This folder contains the C++ and CUDA code of the algorithms and ML primitives of cuML. The build system uses CMake for build configuration, and an out-of-source build is recommended.

The source code of cuML is divided in three main directories: src, src_prims, and comms.

• src contains the source code of the Machine Learning algorithms, and the main cuML C++ API. The main consumable is the shared library libcuml++, that can be used stand alone by C++ consumers or is consumed by our Python package cuml to provide a Python API.
• src_prims contains most of the common components and computational primitives that form part of the machine learning algorithms in cuML, and can be used individually as well in the form of a header only library.
• comms contains the source code of the communications implementations that enable multi-node multi-GPU algorithms. There are currently two communications implementations. The implementation in the mpi directory is for MPI environments. It can also be used for automated tested. The implementation in the std directory is required for running cuML in multi-node multi-GPU Dask environments.

The test directory has subdirectories that reflect this distinction between the src and prims components of cuML.

1. cmake (>= 3.14)
2. CUDA (>= 10.0)
3. gcc (>=5.4.0)
4. BLAS - Any BLAS compatible with cmake's FindBLAS. Note that the blas has to be installed to the same folder system as cmake, for example if using conda installed cmake, the blas implementation should also be installed in the conda environment.
5. clang-format (= 8.0.1) - enforces uniform C++ coding style; required to build cuML from source. The packages clang=8 and clang-tools=8 from the conda-forge channel should be sufficient, if you are on conda. If not using conda, install the right version using your OS package manager.
6. UCX with CUDA support [optional] (>=1.7) - enables point-to-point messaging in the cuML communicator.

The main artifact produced by the build system is the shared library libcuml++. Additionally, executables to run tests for the algorithms can be built. To see detailed steps see the BUILD document of the repository.

Current cmake offers the following configuration options:

• Build Configuration Options:

• Library Location options:

• Debug configuration options:

After running CMake in a build directory, if the BUILD_* options were not turned OFF, the following targets can be built:

$ make -j # Build libcuml++ and all tests
$ make -j sg_benchmark # Build c++ cuml single gpu benchmark
$ make -j cuml++ # Build libcuml++
$ make -j ml # Build ml_test algorithm tests binary
$ make -j ml_mg # Build ml_mg_test multi GPU algorithms tests binary
$ make -j prims # Build prims_test ML primitive unit tests binary

The external folder contains submodules that cuML depends on.

Current external submodules are:

1. CUTLASS
2. CUB
3. Faiss
4. Google Test

After building cuML, you can use its functionality in other C/C++ applications by linking against the generated libraries. The following trivial example shows how to make external use of cuML's logger:

// main.cpp
#include <cuml/common/logger.hpp>

int main(int argc, char *argv[]) {
  CUML_LOG_WARN("This is a warning from the cuML logger!");
  return 0;
}

To compile this example, we must point the compiler to where cuML was installed. Assuming you did not provide a custom $CMAKE_INSTALL_PREFIX, this will default to the $CONDA_PREFIX environment variable.

$ export LD_LIBRARY_PATH="${CONDA_PREFIX}/lib"
$ nvcc \
       main.cpp \
       -o cuml_logger_example \
       "-L${CONDA_PREFIX}/lib" \
       "-I${CONDA_PREFIX}/include" \
       "-I${CONDA_PREFIX}/include/cuml/raft" \
       -lcuml++
$ ./cuml_logger_example
[W] [13:26:43.503068] This is a warning from the cuML logger!