Rodin::MPI namespace

Key Components

• Context::MPI — Wraps a boost::mpi::environment and boost::mpi::communicator to provide a uniform execution context.
• Mesh<Context::MPI> — Distributed mesh storing a rank-local Shard with ownership, ghost, and distributed index metadata. Geometric queries (getPolytopeCount, getVolume, getPerimeter, etc.) perform MPI reductions automatically.
• Sharder — Distributes a global mesh across MPI ranks in three steps: shard(), scatter(), gather(). Also available as one-step distribute().
• Shard — Rank-local view of a partitioned mesh. Classifies every polytope as Owned, Shared, or Ghost and maintains bidirectional local↔distributed index maps.
• reconcile(d) — After local connectivity discovery, reconciles entities of dimension d across ranks so that shared entities have consistent distributed IDs and ownership.

Typical Workflow

#include <Rodin/MPI.h>

boost::mpi::environment env(argc, argv);
boost::mpi::communicator world;
Context::MPI mpi(env, world);

// Option A: UniformGrid (partition + distribute in one call)
auto mesh = Mesh<Context::MPI>::UniformGrid(mpi, Polytope::Type::Tetrahedron, {16,16,16});

// Option B: explicit partition → shard → scatter → gather
MPI::Sharder sharder(mpi);
if (world.rank() == 0)
{
  LocalMesh mesh;
  mesh = mesh.UniformGrid(Polytope::Type::Tetrahedron, {16,16,16});
  mesh.getConnectivity().compute(cellDim, cellDim);
  BalancedCompactPartitioner partitioner(mesh);
  partitioner.partition(world.size());
  sharder.shard(partitioner);
  sharder.scatter(0);
}
auto mesh = sharder.gather(0);

// Compute distributed connectivity and reconcile shared entities
mesh.getConnectivity().compute(2, 3);
mesh.reconcile(2);

Submodules

• Assembly — MPI-parallel assembly strategies for PETSc-backed forms
• Context — Context::MPI execution context
• Geometry — Distributed mesh and sharder
• IO — HDF5 and MFEM IO specializations for distributed meshes
• Variational — Distributed FE spaces (P1 on MPI mesh)

Typedefs

using Sharder = Geometry::Sharder<Context::MPI>

Convenience alias for the MPI mesh sharder specialization.

using P1 = Variational::P1<Real, Geometry::Mesh<Context::MPI>>

Convenience alias for the default distributed scalar P1 space.

Typedef documentation