C++ library¶

Using libmat in a CMake project¶

add_subdirectory(libmat)          # or FetchContent
target_link_libraries(my_app PRIVATE libmat::libmat)

target	content
`libmat::core`	CPU library: functional API and all CPU modules
`libmat::cuda`	CUDA modules (only with `LIBMAT_WITH_CUDA=ON`)
`libmat::libmat`	everything that was built

<libmat/config.h> defines LIBMAT_WITH_CUDA (0 or 1) and libmat::has_cuda().

Functional API¶

Every function of the Python API is a C++ function in namespace libmat, declared in its own header (one function per file, like libigl), taking and returning Eigen matrices:

#include <libmat/read_surface_mesh.h>
#include <libmat/shrink_spheres.h>

Eigen::MatrixXd V, C;
Eigen::MatrixXi F;
Eigen::VectorXd R;
Eigen::VectorXi types;
libmat::read_surface_mesh("data/mesh_69058.obj", V, F);

libmat::ShrinkSpheresOptions options;
options.num_spheres = 2000;
libmat::shrink_spheres(V, F, options, C, R, types);

#include <libmat/libmat.h> includes all of them. The C++ declaration of each function is shown at the bottom of its page in the Python API.

Low-level modules¶

The functional API is a thin layer over lower-level modules, which remain available for writing full pipelines:

include/libmat/
├── <function>.h        functional API (Eigen in / out)
├── common/             common.h, common_cxx.h, common_geogram.h, predicates.h
├── inputs/             TetMesh, SurfaceMesh, AABBWrapper, Parameter,
│                       sharp / concave feature detection
├── matbase/            MedialSphere, MedialMesh, medial primitives (cones, slabs)
├── matfun/             shrinking, updating (sphere optimization), thinning
├── io/                 .msh/.tet/.vtk, .ma, .sph, surface meshes
├── rpd3d/              regular triangulation (CGAL), ConvexCellHost,
│                       power cell update from RPD output
└── cuda/               (LIBMAT_WITH_CUDA=ON)
    ├── common/         CUDA helpers
    ├── rpd3d/          restricted power diagram on the GPU (RPD3D_GPU), kNN
    ├── dist2mat/       GPU distance to medial primitives
    ├── matfun_fix/     topology / feature / geometry fixes
    └── io/             RPD export (Houdini)
src/                    implementations, same layout

module	main entry points
`inputs`	`TetMesh`, `SurfaceMesh`, `detect_mark_sharp_features()`
`matbase`	`MedialSphere`, `MedialMesh`, `get_triangles_from_three_spheres()`
`matfun`	`init_and_shrink()`, `shrink_sphere()`, `iterate_sphere()`, `Thinning::prune()`
`io`	`load_tet()`, `get_surface_from_tet()`, `load_mat_clean()`, `export_ma_clean()`
`rpd3d`	`generate_RT_CGAL_and_mark_valid_spheres()`, `update_power_cells()`
`cuda/rpd3d`	`RPD3D_GPU::calculate()`, `compute_clipped_voro_diagram_GPU()`
`cuda/dist2mat`	`compute_closest_dist2mat()`
`cuda/matfun_fix`	`fix_topo_by_adding_new_sphere()`, `check_and_fix_external_feature()`, `check_and_fix_mm_geo()`

Conventions of the low-level modules

Inputs are expected in [0, 1000]^3 (see normalize_vertices); the functional API normalizes internally.
Call GEO::initialize() once before using them.
MedialMesh and RPD3D_GPU delete the objects they point to (MedialMesh::vertices, RPD3D_GPU::tet_mesh, ...), so pass heap allocated objects.

Viewer¶

apps/viewer is a small polyscope viewer for a tet mesh and a .ma file (-DLIBMAT_BUILD_VIEWER=ON):

./build/apps/viewer/libmat_viewer model.msh model.ma