Removes PME rank dependency on Non-local X H2D copy when buffer ops are on GPU but...
authorGaurav Garg <gaugarg@nvidia.com>
Wed, 25 Aug 2021 08:56:18 +0000 (14:26 +0530)
committerGaurav Garg <gaugarg@nvidia.com>
Fri, 27 Aug 2021 07:51:51 +0000 (07:51 +0000)
commit528af46fb325b35bea933f488d213aa789e47a35
tree6e8b562f57b9b0a83698444d5d47e46cbde307a0
parent1f45888bff3945adb4661ee5b9984c121267dd30
Removes PME rank dependency on Non-local X H2D copy when  buffer ops are on GPU but, Halo exchange happens over host-MPI.

This was not a problem until now because because !stepWork.haveGpuPmeOnThisRank will always evaluate to true in case of domain decomposition as PME will always be on a separate rank.
But now with PME decomposition this bug gets exposed.

Refs #4125
src/gromacs/mdlib/sim_util.cpp