Is anyone else having problems running something as simple as test_basic on
OSX 10.10? Things were running fine about a week ago, but I am not sure if
a recent PR or OSX update has screwed things up. When running test_basic
from the official release of openmc/develop, eigenvalue output is messed up
(tallies are also messed up in other tests with NaNs). See below. Any
help would be greatly appreciated before I dig too much into this.
It appears like the MPI_IN_PLACE used in tally.F90 is not functioning correctly on my mac with mpich-3.1.4. Instead of returning the correct value of the MPI_SUM, it is returning 0.0 . If anyone has experienced this before, quick fixes would be appreciated. Thanks!
Hi Matt- does it work ok with other versions of MPICH?
Paul,
I haven’t tried it with an earlier version of MPICH, but OpenMPI (1.8.4) works just fine.
I found an old thread on OpenMPI where the same issue was seen (http://www.open-mpi.org/community/lists/users/2011/11/17862.php). They suggested to recompile with the additional fortran flags:
compiler_flags=-Wl,-commons,use_dylibs
This seems to fix the error with MPICH. I don’t really have any insights into the underlying reason for this problem. I’ve never had to recompile previous versions of MPICH, so maybe this is just a ‘bug’ introduced in the newest version?