NOTE: For the latest stable README.md ensure you are on the main
branch.
This is a collection of CMake modules that are useful for all CUDA RAPIDS projects. By sharing the code in a single place it makes rolling out CMake fixes easier.
The rapids-cmake
module is designed to be acquired via CMake's Fetch
Content into your project.
cmake_minimum_required(...)
if(NOT EXISTS ${CMAKE_CURRENT_BINARY_DIR}/<PROJECT>_RAPIDS.cmake)
file(DOWNLOAD https://raw.githubusercontent.com/rapidsai/rapids-cmake/branch-<VERSION_MAJOR>.<VERSION_MINOR>/RAPIDS.cmake
${CMAKE_CURRENT_BINARY_DIR}/<PROJECT>_RAPIDS.cmake)
endif()
include(${CMAKE_CURRENT_BINARY_DIR}/<PROJECT>_RAPIDS.cmake)
include(rapids-cmake)
include(rapids-cpm)
include(rapids-cuda)
include(rapids-export)
include(rapids-find)
project(....)
Note that we recommend you install rapids-cmake
into the root CMakeLists.txt
of
your project before the first project
call. This allows us to offer features such as
rapids_cuda_architectures()
rapids-cmake
provides a collection of useful CMake settings that any RAPIDS project may use.
While they maybe common, we know that they aren't universal and might need to be composed in
different ways.
To use function provided by rapids-cmake
projects have two options:
- Call
include(rapids-<component>)
as that imports all commonly used functions for that component - Load each function independently via
include(${rapids-cmake-dir}/<component>/<function_name>.cmake)
Complete online documentation for all components can be found at:
https://docs.rapids.ai/api/rapids-cmake/nightly/api.html
The rapids-cmake
module contains helpful general CMake functionality
rapids_cmake_build_type( )
handles initialization ofCMAKE_BUILD_TYPE
rapids_cmake_support_conda_env( target [MODIFY_PREFIX_PATH])
Establish a target that holds the CONDA environment include and link directories.rapids_cmake_write_version_file( <file> )
Write a C++ header with a projects MAJOR, MINOR, and PATCH defines
The rapids-cpm
module contains CPM functionality to allow projects to acquire dependencies consistently.
For consistency, all targets brought in via rapids-cpm
are GLOBAL targets.
rapids_cpm_init()
handles initialization of the CPM module.rapids_cpm_find(<project> name BUILD_EXPORT_SET <name> INSTALL_EXPORT_SET <name>)
Will search for a module and fall back to installing via CPM. Offers support to track dependencies for easy package exporting
The rapids-cuda
module contains core functionality to allow projects to build CUDA code robustly.
The most commonly used function are:
rapids_cuda_init_architectures(<project_name>)
handles initialization ofCMAKE_CUDA_ARCHITECTURE
. MUST BE CALLED BEFOREPROJECT()
rapids_cuda_init_runtime(<mode>)
handles initialization ofCMAKE_CUDA_RUNTIME_LIBRARY
.rapids_cuda_patch_toolkit()
corrects bugs in the CUDAToolkit module that are being upstreamed.
The rapids_cython
functions allow projects to easily build cython modules using
scikit-build.
rapids_cython_init()
handles initialization of scikit-build and cython.rapids_create_modules([CXX] [SOURCE_FILES <src1> <src2> ...] [LINKED_LIBRARIES <lib1> <lib2> ... ] [INSTALL_DIR <install_path>] [MODULE_PREFIX <module_prefix>] )
will create cython modules for each provided source file
The rapids-export
module contains core functionality to allow projects to easily record and write out
build and install dependencies, that come from find_package
or cpm
rapids_export(<type> <project> EXPORT_SET <name>)
write out all the require components of a projects config module so that theinstall
orbuild
directory can be imported viafind_package
. Seerapids_export
documentation for full documentation
The rapids-find
module contains core functionality to allow projects to easily generate FindModule
or export find_package
calls:
The most commonly used function are:
rapids_find_package(<project_name> BUILD_EXPORT_SET <name> INSTALL_EXPORT_SET <name> )
Combinesfind_package
and support to track dependencies for easy package exportingrapids_generate_module(<PackageName> HEADER_NAMES <paths...> LIBRARY_NAMES <names...> )
Generate a FindModule for the given package. Allows association to export sets so the generated FindModule can be shipped with the project
The rapids_test
functions simplify CTest resource allocation, allowing for
tests to run in parallel without overallocating GPU resources.
The most commonly used functions are:
rapids_test_add(NAME <test_name> GPUS <N> PERCENT <N>)
: State how many GPU resources a single test requires
At times projects or developers will need to verify rapids-cmake
branches. To do this you can set variables that control which repository RAPIDS.cmake
downloads, which should be done like this:
# To override the version that is pulled:
set(rapids-cmake-version "<version>")
# To override the GitHub repository:
set(rapids-cmake-repo "<my_fork>")
# To use an exact Git SHA:
set(rapids-cmake-sha "<my_git_sha>")
# To use a Git tag:
set(rapids-cmake-tag "<my_git_tag>")
# To override the repository branch:
set(rapids-cmake-branch "<my_feature_branch>")
# Or to override the entire repository URL (e.g. to use a GitLab repo):
set(rapids-cmake-url "https://gitlab.com/<my_user>/<my_fork>/-/archive/<my_branch>/<my_fork>-<my_branch>.zip")
file(DOWNLOAD https://raw.githubusercontent.com/rapidsai/rapids-cmake/branch-22.10/RAPIDS.cmake
${CMAKE_CURRENT_BINARY_DIR}/RAPIDS.cmake)
include(${CMAKE_CURRENT_BINARY_DIR}/RAPIDS.cmake)
A few notes:
- An explicitly defined
rapids-cmake-url
will always be used rapids-cmake-sha
takes precedence overrapids-cmake-tag
rapids-cmake-tag
takes precedence overrapids-cmake-branch
- It is advised to always set
rapids-cmake-version
to the version expected by the repo your modifications will pull
Review the CONTRIBUTING.md file for information on how to contribute code and issues to the project.