On Mon, 2012-04-02 at 12:35 -0400, tluchko wrote:
> On 03/29/2012 11:28 PM, David A Case wrote:
> > I have placed updated tarballs here:
> >
> > http://ambermd.org/downloads/AmberTools12.29mar12.tar.bz2
> > http://ambermd.org/downloads/Amber12.29mar12.tar.bz2
> >
>
> I've posted tests on CentOS 5.7 for GNU, Intel and PGI compilers. To
> summarize:
>
> 1) The majority of failures are differences in the last digit and are
> likely just round-off errors. I.e., these should really pass. These
> false positives would go away if we doubled the error tolerance. E.g.,
> dacdif -r 1e-6
> becomes
> dacdif -r 2e-6
> This should let round-offs through but catch everything else.
>
> 2) rem_hybrid has problems with three threads.
That's because it doesn't run with 3 threads. It requires a multiple of
2 threads, but this is surprisingly hard to test for (without explicitly
doing an "if == 2, 4, 6, 8, ... etc) in csh (or bash for that matter).
Parallel compilation dumps a program "numprocs" into the test
directories that will just print out the comm_world size, but it could
be modified to take command-line flags and spit out errors when illegal
processor counts are detected... Too late for this release, though.
Should we be worried about the fact that the rem_hybrid test fails when
we run with an odd number of threads? I feel like 3 is a strange number
to test with, and it's unlikely many people will try.
>
> 3) the fibre-diffraction module has problems with 10 threads when built
> with GNU and PGI.
>
> 4) MMPBSA_PY fails with PGI on two threads.
But from what I saw, just test 06 (Nmode), right? This is the same
issue that was brought up before, and I claim it's not MMPBSA.py's
fault, nor is it something we should worry about (I know there's not
universal agreement on that last part, but the low-frequency modes of
such silly systems are _very_ unstable and can significantly impact
results -- but that's for post-release conversation/investigation, if I
recall correctly).
Thanks!
Jason
--
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
352-392-4032
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Mon Apr 02 2012 - 10:30:03 PDT
