youngkingdom/vllm
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[cpu] Dispatch un-quantized linear to oneDNN/ACL by default for AArch64 (#27183)

Browse Source 
Signed-off-by: Fadi Arafeh <fadi.arafeh@arm.com>
Co-authored-by: Michael Yang <Michael.Yang@arm.com>
This commit is contained in:
Fadi Arafeh
2025-10-21 03:02:58 +01:00
committed by GitHub
parent a03cf9bc70
commit 163965d183
3 changed files with 42 additions and 23 deletions
Download Patch File Download Diff File Expand all files Collapse all files

57
cmake/cpu_extension.cmake
View File

@ -188,16 +188,47 @@ else()
message(FATAL_ERROR "vLLM CPU backend requires AVX512, AVX2, Power9+ ISA, S390X ISA, ARMv8 or RISC-V support.")
endif()
#
# Build oneDNN for W8A8 GEMM kernels (only for x86-AVX512 /ARM platforms)
# Flag to enable ACL kernels for AARCH64 platforms
if (VLLM_BUILD_ACL STREQUAL "ON")
set(USE_ACL ON)
else()
set(USE_ACL OFF)
endif()
# Build oneDNN for GEMM kernels (only for x86-AVX512 /ARM platforms)
if ((AVX512_FOUND AND NOT AVX512_DISABLED) OR (ASIMD_FOUND AND NOT APPLE_SILICON_FOUND) OR POWER9_FOUND OR POWER10_FOUND OR POWER11_FOUND)
# Fetch and build Arm Compute Library (ACL) as oneDNN's backend for AArch64
# TODO [fadara01]: remove this once ACL can be fetched and built automatically as a dependency of oneDNN
if(ASIMD_FOUND)
if(DEFINED ENV{ACL_ROOT_DIR} AND IS_DIRECTORY "$ENV{ACL_ROOT_DIR}")
message(STATUS "Using ACL from specified source directory: $ENV{ACL_ROOT_DIR}")
else()
message(STATUS "Downloading Arm Compute Library (ACL) from GitHub")
FetchContent_Populate(arm_compute
SUBBUILD_DIR "${FETCHCONTENT_BASE_DIR}/arm_compute-subbuild"
SOURCE_DIR "${FETCHCONTENT_BASE_DIR}/arm_compute-src"
GIT_REPOSITORY https://github.com/ARM-software/ComputeLibrary.git
GIT_TAG v52.2.0
GIT_SHALLOW TRUE
GIT_PROGRESS TRUE
)
set(ENV{ACL_ROOT_DIR} "${arm_compute_SOURCE_DIR}")
endif()
# Build ACL with scons
include(ProcessorCount)
ProcessorCount(_NPROC)
execute_process(
COMMAND scons -j${_NPROC}
Werror=0 debug=0 neon=1 examples=0 embed_kernels=0 os=linux
arch=armv8.2-a build=native benchmark_examples=0 fixed_format_kernels=1
multi_isa=1 openmp=1 cppthreads=0
WORKING_DIRECTORY "$ENV{ACL_ROOT_DIR}"
RESULT_VARIABLE _acl_rc
)
if(NOT _acl_rc EQUAL 0)
message(FATAL_ERROR "ACL SCons build failed (exit ${_acl_rc}).")
endif()
set(ONEDNN_AARCH64_USE_ACL "ON")
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wl,-rpath,$ENV{ACL_ROOT_DIR}/build/")
add_compile_definitions(VLLM_USE_ACL)
endif()
set(FETCHCONTENT_SOURCE_DIR_ONEDNN "$ENV{FETCHCONTENT_SOURCE_DIR_ONEDNN}" CACHE PATH "Path to a local oneDNN source directory.")
if(FETCHCONTENT_SOURCE_DIR_ONEDNN)
@ -217,16 +248,6 @@ if ((AVX512_FOUND AND NOT AVX512_DISABLED) OR (ASIMD_FOUND AND NOT APPLE_SILICON
)
endif()
if(USE_ACL)
find_library(ARM_COMPUTE_LIBRARY NAMES arm_compute PATHS $ENV{ACL_ROOT_DIR}/build/)
if(NOT ARM_COMPUTE_LIBRARY)
message(FATAL_ERROR "Could not find ARM Compute Library: please set ACL_ROOT_DIR")
endif()
set(ONEDNN_AARCH64_USE_ACL "ON")
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wl,-rpath,$ENV{ACL_ROOT_DIR}/build/")
add_compile_definitions(VLLM_USE_ACL)
endif()
set(ONEDNN_LIBRARY_TYPE "STATIC")
set(ONEDNN_BUILD_DOC "OFF")
set(ONEDNN_BUILD_EXAMPLES "OFF")