GitLab

Summary

OPAL hangs when running FFA in parallel

Possible fixes

Ring::apply has a call to lossDS_m->save(); which in turn calls ippl::allreduce. This is a type of MPI reduce. At the same time, ParallelCyclotronTracker::deleteParticle calls allreduce(flagNeedUpdate, 1, std::logical_or<bool>());. So the procs can never align and we get a hang.

Fix is to call lossDS_m->save() in the Ring::finalise method, which seems to be the approach taken by Cyclotron.