gaspersic/triton
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
Files
8eb63bcb0189845f34d4db4cfcb08dc1bedde04b
triton/CMakeLists.txt

78 lines
2.4 KiB
CMake
Raw Normal View History

[BUILD] Added automatic nightly build releases to pip in CI; removed build-time dependence on LLVM and PyTorch (#77) Recently there has been more and more report about installation issues: - Installing Triton before upgrading pytorch can create some issues because Triton uses some torch headers - llvm-10-dev not available on some platform; llvm-11-dev not available on e.g. Ubuntu. absence of nightly builds This PR should fix all these issues. Some CMake tricks are used to download and install llvm at build time. Triton Python bindings were modified to remove dependence on pytorch ops. Midnight CI job added to generate binary wheels for all Triton version and update them on pypi's new triton-nightly project. This PR will also make it very easy to use LLVM forks in the future for whatever needs we have.
2021-03-22 20:03:37 -04:00
cmake_minimum_required(VERSION 3.6)
include(ExternalProject)
if(NOT TRITON_LLVM_BUILD_DIR)
set(TRITON_LLVM_BUILD_DIR ${CMAKE_BINARY_DIR})
endif()
History prior to this date belonged to the now deprecated ISAAC project, and was deleted to save space
2021-07-27 12:38:38 -07:00
project(triton)
include(CTest)
list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/cmake")
# Options
[PYTHON][TESTS][DOC] Various improvement of the API and code quality: * Simplified `triton.kernel` API to achieve lower latency: > .data_ptr() must now be passed as kernel argument. No more implicit conversion from torch.tensor > compilation options are now constant attributes, i.e., opt.d('VAR') becomes opt.VAR > torch.device must now be passed explicitly to triton.kernel (no longer inferred from torch.tensor arguments) * C++ tests moved to `python/tests/` * C++ tutorial created in `tutorials/` * Python tutorial created in python/tutorials/ * Version changed to 1.0alpha * No longer copying C++ headers into the Python package * added python/triton/ops/ package for pre-written Triton ops
2021-01-29 17:27:16 -05:00
option(BUILD_TUTORIALS "Build C++ Triton tutorials" ON)
History prior to this date belonged to the now deprecated ISAAC project, and was deleted to save space
2021-07-27 12:38:38 -07:00
option(BUILD_PYTHON_MODULE "Build Python Triton bindings" OFF)
# Default build type
if(NOT CMAKE_BUILD_TYPE)
message(STATUS "Default build type: Release")
set(CMAKE_BUILD_TYPE "Release")
endif()
# Compiler flags
include_directories(${CMAKE_CURRENT_SOURCE_DIR}/include)
[CI] Various improvements to CI (#137) Add clean-up before CI runs. Now using static LLVM-11 libraries from system rather than recompilation. Still no run-time LLVM dependencies
2021-07-22 11:41:51 -07:00
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -D__STDC_FORMAT_MACROS -std=gnu++17")
History prior to this date belonged to the now deprecated ISAAC project, and was deleted to save space
2021-07-27 12:38:38 -07:00
[BUILD] Added automatic nightly build releases to pip in CI; removed build-time dependence on LLVM and PyTorch (#77) Recently there has been more and more report about installation issues: - Installing Triton before upgrading pytorch can create some issues because Triton uses some torch headers - llvm-10-dev not available on some platform; llvm-11-dev not available on e.g. Ubuntu. absence of nightly builds This PR should fix all these issues. Some CMake tricks are used to download and install llvm at build time. Triton Python bindings were modified to remove dependence on pytorch ops. Midnight CI job added to generate binary wheels for all Triton version and update them on pypi's new triton-nightly project. This PR will also make it very easy to use LLVM forks in the future for whatever needs we have.
2021-03-22 20:03:37 -04:00
[CI] Various improvements to CI (#137) Add clean-up before CI runs. Now using static LLVM-11 libraries from system rather than recompilation. Still no run-time LLVM dependencies
2021-07-22 11:41:51 -07:00
# if(APPLE)
# set(CMAKE_OSX_SYSROOT "/")
# set(CMAKE_OSX_DEPLOYMENT_TARGET "")
# endif()
[BUILD] Added automatic nightly build releases to pip in CI; removed build-time dependence on LLVM and PyTorch (#77) Recently there has been more and more report about installation issues: - Installing Triton before upgrading pytorch can create some issues because Triton uses some torch headers - llvm-10-dev not available on some platform; llvm-11-dev not available on e.g. Ubuntu. absence of nightly builds This PR should fix all these issues. Some CMake tricks are used to download and install llvm at build time. Triton Python bindings were modified to remove dependence on pytorch ops. Midnight CI job added to generate binary wheels for all Triton version and update them on pypi's new triton-nightly project. This PR will also make it very easy to use LLVM forks in the future for whatever needs we have.
2021-03-22 20:03:37 -04:00
##########
# LLVM
##########
[CI] Various improvements to CI (#137) Add clean-up before CI runs. Now using static LLVM-11 libraries from system rather than recompilation. Still no run-time LLVM dependencies
2021-07-22 11:41:51 -07:00
find_package(LLVM 11 REQUIRED COMPONENTS "nvptx")
message(STATUS "Found LLVM ${LLVM_PACKAGE_VERSION}")
include_directories("${LLVM_INCLUDE_DIRS}")
if(APPLE)
set(CMAKE_OSX_DEPLOYMENT_TARGET "10.14")
endif()
History prior to this date belonged to the now deprecated ISAAC project, and was deleted to save space
2021-07-27 12:38:38 -07:00
# Python module
if(BUILD_PYTHON_MODULE)
message(STATUS "Adding Python module")
[PYTHON] CUTLASS wrapper for fair benchmarks (#75) Before this commit, the benchmarking infrastructure used heterogeneous protocols between library (e.g., CUTLASS uses a C++ binary that reports mean TFLOPS; torch and triton use python call and report 10th, 50th and 90th quantiles). For the sake of uniformity and fair benchmark practices, this PR adds a python wrapper for auto-tuned CUTLASS matrix multiplication. Benchmarks have been rewritten to use this wrapper with `triton.testing.do_bench` rather than system calls to CUTLASS profiler. Importantly, this also ensures that all the matmuls are done on the *same* input data which should stabilize clock across providers.
2021-03-09 16:32:44 -05:00
# Build CUTLASS python wrapper if requested
[BUILD] Added automatic nightly build releases to pip in CI; removed build-time dependence on LLVM and PyTorch (#77) Recently there has been more and more report about installation issues: - Installing Triton before upgrading pytorch can create some issues because Triton uses some torch headers - llvm-10-dev not available on some platform; llvm-11-dev not available on e.g. Ubuntu. absence of nightly builds This PR should fix all these issues. Some CMake tricks are used to download and install llvm at build time. Triton Python bindings were modified to remove dependence on pytorch ops. Midnight CI job added to generate binary wheels for all Triton version and update them on pypi's new triton-nightly project. This PR will also make it very easy to use LLVM forks in the future for whatever needs we have.
2021-03-22 20:03:37 -04:00
set(PYTHON_SRC_PATH ${CMAKE_CURRENT_SOURCE_DIR}/python/src)
[PYTHON] CUTLASS wrapper for fair benchmarks (#75) Before this commit, the benchmarking infrastructure used heterogeneous protocols between library (e.g., CUTLASS uses a C++ binary that reports mean TFLOPS; torch and triton use python call and report 10th, 50th and 90th quantiles). For the sake of uniformity and fair benchmark practices, this PR adds a python wrapper for auto-tuned CUTLASS matrix multiplication. Benchmarks have been rewritten to use this wrapper with `triton.testing.do_bench` rather than system calls to CUTLASS profiler. Importantly, this also ensures that all the matmuls are done on the *same* input data which should stabilize clock across providers.
2021-03-09 16:32:44 -05:00
set(CUTLASS_INCLUDE_DIR "$ENV{CUTLASS_INCLUDE_DIR}")
set(CUTLASS_LIBRARY_DIR "$ENV{CUTLASS_LIBRARY_DIR}")
if(NOT("${CUTLASS_INCLUDE_DIR}" STREQUAL "") AND NOT("${CUTLASS_LIBRARY_DIR}" STREQUAL ""))
[BUILD] Added automatic nightly build releases to pip in CI; removed build-time dependence on LLVM and PyTorch (#77) Recently there has been more and more report about installation issues: - Installing Triton before upgrading pytorch can create some issues because Triton uses some torch headers - llvm-10-dev not available on some platform; llvm-11-dev not available on e.g. Ubuntu. absence of nightly builds This PR should fix all these issues. Some CMake tricks are used to download and install llvm at build time. Triton Python bindings were modified to remove dependence on pytorch ops. Midnight CI job added to generate binary wheels for all Triton version and update them on pypi's new triton-nightly project. This PR will also make it very easy to use LLVM forks in the future for whatever needs we have.
2021-03-22 20:03:37 -04:00
set(CUTLASS_SRC ${PYTHON_SRC_PATH}/cutlass.cc)
[PYTHON] CUTLASS wrapper for fair benchmarks (#75) Before this commit, the benchmarking infrastructure used heterogeneous protocols between library (e.g., CUTLASS uses a C++ binary that reports mean TFLOPS; torch and triton use python call and report 10th, 50th and 90th quantiles). For the sake of uniformity and fair benchmark practices, this PR adds a python wrapper for auto-tuned CUTLASS matrix multiplication. Benchmarks have been rewritten to use this wrapper with `triton.testing.do_bench` rather than system calls to CUTLASS profiler. Importantly, this also ensures that all the matmuls are done on the *same* input data which should stabilize clock across providers.
2021-03-09 16:32:44 -05:00
add_definitions(-DWITH_CUTLASS_BINDINGS)
[DOCS] Added matrix multiplication tutorial
2021-03-14 18:49:59 -04:00
set(CUTLASS_LIBRARIES "cutlass.a")
[PYTHON] CUTLASS wrapper for fair benchmarks (#75) Before this commit, the benchmarking infrastructure used heterogeneous protocols between library (e.g., CUTLASS uses a C++ binary that reports mean TFLOPS; torch and triton use python call and report 10th, 50th and 90th quantiles). For the sake of uniformity and fair benchmark practices, this PR adds a python wrapper for auto-tuned CUTLASS matrix multiplication. Benchmarks have been rewritten to use this wrapper with `triton.testing.do_bench` rather than system calls to CUTLASS profiler. Importantly, this also ensures that all the matmuls are done on the *same* input data which should stabilize clock across providers.
2021-03-09 16:32:44 -05:00
endif()
message(STATUS ${CUTLASS_INCLUDE_PATH})
[BUILD] Added automatic nightly build releases to pip in CI; removed build-time dependence on LLVM and PyTorch (#77) Recently there has been more and more report about installation issues: - Installing Triton before upgrading pytorch can create some issues because Triton uses some torch headers - llvm-10-dev not available on some platform; llvm-11-dev not available on e.g. Ubuntu. absence of nightly builds This PR should fix all these issues. Some CMake tricks are used to download and install llvm at build time. Triton Python bindings were modified to remove dependence on pytorch ops. Midnight CI job added to generate binary wheels for all Triton version and update them on pypi's new triton-nightly project. This PR will also make it very easy to use LLVM forks in the future for whatever needs we have.
2021-03-22 20:03:37 -04:00
include_directories("." ${PYTHON_SRC_PATH} ${PYTHON_INCLUDE_DIRS} ${CUTLASS_INCLUDE_DIR})
[PYTHON] CUTLASS wrapper for fair benchmarks (#75) Before this commit, the benchmarking infrastructure used heterogeneous protocols between library (e.g., CUTLASS uses a C++ binary that reports mean TFLOPS; torch and triton use python call and report 10th, 50th and 90th quantiles). For the sake of uniformity and fair benchmark practices, this PR adds a python wrapper for auto-tuned CUTLASS matrix multiplication. Benchmarks have been rewritten to use this wrapper with `triton.testing.do_bench` rather than system calls to CUTLASS profiler. Importantly, this also ensures that all the matmuls are done on the *same* input data which should stabilize clock across providers.
2021-03-09 16:32:44 -05:00
link_directories(${PYTHON_LINK_DIRS} ${CUTLASS_LIBRARY_DIR})
[BUILD] Added automatic nightly build releases to pip in CI; removed build-time dependence on LLVM and PyTorch (#77) Recently there has been more and more report about installation issues: - Installing Triton before upgrading pytorch can create some issues because Triton uses some torch headers - llvm-10-dev not available on some platform; llvm-11-dev not available on e.g. Ubuntu. absence of nightly builds This PR should fix all these issues. Some CMake tricks are used to download and install llvm at build time. Triton Python bindings were modified to remove dependence on pytorch ops. Midnight CI job added to generate binary wheels for all Triton version and update them on pypi's new triton-nightly project. This PR will also make it very easy to use LLVM forks in the future for whatever needs we have.
2021-03-22 20:03:37 -04:00
set(PYTHON_SRC ${PYTHON_SRC_PATH}/main.cc ${PYTHON_SRC_PATH}/triton.cc ${PYTHON_SRC_PATH}/superblock.cc ${CUTLASS_SRC})
History prior to this date belonged to the now deprecated ISAAC project, and was deleted to save space
2021-07-27 12:38:38 -07:00
endif()
# Triton
file(GLOB_RECURSE LIBTRITON_SRC lib/*.cc)
add_library(triton SHARED ${LIBTRITON_SRC} ${PYTHON_SRC})
[CI] Various improvements to CI (#137) Add clean-up before CI runs. Now using static LLVM-11 libraries from system rather than recompilation. Still no run-time LLVM dependencies
2021-07-22 11:41:51 -07:00
target_link_options(triton PRIVATE ${LLVM_LDFLAGS})
target_link_libraries(triton ${LLVM_LIBRARIES} ${LLVM_SYSTEM_LIBS})
message(STATUS ${LLVM_LDFLAGS})
History prior to this date belonged to the now deprecated ISAAC project, and was deleted to save space
2021-07-27 12:38:38 -07:00
[RUNTIME] Lower-level interface for executing functions
2020-08-11 20:10:39 -04:00
if(BUILD_PYTHON_MODULE)
[CI] Various improvements to CI (#137) Add clean-up before CI runs. Now using static LLVM-11 libraries from system rather than recompilation. Still no run-time LLVM dependencies
2021-07-22 11:41:51 -07:00
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})
[RUNTIME] Lower-level interface for executing functions
2020-08-11 20:10:39 -04:00
endif()
Reference in New Issue Copy Permalink