Enable parallel runs
Dorie requires MPI as a dependency and supports creating a random field in parallel. The main routine is always compiled with parallel support. However, the current
SEQ_SuperLU linear solver backend lacks parallel computation abilities. We intend to switch to the OVLP_AMG_4_DG backend for this task. But assembling the (processor) boundary constraints correctly is quite the effort.
Also: In the past, the parallel solver proved to be inferior to the sequential one when computing a 'small' (not memory-bound) problem.
Tasks for going parallel:
- Enable collective read of HDF5 parameter field file
revisit verbosity: only rank 0 should print into the
- find working combinations of
Hint by Marian: Use
OVLP_AMG_4_DGfor parallel UG with an
AllEntitySet. Contraints are probably
P0ParallelConstraints, the subspace has to be P0 (which is already implemented for parallel YaspGrid).
- employ NOVLP solver when using UG. Need new templated type helper