wehub/alibaba--mnn
1
0
Fork 0
Code Issues Pull Requests Actions Deployments Models Agent Notes Packages Projects Releases Wiki Activity
Files
main
alibaba--mnn/source/backend/cuda/CMakeLists.txt
T
wehub-resource-sync 5b57521aa1 chore: import upstream snapshot with attribution
2026-07-13 13:33:03 +08:00

223 lines
8.4 KiB
CMake
Raw Permalink Blame History

set(CUDA_MIN_VERSION "8.0")
find_package(CUDA ${CUDA_MIN_VERSION})
set (EXTRA_LIBS "")
if(MNN_CUDA_PROFILE)
set(EXTRA_LIBS -lnvToolsExt)
endif()
if(CUDA_FOUND)
set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -D_FORCE_INLINES -Wno-deprecated-gpu-targets -w ${EXTRA_LIBS}")
set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} --std=c++17")
if(CMAKE_BUILD_TYPE MATCHES Debug)
set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -O0 -g")
else()
set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -O3")
endif()
if (WIN32)
set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -Xcompiler /FS -Xcompiler /Zc:__cplusplus")
endif ()
include(${CMAKE_CURRENT_SOURCE_DIR}/SelectCudaComputeArch.cmake)
CUDA_SELECT_NVCC_ARCH_FLAGS(CUDA_ARCH_FLAGS ${CUDA_ARCHS})
list(LENGTH CUDA_ARCH_FLAGS_readable_code arch_count)
# Current Supported Arch List
IF (MSVC)
set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -allow-unsupported-compiler")
ENDIF()
IF (${arch_count} EQUAL 1)
set(support_archs 60 61 62 70 72 75 80 86 87 89 120)
list(FIND support_archs ${CUDA_ARCH_FLAGS_readable_code} list_index)
IF (${list_index} EQUAL -1)
message(FATAL_ERROR "Please add your own sm arch ${CUDA_ARCH_FLAGS_readable_code} to CmakeLists.txt!")
ENDIF()
ENDIF()
# When MNN_CUDA_NATIVE_ARCH is ON, only compile for the auto-detected local
# arch (skips the broad gencode list below). Saves compile time and disk.
option(MNN_CUDA_NATIVE_ARCH "Only emit gencode for local GPU arch" OFF)
IF (MNN_CUDA_NATIVE_ARCH AND ${arch_count} EQUAL 1)
set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} ${CUDA_ARCH_FLAGS}")
message(STATUS "MNN_CUDA_NATIVE_ARCH=ON, only compile for sm_${CUDA_ARCH_FLAGS_readable_code}")
ELSE()
IF ((CUDA_VERSION VERSION_GREATER "8.0") OR (CUDA_VERSION VERSION_EQUAL "8.0"))
set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -gencode arch=compute_60,code=sm_60")
set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -gencode arch=compute_61,code=sm_61")
set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -gencode arch=compute_62,code=sm_62")
ENDIF()
IF ((CUDA_VERSION VERSION_GREATER "10.1") OR (CUDA_VERSION VERSION_EQUAL "10.1"))
set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -gencode arch=compute_70,code=sm_70")
set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -gencode arch=compute_72,code=sm_72")
ENDIF()
IF ((CUDA_VERSION VERSION_GREATER "10.2") OR (CUDA_VERSION VERSION_EQUAL "10.2"))
set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -gencode arch=compute_75,code=sm_75")
add_definitions(-DMNN_CUDA_ENABLE_SM75)
ENDIF()
IF ((CUDA_VERSION VERSION_GREATER "11.2") OR (CUDA_VERSION VERSION_EQUAL "11.2"))
set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -gencode arch=compute_80,code=sm_80")
set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -gencode arch=compute_86,code=sm_86")
add_definitions(-DMNN_CUDA_ENABLE_SM80 -DMNN_CUDA_ENABLE_SM86)
ENDIF()
IF ((CUDA_VERSION VERSION_GREATER "11.4") OR (CUDA_VERSION VERSION_EQUAL "11.4"))
set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -gencode arch=compute_87,code=sm_87")
add_definitions(-DMNN_CUDA_ENABLE_SM87)
ENDIF()
IF ((CUDA_VERSION VERSION_GREATER "12.2") OR (CUDA_VERSION VERSION_EQUAL "12.2"))
set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -gencode arch=compute_89,code=sm_89")
add_definitions(-DMNN_CUDA_ENABLE_SM89 -DMNN_CUDA_ENABLE_SM89)
ENDIF()
IF (CUDA_VERSION VERSION_GREATER_EQUAL "12.8")
# Blackwell consumer (e.g. RTX 5090) — sm_120
set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -gencode arch=compute_120,code=sm_120")
ENDIF()
ENDIF()
# Limit minimum cuda version for each archs
IF (${arch_count} EQUAL 1)
IF (MNN_SUPPORT_TRANSFORMER_FUSE OR (CUDA_ARCH_FLAGS_readable_code VERSION_GREATER "80") OR (CUDA_ARCH_FLAGS_readable_code VERSION_EQUAL "80"))
IF (CUDA_VERSION VERSION_LESS "11.2")
message(FATAL_ERROR "Please update cuda version to 11.2 or higher!")
ENDIF()
ENDIF()
IF ((CUDA_ARCH_FLAGS_readable_code VERSION_GREATER "75") OR (CUDA_ARCH_FLAGS_readable_code VERSION_EQUAL "75"))
IF (CUDA_VERSION VERSION_LESS "10.2")
message(FATAL_ERROR "Please update cuda version to 10.2 or higher!")
ENDIF()
ENDIF()
IF ((CUDA_ARCH_FLAGS_readable_code VERSION_GREATER "70") OR (CUDA_ARCH_FLAGS_readable_code VERSION_EQUAL "70"))
IF (CUDA_VERSION VERSION_LESS "10.1")
message(FATAL_ERROR "Please update cuda version to 10.1 or higher!")
ENDIF()
ENDIF()
ENDIF()
message(STATUS "Enabling CUDA support (version: ${CUDA_VERSION_STRING},"
" archs: ${CUDA_ARCH_FLAGS_readable})")
else()
message(FATAL_ERROR "CUDA not found >= ${CUDA_MIN_VERSION} required)")
endif()
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fexceptions")
option(MNN_CUDA_QUANT "Enable MNN CUDA Quant File" OFF)
option(MNN_CUDA_BF16 "Enable MNN CUDA Bfloat16 File" OFF)
option(MNN_CUDA_TUNE_PARAM "Enable MNN CUDA Tuning Cutlass Params" OFF)
IF (MNN_CUDA_QUANT)
add_definitions(-DENABLE_CUDA_QUANT)
ENDIF()
IF (MNN_CUDA_BF16)
add_definitions(-DENABLE_CUDA_BF16)
ENDIF()
IF (MNN_CUDA_NATIVE_ARCH AND CUDA_ARCH_FLAGS_readable_code)
add_definitions(-DMNN_CUDA_ENABLE_SM${CUDA_ARCH_FLAGS_readable_code})
ENDIF()
IF (MNN_CUDA_TUNE_PARAM)
add_definitions(-DENABLE_CUDA_TUNE_PARAM)
ENDIF()
IF (MNN_LOW_MEMORY)
add_definitions(-DMNN_LOW_MEMORY)
ENDIF()
include(FetchContent)
if(MNN_SUPPORT_TRANSFORMER_FUSE)
set(CUTLASS_COMMIT_HASH "b995f933179c22d3fe0d871c3a53d11e4681950f")
set(CUTLASS_VERSION_NAME "v4.0.0")
else()
set(CUTLASS_COMMIT_HASH "319a389f42b776fae5701afcb943fc03be5b5c25")
set(CUTLASS_VERSION_NAME "v2.9.0")
endif()
set(CUTLASS_SOURCE_DIR "${CMAKE_SOURCE_DIR}/3rd_party/cutlass/${CUTLASS_VERSION_NAME}")
FetchContent_Declare(
cutlass
GIT_REPOSITORY https://github.com/NVIDIA/cutlass.git
GIT_TAG ${CUTLASS_COMMIT_HASH}
SOURCE_DIR ${CUTLASS_SOURCE_DIR}
)
FetchContent_GetProperties(cutlass)
if(NOT cutlass_POPULATED)
FetchContent_Populate(cutlass)
message(STATUS "Cloned cutlass commit ${CUTLASS_COMMIT_HASH} to ${CUTLASS_SOURCE_DIR}")
endif()
set(CUTLASS_SRC ${CUTLASS_SOURCE_DIR})
file(GLOB_RECURSE MNN_CUDA_SRC ${CMAKE_CURRENT_LIST_DIR}/core/* ${CMAKE_CURRENT_SOURCE_DIR}/execution/*)
if(NOT MNN_SUPPORT_TRANSFORMER_FUSE)
file(GLOB_RECURSE MNN_CUDA_TRANSFORMER_FUSE_SRC ${CMAKE_CURRENT_SOURCE_DIR}/execution/plugin/*)
list(REMOVE_ITEM MNN_CUDA_SRC ${MNN_CUDA_TRANSFORMER_FUSE_SRC})
endif()
if(NOT MNN_SUPPORT_RENDER)
file(GLOB_RECURSE MNN_CUDA_RENDER_SRC ${CMAKE_CURRENT_SOURCE_DIR}/execution/render/*)
if(MNN_CUDA_RENDER_SRC AND MNN_CUDA_SRC)
list(REMOVE_ITEM MNN_CUDA_SRC ${MNN_CUDA_RENDER_SRC})
endif()
endif()
if(NOT MNN_LOW_MEMORY)
file(GLOB_RECURSE MNN_CUDA_LOW_MEM_SRC ${CMAKE_CURRENT_SOURCE_DIR}/execution/weight_only_quant/*)
list(REMOVE_ITEM MNN_CUDA_SRC ${MNN_CUDA_LOW_MEM_SRC})
endif()
message(STATUS "message ${CUDA_NVCC_FLAGS} !!!!!!!!!!! ${CUDA_INCLUDE_DIRS}")
if(WIN32)
cuda_add_library(MNN_CUDA STATIC Register.cpp ${MNN_CUDA_SRC})
find_library(CUDA_CUBLAS_LIBRARY cublas HINTS ${CUDA_TOOLKIT_ROOT_DIR}/lib/x64 ${CUDA_TOOLKIT_ROOT_DIR}/lib)
if(CUDA_CUBLAS_LIBRARY)
set(MNN_CUDA_LIBS MNN_CUDA ${CUDA_LIBRARIES} ${CUDA_CUBLAS_LIBRARY} PARENT_SCOPE)
else()
set(MNN_CUDA_LIBS MNN_CUDA ${CUDA_LIBRARIES} PARENT_SCOPE)
endif()
else()
cuda_add_library(MNN_Cuda_Main SHARED ${MNN_CUDA_SRC})
# Link cuBLAS for prefill GEMM acceleration
find_library(CUDA_CUBLAS_LIBRARY cublas HINTS ${CUDA_TOOLKIT_ROOT_DIR}/lib64 ${CUDA_TOOLKIT_ROOT_DIR}/lib)
if(CUDA_CUBLAS_LIBRARY)
target_link_libraries(MNN_Cuda_Main ${CUDA_CUBLAS_LIBRARY})
else()
target_link_libraries(MNN_Cuda_Main -lcublas)
endif()
if(MNN_CUDA_PROFILE)
target_compile_options(MNN_Cuda_Main PRIVATE -DMNN_CUDA_PROFILE)
target_link_libraries(MNN_Cuda_Main ${CUDA_INCLUDE_DIRS}/../lib/libnvToolsExt.so)
endif()
if(MNN_CODEGEN_CUDA)
target_link_libraries(MNN_Cuda_Main ${CUDA_INCLUDE_DIRS}/../lib/libnvrtc.so ${CUDA_INCLUDE_DIRS}/../lib/stubs/libcuda.so)
endif()
set(MNN_CUDA_LIBS MNN_Cuda_Main PARENT_SCOPE)
add_library(MNN_CUDA OBJECT Register.cpp)
endif()
include_directories(
${CMAKE_CURRENT_LIST_DIR}/
${CUDA_INCLUDE_DIRS}
${CMAKE_SOURCE_DIR}/include/
${CUTLASS_SRC}/include
)
Reference in New Issue View Git Blame Copy Permalink