTIES COMPILE_OPTIONS "${PRIVATE_ARCH_FLAGS}")
|
||||
list(APPEND GGML_CPU_SOURCES ${GGML_KLEIDIAI_SOURCES})
|
||||
endif()
|
||||
|
||||
message(STATUS "Adding CPU backend variant ${GGML_CPU_NAME}: ${ARCH_FLAGS} ${ARCH_DEFINITIONS}")
|
||||
target_sources(${GGML_CPU_NAME} PRIVATE ${GGML_CPU_SOURCES})
|
||||
target_compile_options(${GGML_CPU_NAME} PRIVATE ${ARCH_FLAGS})
|
||||
|
|
Loading…
Add table
Add a link
Reference in a new issue