2019.1 mpiexec worked as expected - mpiexec -np 2 -hosts A,B P issued one process P on each of hosts A and B.
2019.5 mpiexec -np 2 -hosts A,B -P issues two P's on A.
It's maddening because this is not how any other MPI I've experienced works...
mpiexec -configfile CF works the same in both releases. With this I can make things work, but it's very, VERY annoying.
Meanwhile, 2019.5 MPI_BCast() appears to be better behaved than 2019.1 MPI_BCast(). This is a good thing, and worth the upgrade.
I'm not in to spending 2-1/2 hours on each release to see if 2019.2 through 2019.4 might have better overall behavior.
TCE Level:
TCE Open Date:
Wednesday, January 15, 2020 - 18:07