[gmx-developers] gromacs 5.1rc1 OpenCL problem with Parrinello-Rahman

[Szilárd Páll]

Hi Carlo,

Do the regressiontests pass on your build? There are two tests that
use Parrinello-Rahman, so I expect those to also fail.

As you use clang (=>no OpenMP support) I assume you are running 1 CPU
thread in the GPU run (multiple ranks per GPU is not supported), right?

If your regressiontest run fails, can you please file a bug in redmine?

Cheers,

--
Szilárd

On Tue, Jul 14, 2015 at 4:52 PM, Carlo Camilloni <carlo.camilloni at gmail.com>
wrote:

> Dear Gromacs devs
>
> I am trying to test the OpenCL implementation of the new gromacs. I am
> using a topol.tpr for a production run from gromacs 5.0.4
> and running it either using a single cpu with gmx51rc1 or with a single
> cpu and one gpu on a MacPro using OpenCL (everything without openmp)
>
> The code is compiled with clang, open-mpi and avx_256 simd, the
> compilation is straightforward!
>
> Running the system with 1 cpu works well as expected  while as soon as I
> use the GPU/OpenCL (removing the -nb cpu flag)
> it crashes after ~100 steps. I generated the same topol.tpr  without
> pressure coupling (none instead of Parrinello-Rahman) and it works well.
>
> Is there a problem between OpenCL and pressure coupling? Because I
> observed this same behaviour with other systems.
>
> Apart from this, I think you are doing a great work with Gromacs
> development!
>
> Best,
> Carlo
>
>
>
> --
> Gromacs Developers mailing list
>
> * Please search the archive at
> http://www.gromacs.org/Support/Mailing_Lists/GMX-developers_List before
> posting!
>
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>
> * For (un)subscribe requests visit
> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-developers
> or send a mail to gmx-developers-request at gromacs.org.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-developers/attachments/20150714/0e817136/attachment.html>