[Q-e-developers] Fwd: vdw-d failing test (./check-pw.x.j)

[Alin Marin Elena]

Hi,

just a thought... it smells like Openmp runs out of stack. can you try
to run it with OMP_STACKSIZE= 256M or some ridiculous big value for
stack. As far as I know the standard does not impose any minimum value
so it is a free for all.

Also my experience is reducing automatically multidimensional arrays
in openmp is not a great idea for performance. you may consider to do
it by hand, by keeping a force array for each thread.

regards,
Alin
Without Questions there are no Answers!
______________________________________________________________________
Dr. Alin Marin ELENA
http://alin.elenaworld.net/
______________________________________________________________________


On 1 November 2014 12:14, Filippo Spiga <spiga.filippo at gmail.com> wrote:
> FYI Gabriele's email bounced back, here below his message!
>
> F
>
> Begin forwarded message:
>
> From: Gabriele Sclauzero <gurlomail at gmail.com>
> Subject: Re: [Q-e-developers] vdw-d failing test (./check-pw.x.j)
> Date: October 31, 2014 at 7:13:37 PM GMT
> Cc: General discussion list for Quantum ESPRESSO developers
> <q-e-developers at qe-forge.org>
> To: spiga.filippo at gmail.com
>
> Dear Filippo,
>
>     Indeed, that commit was intended to fix a problem of OpenMP
> parallelization in the calculation of forces with Grimme’s dispersion
> (mm_dispersion.f90 module). The problem shows up with some Intel compilers,
> but it doesn’t with the GNU ones (so, it might be a bug of the Intel
> compiler, perhaps). If you look inside vdw-d.out and you find that the
> output stops just before printing out the forces, it is likely the same
> problem that I have met.
> The workaround consists in doing the reduction on an auxiliary array,
> aux(:), instead of doing it directly on force_london(:,:).
> The latter is an array with two dimensions, so I think this confuses the
> OpenMP compilation part (for Intel, at least). Anyway, the fix seems to do
> the job without sacrificing performance.
>
> Ciao,
>
> Gabriele
>
>
> --
> Mr. Filippo SPIGA, M.Sc.
> http://filippospiga.info ~ skype: filippo.spiga
>
> «Nobody will drive us out of Cantor's paradise.» ~ David Hilbert
>
> *****
> Disclaimer: "Please note this message and any attachments are CONFIDENTIAL
> and may be privileged or otherwise protected from disclosure. The contents
> are not to be disclosed to anyone other than the addressee. Unauthorized
> recipients are requested to preserve this confidentiality and to advise the
> sender immediately of any error in transmission."
>
>
>
> _______________________________________________
> Q-e-developers mailing list
> Q-e-developers at qe-forge.org
> http://qe-forge.org/mailman/listinfo/q-e-developers
>