Files
ROCm/CMakeLists.txt
goostavz f1512bded1 Initial code merge of Hopper support (#2036)
The initial code merge of Nvidia Hopper features support. Please be
aware that the code merge is not finished yet and the trouble-shooting
is still ongoing. The new hardware features (GMMA, TMA, STMATRIX etc.)
and automatic warp-specialization are experimental for now and turned
off by default. It is recommended for a trial when version 3.0 is
released.

The work is contributed by:
ben-zhang-609, bealwang, donproc, qliu93, jsh20, allatit23, LyricZhao,
ivanyinwz, goostavz & yangjunpro
from Nvidia, in cooperation with:
ptillet, Jokeren, ThomasRaoux & zahimoud
from OpenAI.

Co-authored-by: Goostav Zhu <gzhu@nvidia.com>
2023-08-07 09:53:04 +08:00

279 lines
7.2 KiB
CMake

cmake_minimum_required(VERSION 3.18)
if(POLICY CMP0116)
# Introduced in cmake 3.20
# https://cmake.org/cmake/help/latest/policy/CMP0116.html
cmake_policy(SET CMP0116 OLD)
endif()
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 "${CMAKE_CURRENT_SOURCE_DIR}/cmake")
endif()
# Options
option(TRITON_BUILD_TUTORIALS "Build C++ Triton tutorials" ON)
option(TRITON_BUILD_PYTHON_MODULE "Build Python Triton bindings" OFF)
set(TRITON_CODEGEN_BACKENDS "" CACHE STRING "Enable different codegen backends")
# Ensure Python3 vars are set correctly
# used conditionally in this file and by lit tests
# Customized release build type with assertions: TritonRelBuildWithAsserts
set(CMAKE_C_FLAGS_TRITONRELBUILDWITHASSERTS "-O2 -g")
set(CMAKE_CXX_FLAGS_TRITONRELBUILDWITHASSERTS "-O2 -g")
# Default build type
if(NOT CMAKE_BUILD_TYPE)
message(STATUS "Default build type: Release")
set(CMAKE_BUILD_TYPE "Release")
endif()
if(NOT WIN32)
find_library(TERMINFO_LIBRARY tinfo)
endif()
# Compiler flags
include_directories(${CMAKE_CURRENT_SOURCE_DIR}/include)
# Third-party
include_directories(${PYBIND11_INCLUDE_DIR})
set(CMAKE_CXX_FLAGS "${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}")
# FindLLVM outputs LLVM_LIBRARY_DIRS but we expect LLVM_LIBRARY_DIR here
set(LLVM_LIBRARY_DIR ${LLVM_LIBRARY_DIRS})
if(APPLE)
set(CMAKE_OSX_DEPLOYMENT_TARGET "10.14")
endif()
# sometimes we don't want to use llvm-config, since it may have been downloaded for some specific linux distros
else()
set(LLVM_LDFLAGS "-L${LLVM_LIBRARY_DIR}")
set(LLVM_LIBRARIES
LLVMNVPTXCodeGen
LLVMNVPTXDesc
LLVMNVPTXInfo
LLVMAMDGPUDisassembler
LLVMMCDisassembler
LLVMAMDGPUCodeGen
LLVMMIRParser
LLVMGlobalISel
LLVMSelectionDAG
LLVMipo
LLVMInstrumentation
LLVMVectorize
LLVMLinker
LLVMIRReader
LLVMAsmParser
LLVMFrontendOpenMP
LLVMAsmPrinter
LLVMDebugInfoDWARF
LLVMCodeGen
LLVMTarget
LLVMScalarOpts
LLVMInstCombine
LLVMAggressiveInstCombine
LLVMTransformUtils
LLVMBitWriter
LLVMAnalysis
LLVMProfileData
LLVMObject
LLVMTextAPI
LLVMBitReader
LLVMAMDGPUAsmParser
LLVMMCParser
LLVMAMDGPUDesc
LLVMAMDGPUUtils
LLVMMC
LLVMDebugInfoCodeView
LLVMDebugInfoMSF
LLVMCore
LLVMRemarks
LLVMBitstreamReader
LLVMBinaryFormat
LLVMAMDGPUInfo
LLVMSupport
LLVMDemangle
LLVMPasses
LLVMAnalysis
LLVMTransformUtils
LLVMScalarOpts
LLVMTransformUtils
LLVMipo
LLVMObjCARCOpts
LLVMCoroutines
LLVMAnalysis
)
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()
find_package(Python3 REQUIRED COMPONENTS Development Interpreter)
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(${LLVM_LIBRARY_DIR})
add_subdirectory(include)
add_subdirectory(lib)
add_subdirectory(bin)
# find_package(PythonLibs REQUIRED)
set(TRITON_SOURCE_DIR "${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)
if(TRITON_BUILD_PYTHON_MODULE)
add_library(triton SHARED ${PYTHON_SRC})
set(TRITON_LIBRARIES
TritonAnalysis
TritonTransforms
TritonGPUTransforms
TritonNvidiaGPUTransforms
TritonLLVMIR
TritonPTX
TritonHSACO
${dialect_libs}
${conversion_libs}
# optimizations
MLIRBytecodeWriter
MLIRPass
MLIRTransforms
MLIRLLVMDialect
MLIRSupport
MLIRTargetLLVMIRExport
MLIRExecutionEngine
MLIRMathToLLVM
MLIRNVVMToLLVMIRTranslation
MLIRROCDLToLLVMIRTranslation
MLIRIR
)
if(WIN32)
target_link_libraries(triton PRIVATE ${LLVM_LIBRARIES} ${CMAKE_DL_LIBS}
${TRITON_LIBRARIES}
)
elseif(APPLE)
target_link_libraries(triton ${LLVM_LIBRARIES} z
${TRITON_LIBRARIES}
)
else()
target_link_libraries(triton ${LLVM_LIBRARIES} z
${TRITON_LIBRARIES}
)
# TODO: Figure out which target is sufficient to fix errors; triton is
# apparently not enough
link_libraries(stdc++fs)
endif()
target_link_options(triton PRIVATE ${LLVM_LDFLAGS})
endif()
if(UNIX AND NOT APPLE)
set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} -Wl,--exclude-libs,ALL")
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} ${PYTHON_LDFLAGS})
endif()
list(LENGTH TRITON_CODEGEN_BACKENDS CODEGEN_BACKENDS_LEN)
if (${CODEGEN_BACKENDS_LEN} GREATER 0)
set(PYTHON_THIRD_PARTY_PATH ${CMAKE_CURRENT_SOURCE_DIR}/python/triton/third_party)
foreach(CODEGEN_BACKEND ${TRITON_CODEGEN_BACKENDS})
add_subdirectory(third_party/${CODEGEN_BACKEND})
endforeach()
endif()
add_subdirectory(test)
add_subdirectory(unittest)