[MITgcm-devel] [MITgcm-cvs] MITgcm/verification/halfpipe_streamice/results CVS Commit

[Daniel Goldberg]

OK, thank you.

I might have found the bug, but to be certain, what is the standard mpi
test? Will

./testresport -t halfpipe_streamice -mpi

be sufficient? I ask because the standard test already uses 2 tiles, and so
i would think that the 2-proc mpi test (above) would not give different
results.



On Sat, Sep 6, 2014 at 4:18 PM, Jean-Michel Campin <jmc at ocean.mit.edu>
wrote:

> Hi Dan,
>
> On Sat, Sep 06, 2014 at 03:39:22PM +0100, Daniel Goldberg wrote:
> > Hi J-M,
> >
> > i did not actually see your last email, but i knew that i had broken
> openad
> > so i fixed it.
> >
> > i did not realize mpi was breaking, i did not do a -mpi verification test
> > to check, but will in the future. i did not know about the mpi test, but
> it
> > is matched against output.txt?
> testreport compares with the same output.txt
>
> >i didn't realize i had broken it, it could be due to one of my recent
> changes.
> OK, no problems.
>
> > Is the STREAMICE_FP_ERROR the only
> > difference? if so i suspect i know the reason but will need to look into
> > this. thank you for letting me know.
> >
> > dan
>
> To be more presice regarding MPI test:
> 1) all the other than STREAMICE_FP_ERROR numbers that testreport collect
>  have not changed, so it's good.
> 2) STREAMICE_FP_ERROR output are sensitive to which compiler we
> use, but the differences from compilers used to be (and, without MPI,
>  are still) matching for 6 or 7 digits
> 3) with the latest changes, the STREAMICE_FP_ERROR that we get with
>  MPI are different enough from a non MPI output that we don't get
>  any digits matching (whereas it use to be perfect matching between
>  MPI and non MPI results).
>
> Cheers,
> Jean-Michel
>
> > On Sat, Sep 6, 2014 at 3:24 PM, Jean-Michel Campin <jmc at ocean.mit.edu>
> > wrote:
> >
> > > Hi Dan,
> > >
> > > OK, you fixed the OpenAD halfpipe_streamice, and it worked well
> > > (last night daily testing).
> > >
> > > But there might be something strange with the forward test
> > > when using MPI:
> > > As of this morning, all the MPI halfpipe_streamice tests fail because
> > > the STREAMICE_FP_ERROR (that testreport checks to decide for a fail of
> > > pass)
> > > is different enough (it used to be identical between a 2 procs MPI run
> and
> > > the reference 1 proc output):
> > > for instance, on baudelaire with gfortran & mpich2-1.3,
> > > the 1rst STREAMICE_FP_ERROR is now:
> > > MPI 2 procs:
> > > < (PID.TID 0000.0001) STREAMICE_FP_ERROR =  5.42568206198871E-05
> > > --- reference 1 proc output (also = old MPI 2 procs):
> > > > (PID.TID 0000.0001) STREAMICE_FP_ERROR =  6.72968409916551E-05
> > >
> > > Cheers,
> > > Jean-Michel
> > >
> > > On Fri, Sep 05, 2014 at 02:57:53PM -0400, Jean-Michel Campin wrote:
> > > > Hi Dan,
> > > > Could you check also the Open-AD halfpipe_streamice test ?
> > > > For some reasons, I am now getting some errors when trying to
> > > > compile with OpenAD (on baudelaire).
> > > > Cheers,
> > > > Jean-Michel
> > > >
> > > > On Fri, Sep 05, 2014 at 02:43:24PM -0400, Daniel N. Goldberg,
> 54-1423,
> > > 6172532977,  wrote:
> > > > > Update of /u/gcmpack/MITgcm/verification/halfpipe_streamice/results
> > > > > In directory forge:/tmp/cvs-serv9683
> > > > >
> > > > > Modified Files:
> > > > >     output_adm.txt
> > > > > Log Message:
> > > > > values changed after last commit, not sure why
> > > > >
> > > > >
> > > > > _______________________________________________
> > > > > MITgcm-cvs mailing list
> > > > > MITgcm-cvs at mitgcm.org
> > > > > http://mitgcm.org/mailman/listinfo/mitgcm-cvs
> > > >
> > > > _______________________________________________
> > > > MITgcm-devel mailing list
> > > > MITgcm-devel at mitgcm.org
> > > > http://mitgcm.org/mailman/listinfo/mitgcm-devel
> > >
> > > _______________________________________________
> > > MITgcm-devel mailing list
> > > MITgcm-devel at mitgcm.org
> > > http://mitgcm.org/mailman/listinfo/mitgcm-devel
> > >
> >
> >
> >
> > --
> >
> > Daniel Goldberg, PhD
> > Lecturer in Glaciology
> > School of Geosciences, University of Edinburgh
> > Geography Building, Drummond Street, Edinburgh EH8 9XP
> >
> >
> > em: D <dgoldber at mit.edu>an.Goldberg at ed.ac.uk
> > web: http://ocean.mit.edu/~dgoldberg
>
> > _______________________________________________
> > MITgcm-devel mailing list
> > MITgcm-devel at mitgcm.org
> > http://mitgcm.org/mailman/listinfo/mitgcm-devel
>
>
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel
>



-- 

Daniel Goldberg, PhD
Lecturer in Glaciology
School of Geosciences, University of Edinburgh
Geography Building, Drummond Street, Edinburgh EH8 9XP


em: D <dgoldber at mit.edu>an.Goldberg at ed.ac.uk
web: http://ocean.mit.edu/~dgoldberg
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mitgcm.org/pipermail/mitgcm-devel/attachments/20140906/9021e3cd/attachment.htm>