When I built HDF5 1.10.0 with MVAPICH2 2.2b (GPUdirect version) and PGI
16.4, I got this hang during the make check:
Testing t_cache
I made the hang go away with this setting:
export MV2_ENABLE_AFFINITY=0
I'd seen a similar problem (and fix) with the make check for the Score-P
profiler, built under the same set of conditions.
What I *think* is going on, is that the configurator recognizes the
MVAPICH2 libraries as being MPICH-compatible, and tries to use MPICH
environment settings to manage the process-to-core bindings.
But the MVAPICH2 environment settings use different names for the
settings, so it falls back to default behavior, which could cause an
oversubscription of the processor cores and a hang in this case.
Can you confirm this for me?
Thanks,
Carl Ponder
-----------------------------------------------------------------------------------
This email message is for the sole use of the intended recipient(s) and may
contain
confidential information. Any unauthorized review, use, disclosure or
distribution
is prohibited. If you are not the intended recipient, please contact the
sender by
reply email and destroy all copies of the original message.
-----------------------------------------------------------------------------------
_______________________________________________
Hdf-forum is for HDF software users discussion.
[email protected]
http://lists.hdfgroup.org/mailman/listinfo/hdf-forum_lists.hdfgroup.org
Twitter: https://twitter.com/hdf5
