Re: [AMBER-Developers] incompatible processor requirements in parallel tests

From: Jason Swails <>
Date: Thu, 18 Feb 2010 14:32:26 -0500

On Thu, Feb 18, 2010 at 11:55 AM, Lachele Foley <> wrote:
> Even though I now have CVS access, I don't feel qualified to fix this myself (except maybe to add some documentation).
> Some parallel tests require np=2, some require np>2, at least one requires np=1.  But, these skipped tests don't show up in "TEST_FAILURES" and aren't documented.  So, unless the user runs the stdout from the tests into a file and checks it (like I do), they never know.  I think this could be fixed with some documentation and perhaps a segregation of parallel tests (make test.parallel.np2, etc.).  If there is documentation, I didn't find it in the places I expected it.

This is what I do (make test.parallel | tee ampar.test) and then i can
look at ampar.test and see how it went. After seeing several of those
messages, I just decided that whenever I test in parallel, I'll just
run 2 threads, 4 threads, and 8 threads (though there is one that uses
1 and one that uses 32). I grep-ed out those that claim to need a
specific number of threads and created rules in the Makefile for that
number of processors (i.e. make test.parallel.4proc
test.parallel.8proc test.parallel.32proc). I did this because in
several cases, using 4/8/32 processors, you wouldn't necessarily want
to re-run every test (since for me, several failed when i went to 8
threads that passed on 2 and 4, though I only have a 4-core system).
But this doesn't prevent you from doing that if you want.

I checked in the new rules to Makefile. People can feel free to
remove it if they think it's pointless. But I agree with Lachele that
there should probably be some documentation about it (at least some
blurb in the README).


Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Graduate Student
AMBER-Developers mailing list
Received on Thu Feb 18 2010 - 12:00:02 PST
Custom Search