Advanced Parallel Methods

The methods here build on the basic MPI workflow described in MPI Distribute Domain-first parallel workflow (distribute). They are intended for large production runs, HPC cluster deployments, or situations where the standard distribute approach is not sufficient (e.g. meshes too large to fit in memory on a single node).

• MPI Distribute Mesh-first parallel workflow (distribute_basic_mesh)
• Offline domain partitioning (sequential_distribute_dump / sequential_distribute_load)
• Offline mesh partitioning (sequential_mesh_dump / sequential_mesh_load)
• Uniform mesh refinement
• GPU Acceleration (OpenMP target offloading)

See also

Parallelisation

The simple parallel workflow (OpenMP and basic MPI distribute) that most users should start with.