Hi,
On Tue, Apr 03, 2012 at 01:38:32AM +0000, B. Lachele Foley wrote:
> Patch fixed it. I can't help you understand why, but I will gladly applaud you for having made it so.
> > From: Jason Swails [jason.swails.gmail.com]
> >
> > GNU 4.5.3 + OpenMPI 1.4.5
> > GNU 4.6.1 + OpenMPI 1.4.3
> > (It did NOT happen with GNU 4.1.2 + OpenMPI 1.4.5)
> >
> > (I believe it also happened with OpenMPI 1.5.4). The issue does not crop
> > up with those compilers using mpich2-1.4.1p1, so it is strictly the GNU 4.5
> > (maybe 4.4?) or GNU 4.6 compilers combined with OpenMPI 1.4+. (The issue
> > is 100% reproducible on these systems, though). The NEB tests require 4+
> > processors to run, so you should be able to test this by going to
> >
> > $AMBERHOME/test/neb-testcases/neb_gb_partial/ and doing the following:
> >
> > export TESTsander=$AMBERHOME/bin/sander.MPI
> > export DO_PARALLEL='mpirun -np 4' # or 8
> > ./Run.neb_gb_partial
> >
> > The test should pass almost instantly. I've attached a patch that fixes
> > the issue for me (which was checked into the master branch recently), but I
> > do not understand why the patch is necessary in the first place. (apply
> > the patch from $AMBERHOME using 'patch -p0 -N < neb.patch')
> >
> > Any indication that I'm not going crazy is appreciated (or even better, if
> > you can shed light on the actual origin of the problem).
Based on your commit log, it looks like you have identified the likely
origin as a bug, likely in openmpi:
commit 641f88d3abb87109f026de34ec32996789c48211
While glancing at this i noticed a block of commented code in
subroutine pimd_full_spring_force ( x, f, mass, Epot_spring, Epot_deriv, dvdl )
Such blocks should have a description of their purpose.
Since i found no such description, i removed the block.
scott
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Mon Apr 23 2012 - 17:30:03 PDT
