[gmx-developers] Configure warning for MPI version, was: [gmx-users] New beta release: 4.5-beta3

[Roland Schulz]

Hi Bogdan,

this is exactly the OpenMPI problem I meant (
https://svn.open-mpi.org/trac/ompi/ticket/2043). I only suggested a warning
for GCC 4.4.x on x86/x64 because only this is affected. According to the bug
descriptions also older Intel CPUs are affected (and not only Nehalem).

Roland

On Thu, Aug 12, 2010 at 9:55 AM, Bogdan Costescu <bcostescu at gmail.com>wrote:

> On Wed, Aug 11, 2010 at 11:58 PM, Roland Schulz <roland at utk.edu> wrote:
> > OpenMPI 1.4.0 hangs on GCC 4.4.x on x86/x64
>
> Was this by any chance run on a Nehalem CPU using shared memory ? (not
> necessarily sm alone). This particular combination (gcc >= 4.4,
> Nehalem, sm) has been known to cause problems which are supposed to be
> fixed in OpenMPI 1.4.2. However this problem doesn't manifest itself
> on the same computers with older gcc, when not using sm or on
> computers with other CPUs. So I think that banning OpenMPI 1.4.0
> outright is a too tough measure.
>
> If you still have access to this combination, could you try running
> with 'mpirun -mca btl ^sm ...' ? This should disable sm and allow
> other mechanisms to be used for on-node communication (openib or tcp
> most likely).
>
> Cheers,
> Bogdan
> --
> gmx-developers mailing list
> gmx-developers at gromacs.org
> http://lists.gromacs.org/mailman/listinfo/gmx-developers
> Please don't post (un)subscribe requests to the list. Use the
> www interface or send it to gmx-developers-request at gromacs.org.
>



-- 
ORNL/UT Center for Molecular Biophysics cmb.ornl.gov
865-241-1537, ORNL PO BOX 2008 MS6309
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-developers/attachments/20100812/dcc28e19/attachment.html>