mirror of
https://github.com/vllm-project/vllm-ascend.git
synced 2025-10-20 05:33:51 +08:00
bcc313e8f2
### What this PR does / why we need it? - Adds the `mla_preprocess` custom kernel to provide an optimized pre-processing operator for Multi-head Latent Attention (MLA) on Ascend NPUs. - Wires the new kernel into the C++ extension pipeline so vLLM can invoke it directly, cutting Python-side tensor shuffling and memory copies that previously bottlenecked MLA compilation paths. ### Does this PR introduce any user-facing change? - No. The change only introduces a low-level kernel; public APIs and inference behavior remain unchanged. ### How was this patch tested? - Dedicated Ascend kernels are not covered by our CI yet, so no extra automated tests were added. Future MLA-focused regression runs will cover this path. - vLLM version: v0.11.0 Signed-off-by: Chen Chen <0109chenchen@gmail.com>
105 lines
3.0 KiB
CMake
105 lines
3.0 KiB
CMake
cmake_minimum_required(VERSION 3.16)
|
|
project(vllm_ascend_C)
|
|
|
|
# include(CheckCXXcompilerFlag)
|
|
# check_cxx_compiler_flag("-std=c++17", COMPILER_SUPPORTS_CXX17)
|
|
set(CMAKE_CXX_STANDARD 17)
|
|
|
|
include(${CMAKE_CURRENT_LIST_DIR}/cmake/utils.cmake)
|
|
|
|
# Suppress potential warnings about unused manually-specified variables
|
|
set(ignoreMe "${VLLM_PYTHON_PATH}")
|
|
|
|
# TODO: Add 3.12 back when torch-npu support 3.12
|
|
set(PYTHON_SUPPORTED_VERSIONS "3.9" "3.10" "3.11")
|
|
|
|
find_package(pybind11 REQUIRED)
|
|
|
|
append_cmake_prefix_path("torch" "torch.utils.cmake_prefix_path")
|
|
set(VLLM_ASCEND_INSTALL_PATH "${CMAKE_INSTALL_PREFIX}")
|
|
|
|
find_package(Torch REQUIRED)
|
|
|
|
set(RUN_MODE "npu" CACHE STRING "cpu/sim/npu")
|
|
set(SOC_VERSION ${SOC_VERSION})
|
|
message(STATUS "Detected SOC version: ${SOC_VERSION}")
|
|
|
|
if (NOT CMAKE_BUILD_TYPE)
|
|
set(CMAKE_BUILD_TYPE "Release" CACHE STRINGS "Build type Release/Debug (default Release)" FORCE)
|
|
endif()
|
|
|
|
if (CMAKE_INSTALL_PREFIX STREQUAL /usr/local)
|
|
set(CMAKE_INSTALL_PREFIX "${CMAKE_CURRENT_LIST_DIR}/out" CACHE STRINGS "path to install()")
|
|
endif()
|
|
|
|
set(ASCEND_CANN_PACKAGE_PATH ${ASCEND_HOME_PATH})
|
|
if(EXISTS ${ASCEND_HOME_PATH}/tools/tikcpp/ascendc_kernel_cmake)
|
|
set(ASCENDC_CMAKE_DIR ${ASCEND_HOME_PATH}/tools/tikcpp/ascendc_kernel_cmake)
|
|
elseif(EXISTS ${ASCEND_HOME_PATH}/compiler/tikcpp/ascendc_kernel_cmake)
|
|
set(ASCENDC_CMAKE_DIR ${ASCEND_HOME_PATH}/compiler/tikcpp/ascendc_kernel_cmake)
|
|
elseif(EXISTS ${ASCEND_HOME_PATH}/ascendc_devkit/tikcpp/samples/cmake)
|
|
set(ASCENDC_CMAKE_DIR ${ASCEND_HOME_PATH}/ascendc_devkit/tikcpp/samples/cmake)
|
|
else()
|
|
message(FATAL_ERROR "ascendc_kernel_cmake does not exist, please check whether the cann package is installed.")
|
|
endif()
|
|
|
|
include(${ASCENDC_CMAKE_DIR}/ascendc.cmake)
|
|
|
|
file(GLOB KERNEL_FILES
|
|
${CMAKE_CURRENT_SOURCE_DIR}/csrc/kernels/*.cpp)
|
|
|
|
ascendc_library(vllm_ascend_kernels SHARED
|
|
${KERNEL_FILES}
|
|
${CMAKE_CURRENT_SOURCE_DIR}/csrc/mla_preprocess/op_kernel/mla_preprocess_kernel.cpp
|
|
)
|
|
|
|
message("TORCH_NPU_PATH is ${TORCH_NPU_PATH}")
|
|
|
|
file(GLOB VLLM_ASCEND_SRC
|
|
${CMAKE_CURRENT_SOURCE_DIR}/csrc/*.cpp)
|
|
|
|
include_directories(
|
|
${pybind11_INCLUDE_DIRS}
|
|
${PYTHON_INCLUDE_PATH}
|
|
${TORCH_INCLUDE_DIRS}
|
|
${TORCH_NPU_PATH}/include
|
|
${ASCEND_HOME_PATH}/include
|
|
${ASCEND_HOME_PATH}/aarch64-linux/include/experiment/platform
|
|
${ASCEND_HOME_PATH}/x86_64-linux/include/experiment/platform
|
|
)
|
|
|
|
set(
|
|
INCLUDES
|
|
${TORCH_INCLUDE_DIRS}
|
|
${TORCH_NPU_INCLUDE_DIRS}
|
|
${ASCEND_HOME_PATH}/include
|
|
${ASCEND_HOME_PATH}/aarch64-linux/include/experiment/platform
|
|
)
|
|
|
|
pybind11_add_module(vllm_ascend_C ${VLLM_ASCEND_SRC})
|
|
|
|
target_link_directories(
|
|
vllm_ascend_C
|
|
PRIVATE
|
|
${TORCH_NPU_PATH}/lib/
|
|
${ASCEND_HOME_PATH}/lib64
|
|
)
|
|
|
|
target_link_libraries(
|
|
vllm_ascend_C
|
|
PUBLIC
|
|
${TORCH_LIBRARIES}
|
|
libtorch_npu.so
|
|
vllm_ascend_kernels
|
|
ascendcl
|
|
tiling_api
|
|
register
|
|
platform
|
|
ascendalog
|
|
dl
|
|
)
|
|
|
|
target_link_options(vllm_ascend_C PRIVATE "-Wl,-rpath,$ORIGIN:$ORIGIN/lib")
|
|
|
|
install(TARGETS vllm_ascend_C vllm_ascend_kernels DESTINATION ${VLLM_ASCEND_INSTALL_PATH})
|