Installation¶

Python package¶

From PyPI¶

python -m pip install libmat

Prebuilt wheels (CPU version) are available for Linux x86_64 and macOS (Intel, Apple silicon), Python ≥ 3.10. They have no dependencies besides numpy. Linux on ARM is not supported yet (geogram 1.7 uses x86 assembly there).

From source¶

Needed on other platforms and for the CUDA version; see the dependencies below.

git clone https://github.com/ningnawang/libmat.git
cd libmat
pip install .

CUDA version¶

On a machine with an NVIDIA GPU and the CUDA toolkit, build from source:

LIBMAT_WITH_CUDA=ON pip install .                          # from a clone
LIBMAT_WITH_CUDA=ON pip install libmat --no-binary libmat  # from the PyPI sdist

import libmat
libmat.has_cuda()   # True

The CUDA build contains everything of the CPU build, plus medial_mesh_from_spheres and the GPU path of distance_to_medial_mesh. Set CMAKE_CUDA_ARCHITECTURES (e.g. -C cmake.define.CMAKE_CUDA_ARCHITECTURES=86) to build for a GPU other than the one of the build machine.

Optional extras¶

pip install ".[viz]"    # polyscope, used by the examples
pip install ".[test]"   # pytest:  python -m pytest tests
pip install ".[docs]"   # mkdocs:  mkdocs serve

Development build¶

Build the extension in place and point PYTHONPATH to the sources:

cmake -B build -DLIBMAT_BUILD_PYTHON=ON
cmake --build build -j
export PYTHONPATH=$PWD/python
python -m pytest tests

Dependencies¶

Only needed to build from source.

dependency	how it is obtained
C++17 compiler, CMake ≥ 3.18	system
Boost headers	system: `brew install boost` / `apt install libboost-dev`
CGAL ≥ 5.6	system (`brew install cgal` / `apt install libcgal-dev`), downloaded if missing
GMP / MPFR	optional, used by CGAL if found (`-DCGAL_DISABLE_GMP=ON` to skip)
Eigen 3.4	system, downloaded if missing
geogram 1.7.9	downloaded and built statically
pybind11	pip build dependency / downloaded
CUDA toolkit ≥ 11	only for the CUDA build

C++ library¶

cmake -B build -DCMAKE_BUILD_TYPE=Release            # CPU only
cmake -B build -DCMAKE_BUILD_TYPE=Release -DLIBMAT_WITH_CUDA=ON
cmake --build build -j

CMake option	default
`LIBMAT_WITH_CUDA`	`OFF`	build the CUDA modules (`libmat::cuda`)
`LIBMAT_BUILD_PYTHON`	`OFF`	build the python extension module
`LIBMAT_BUILD_VIEWER`	`OFF`	build the polyscope viewer in `apps/viewer`
`LIBMAT_GEOGRAM_DIR`	empty	use an existing geogram checkout instead of downloading it
`LIBMAT_AUTO_DOWNLOAD`	`ON`	download missing dependencies

The old option names LIBMAT_IS_USE_GPU and LIBMAT_IS_BUILD_TEST are still accepted. See C++ library to use libmat from your own project.