[gmx-developers] 4.6 Binaries and Acceleration levels

[Christoph Junghans]

2012/7/5 Szilárd Páll <szilard.pall at cbr.su.se>:
> On Thu, Jul 5, 2012 at 8:11 PM, Roland Schulz <roland at utk.edu> wrote:
>> Hi,
>>
>> with the new large number of GMX_ACCELERATION options (SSE2, SSE4.1,
>> AVX_128_FMA, AVX_256) for X86 it's going to be very difficult to make
>> efficient binary packages. This effects both us if we provide binaries
>> and the Linux distributions. Either one only has SSE2 binaries or one
>> has to create a large number of binaries and let the user choose the
>> correct one. In the first case the user might get significant lower
>> performance then would be possible.
>>
>> Has the option to allow setting GMX_ACCELERATION to more than one
>> value been discussed?  Wouldn't it be relatively easy to compile the
>> same kernel more than once with different defines and use a macro to
>> add the acceleration in the function name of the kernel (E.g.
>> nbnxn_kernel_x86_sse41_simd128)? That way one could create one binary
>> which has all accelerations and the correct one would be chosen at
>> runtime.
>>
>> Otherwise I expect that everyone using mdrun from a binary Linux
>> distribution is getting only SSE2 acceleration. Obviously I'm very
>> late with suggesting that but the new GMX_ACCELERATION options have
>> also only just been added given little time to think about how to make
>> binary distributions.
>
> Indeed, that's a *major* drawback and we haven't thought of it at all.
> Additionally, all these users will be annoyed by the run-time
> detection which will keep on telling them that mdrun is not using the
> optimal level of acceleration.
Roland has a good point, but Debian and Fedora already compile Gromacs
for different mpi version.

4 acc. x 3 (serial,openmpi & mpich2) = 12 packages!

@ Jussi: Would that still be possible?


Cheers,

Christoph


>
> --
> Szilárd
>
>> Roland
>>
>> --
>> ORNL/UT Center for Molecular Biophysics cmb.ornl.gov
>> 865-241-1537, ORNL PO BOX 2008 MS6309
>> --
>> gmx-developers mailing list
>> gmx-developers at gromacs.org
>> http://lists.gromacs.org/mailman/listinfo/gmx-developers
>> Please don't post (un)subscribe requests to the list. Use the
>> www interface or send it to gmx-developers-request at gromacs.org.
> --
> gmx-developers mailing list
> gmx-developers at gromacs.org
> http://lists.gromacs.org/mailman/listinfo/gmx-developers
> Please don't post (un)subscribe requests to the list. Use the
> www interface or send it to gmx-developers-request at gromacs.org.



-- 
Christoph Junghans
Web: http://www.compphys.de