[gmx-developers] lots of new SIMD code but very poor jenkins coverage!

Erik Lindahl erik.lindahl at gmail.com
Wed Dec 30 15:01:36 CET 2015


I would actually argue the SIMD implementations themselves are a whole lot
less dangerous than other stuff since we've tested each of them manually on
virtually all compiler/hardware combinations where they are likely be used.
It's not like you'll have problems with BlueGene/Q SIMD in the Microsoft
Visual Studio compilers.

As people start to use the SIMD interface in other parts of the code it
will be more important to test things so we properly check the properties
of each implementation (for instance if unaligned loads are available), but
right now I'm not particularly worried about it. As I'll be working on
NBNXN kernels in january I expect to catch any problems there.


Having said that: Of course it's fine to add more coverage, but that will
require people to step up and volunteer to get it done, rather than hoping
Mark will do everything ;-)

Cheers,

Erik

On Wed, Dec 30, 2015 at 2:45 PM, Szilárd Páll <pall.szilard at gmail.com>
wrote:

> Hi,
>
> It seems to be that loads of new SIMD code is getting merged, but our
> current jenkins setup is very stripped down (to put it mildly). The current
> jenkins setup essentially tests nothing else but AVX2_256 and only with
> FFTPACK and without GPUs. Because of the default is explicitly set to
> GMX_SIMD=None unless specified otherwise, the rest of the tests all run
> with SIMD off.
>
> Am I missing something? This seems very dangerous. Could we not add a few
> more configs to get better SIMD coverage? SSE4.1, AVX, KNC, AVX_12_FMA is
> readily available and a 32-bit ARM host could be added easily (though
> perhaps not for verification but rather as post-submit trigger which AFAIK
> works now).
>
> Cheers,
> --
> Szilárd
>
> --
> 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.
>



-- 
--
Erik Lindahl <erik.lindahl at gmail.com>
Professor of Biophysics, Dept. Biochemistry & Biophysics, Stockholm
University
Professor of Theoretical biophysics, Dept. Theoretical Physics, Royal Inst.
Technology
Science for Life Laboratory, Box 1031, 17121 Solna, Sweden
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-developers/attachments/20151230/c961d4ce/attachment.html>


More information about the gromacs.org_gmx-developers mailing list