This PR merges the `triton-mlir` branch, in which we have been quietly rewriting the Triton backend from scratch to increase maintainability, stability and ultimately performance. Changes to the runtime are minimal, and this new version aims to remain backward-compatible with the previous commit. The legacy backend is now officially deprecated, but can still be accessed via the `legacy-backend` tag. Co-authored-by: Keren Zhou <kerenzhou@openai.com> Co-authored-by: Yan Chunwei <yanchunwei@outlook.com> Co-authored-by: goostavz <109190422+goostavz@users.noreply.github.com> Co-authored-by: Shintaro Iwasaki <siwasaki@fb.com> Co-authored-by: Yan Da <dyanab@connect.ust.hk> Co-authored-by: Jun Yang <yangjunpro@gmail.com> Co-authored-by: Ian Bearman <ianb@microsoft.com> Co-authored-by: Jason Ansel <jansel@jansel.net> Co-authored-by: Qingyi Liu <qingyil@nvidia.com> Co-authored-by: ben-zhang-609 <110140741+ben-zhang-609@users.noreply.github.com> Co-authored-by: Chenggang Zhao <lyricz@yeah.net> Co-authored-by: ben-zhang-609 <benzh609@gmail.com> Co-authored-by: dongdongl <dongdongl@nvidia.com>
242 lines
6.5 KiB
CMake
242 lines
6.5 KiB
CMake
cmake_minimum_required(VERSION 3.6)
|
|
include(ExternalProject)
|
|
|
|
set(CMAKE_CXX_STANDARD 17)
|
|
|
|
set(CMAKE_INCLUDE_CURRENT_DIR ON)
|
|
|
|
project(triton)
|
|
include(CTest)
|
|
if(NOT WIN32)
|
|
list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/cmake")
|
|
endif()
|
|
|
|
# Options
|
|
option(TRITON_BUILD_TUTORIALS "Build C++ Triton tutorials" ON)
|
|
option(TRITON_BUILD_PYTHON_MODULE "Build Python Triton bindings" OFF)
|
|
|
|
# Ensure Python3 vars are set correctly
|
|
# used conditionally in this file and by lit tests
|
|
find_package(Python3 REQUIRED COMPONENTS Development Interpreter)
|
|
|
|
# Customized release build type with assertions: TritonRelBuildWithAsserts
|
|
set(CMAKE_C_FLAGS_TRITONRELBUILDWITHASSERTS "-O2 -g")
|
|
set(CMAKE_CXX_FLAGS_TRITONRELBUILDWITHASSERTS "-O2 -g")
|
|
|
|
# Default build type
|
|
if(NOT CMAKE_BUILD_TYPE)
|
|
message(STATUS "Default build type: Release")
|
|
set(CMAKE_BUILD_TYPE "Release")
|
|
endif()
|
|
|
|
if(NOT WIN32)
|
|
find_library(TERMINFO_LIBRARY tinfo)
|
|
endif()
|
|
|
|
# Compiler flags
|
|
include_directories(${CMAKE_CURRENT_SOURCE_DIR}/include)
|
|
|
|
# Third-party
|
|
include_directories(${PYBIND11_INCLUDE_DIR})
|
|
|
|
if(WIN32)
|
|
SET(BUILD_SHARED_LIBS OFF)
|
|
include_directories(${CMAKE_CURRENT_SOURCE_DIR}/deps/dlfcn-win32/src)
|
|
add_subdirectory(deps/dlfcn-win32/src ${CMAKE_BINARY_DIR}/dlfcn-win32)
|
|
endif()
|
|
|
|
set(CMAKE_CXX_FLAGS "${CMAKE_C_FLAGS} -D__STDC_FORMAT_MACROS -fPIC -std=gnu++17 -fvisibility=hidden -fvisibility-inlines-hidden")
|
|
if(APPLE)
|
|
set(CMAKE_OSX_DEPLOYMENT_TARGET 11.6)
|
|
endif()
|
|
|
|
|
|
|
|
##########
|
|
# LLVM
|
|
##########
|
|
if (NOT MLIR_DIR)
|
|
if(NOT LLVM_LIBRARY_DIR)
|
|
if(WIN32)
|
|
find_package(LLVM 13 REQUIRED COMPONENTS nvptx amdgpu)
|
|
|
|
include_directories(${LLVM_INCLUDE_DIRS})
|
|
separate_arguments(LLVM_DEFINITIONS_LIST NATIVE_COMMAND ${LLVM_DEFINITIONS})
|
|
add_definitions(${LLVM_DEFINITIONS_LIST})
|
|
|
|
llvm_map_components_to_libnames(LLVM_LIBRARIES support core
|
|
NVPTXInfo nvptxcodegen
|
|
AMDGPUInfo AMDGPUcodegen
|
|
)
|
|
else()
|
|
find_package(LLVM 11 REQUIRED COMPONENTS "nvptx;amdgpu")
|
|
endif()
|
|
message(STATUS "Found LLVM ${LLVM_PACKAGE_VERSION}")
|
|
if(APPLE)
|
|
set(CMAKE_OSX_DEPLOYMENT_TARGET "10.14")
|
|
endif()
|
|
# sometimes we don't want to use llvm-config, since it may have been downloaded for some specific linux distros
|
|
else()
|
|
set(LLVM_LDFLAGS "-L${LLVM_LIBRARY_DIR}")
|
|
set(LLVM_LIBRARIES
|
|
libLLVMNVPTXCodeGen.a
|
|
libLLVMNVPTXDesc.a
|
|
libLLVMNVPTXInfo.a
|
|
libLLVMAMDGPUDisassembler.a
|
|
libLLVMMCDisassembler.a
|
|
libLLVMAMDGPUCodeGen.a
|
|
libLLVMMIRParser.a
|
|
libLLVMGlobalISel.a
|
|
libLLVMSelectionDAG.a
|
|
libLLVMipo.a
|
|
libLLVMInstrumentation.a
|
|
libLLVMVectorize.a
|
|
libLLVMLinker.a
|
|
libLLVMIRReader.a
|
|
libLLVMAsmParser.a
|
|
libLLVMFrontendOpenMP.a
|
|
libLLVMAsmPrinter.a
|
|
libLLVMDebugInfoDWARF.a
|
|
libLLVMCodeGen.a
|
|
libLLVMTarget.a
|
|
libLLVMScalarOpts.a
|
|
libLLVMInstCombine.a
|
|
libLLVMAggressiveInstCombine.a
|
|
libLLVMTransformUtils.a
|
|
libLLVMBitWriter.a
|
|
libLLVMAnalysis.a
|
|
libLLVMProfileData.a
|
|
libLLVMObject.a
|
|
libLLVMTextAPI.a
|
|
libLLVMBitReader.a
|
|
libLLVMAMDGPUAsmParser.a
|
|
libLLVMMCParser.a
|
|
libLLVMAMDGPUDesc.a
|
|
libLLVMAMDGPUUtils.a
|
|
libLLVMMC.a
|
|
libLLVMDebugInfoCodeView.a
|
|
libLLVMDebugInfoMSF.a
|
|
libLLVMCore.a
|
|
libLLVMRemarks.a
|
|
libLLVMBitstreamReader.a
|
|
libLLVMBinaryFormat.a
|
|
libLLVMAMDGPUInfo.a
|
|
libLLVMSupport.a
|
|
libLLVMDemangle.a
|
|
libLLVMPasses.a
|
|
libLLVMAnalysis.a
|
|
libLLVMTransformUtils.a
|
|
libLLVMScalarOpts.a
|
|
libLLVMTransformUtils.a
|
|
libLLVMipo.a
|
|
libLLVMObjCARCOpts.a
|
|
libLLVMCoroutines.a
|
|
libLLVMAnalysis.a
|
|
)
|
|
endif()
|
|
set (MLIR_DIR ${LLVM_LIBRARY_DIR}/cmake/mlir)
|
|
endif()
|
|
|
|
# Python module
|
|
if(TRITON_BUILD_PYTHON_MODULE)
|
|
message(STATUS "Adding Python module")
|
|
set(PYTHON_SRC_PATH ${CMAKE_CURRENT_SOURCE_DIR}/python/src)
|
|
set(PYTHON_SRC ${PYTHON_SRC_PATH}/main.cc ${PYTHON_SRC_PATH}/triton.cc)
|
|
include_directories("." ${PYTHON_SRC_PATH})
|
|
if (PYTHON_INCLUDE_DIRS)
|
|
include_directories(${PYTHON_INCLUDE_DIRS})
|
|
else()
|
|
include_directories(${Python3_INCLUDE_DIRS})
|
|
link_directories(${Python3_LIBRARY_DIRS})
|
|
link_libraries(${Python3_LIBRARIES})
|
|
add_link_options(${Python3_LINK_OPTIONS})
|
|
endif()
|
|
endif()
|
|
|
|
|
|
# # Triton
|
|
# file(GLOB_RECURSE LIBTRITON_SRC lib/*.cc)
|
|
# if (WIN32 AND TRITON_BUILD_PYTHON_MODULE)
|
|
# Python3_add_library(triton SHARED ${LIBTRITON_SRC} ${PYTHON_SRC})
|
|
# set_target_properties(triton PROPERTIES SUFFIX ".pyd")
|
|
# set_target_properties(triton PROPERTIES PREFIX "lib")
|
|
# else()
|
|
# add_library(triton SHARED ${LIBTRITON_SRC} ${PYTHON_SRC})
|
|
# endif()
|
|
|
|
|
|
# MLIR
|
|
find_package(MLIR REQUIRED CONFIG PATHS ${MLIR_DIR})
|
|
|
|
list(APPEND CMAKE_MODULE_PATH "${MLIR_CMAKE_DIR}")
|
|
list(APPEND CMAKE_MODULE_PATH "${LLVM_CMAKE_DIR}")
|
|
|
|
include(TableGen) # required by AddMLIR
|
|
include(AddLLVM)
|
|
include(AddMLIR)
|
|
|
|
# Disable warnings that show up in external code (gtest;pybind11)
|
|
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Werror -Wno-covered-switch-default")
|
|
|
|
include_directories(${MLIR_INCLUDE_DIRS})
|
|
include_directories(${LLVM_INCLUDE_DIRS})
|
|
include_directories(${PROJECT_SOURCE_DIR}/include)
|
|
include_directories(${PROJECT_BINARY_DIR}/include) # Tablegen'd files
|
|
# link_directories(${LLVM_LIBRARY_DIR})
|
|
|
|
add_subdirectory(include)
|
|
add_subdirectory(lib)
|
|
add_subdirectory(bin)
|
|
|
|
add_library(triton SHARED ${PYTHON_SRC})
|
|
|
|
# find_package(PythonLibs REQUIRED)
|
|
|
|
set(TRITON_SOURCE_DIR "${CMAKE_CURRENT_SOURCE_DIR}")
|
|
set(TRITON_BINARY_DIR "${CMAKE_CURRENT_BINARY_DIR}")
|
|
|
|
get_property(dialect_libs GLOBAL PROPERTY MLIR_DIALECT_LIBS)
|
|
get_property(conversion_libs GLOBAL PROPERTY MLIR_CONVERSION_LIBS)
|
|
|
|
target_link_libraries(triton
|
|
TritonAnalysis
|
|
TritonTransforms
|
|
TritonGPUTransforms
|
|
TritonLLVMIR
|
|
TritonPTX
|
|
${dialect_libs}
|
|
${conversion_libs}
|
|
# optimizations
|
|
MLIRPass
|
|
MLIRTransforms
|
|
MLIRLLVMIR
|
|
MLIRSupport
|
|
MLIRTargetLLVMIRExport
|
|
MLIRExecutionEngine
|
|
MLIRMathToLLVM
|
|
MLIRNVVMToLLVMIRTranslation
|
|
MLIRIR
|
|
)
|
|
|
|
target_link_options(triton PRIVATE ${LLVM_LDFLAGS})
|
|
|
|
if(WIN32)
|
|
target_link_libraries(triton PRIVATE ${LLVM_LIBRARIES} dl) # dl is from dlfcn-win32
|
|
else()
|
|
target_link_libraries(triton ${LLVM_LIBRARIES} z)
|
|
endif()
|
|
|
|
|
|
if(TRITON_BUILD_PYTHON_MODULE AND NOT WIN32)
|
|
set(CMAKE_SHARED_LIBRARY_SUFFIX ".so")
|
|
# Check if the platform is MacOS
|
|
if(APPLE)
|
|
set(PYTHON_LDFLAGS "-undefined dynamic_lookup -flto")
|
|
endif()
|
|
target_link_libraries(triton ${CUTLASS_LIBRARIES} ${PYTHON_LDFLAGS})
|
|
endif()
|
|
|
|
add_subdirectory(test)
|
|
|
|
add_subdirectory(unittest)
|