Hi, On Fri, Jul 07, 2017 at 06:15:28PM +0800, Drew Parsons wrote: > On Fri, 2017-07-07 at 11:47 +0200, Michael Banck wrote: > > We (Debichem) also have a few upstreams that claim OpenMPI is more > > buggy > > than MPICH. > > Interesting. Can you check in with them to see whether they still > think that's the case, or was that poor experience with older OpenMPI > versions?
Well, I think it is more a general feeling towards the MPI_THREAD_MULTIPLE issues that plagued OpenMPI for a long time and nobody at OpenMPI seemed to take serious. See e.g. https://github.com/ValeevGroup/tiledarray which references https://github.com/open-mpi/ompi/issues/157 (since closed) or https://github.com/nubakery/bagel/issues/8#issuecomment-32286225 or https://github.com/jeffhammond/armci-mpi#status All of those might be obsolete, but it might be difficult to get everybody to validate their codes on OpenMPI again. Michael
