Skip to content

Use MPI-3 shared memory windows for global index mappings

Sebastian Ohlmann requested to merge mpi_shared_memory into develop

Description

Like this, the corresponding memory for the global index mappings needs to be allocated only once per compute node and all other ranks on the node can access this memory. Especially on newer HPC systems with many cores per node, this can save substantial amounts of memory.

News snippet

Use MPI-3 shared memory windows for global index mappings

Checklist

  • I have checked that my code follows the Octopus coding standards
  • I have added tests for all the new features added in this request.

Merge request reports