[gmx-users] GMX 4.0.5 on Altix 3700 BX2: Assembly Optimizations fixed?

[Steve Cousins]

On Thu, 17 Sep 2009, Carsten Kutzner wrote:

> Hi Steve,
>
> I made some tests on an Altix 4700 some time ago because I ran into
> the same problem. If you use the fortran loops (--enable-fortran, which
> is the default on sgi) you do not loose much performance. Here are
> my benchmark results with a 80000 atom protein+bilayer+water
> simulation on a single core:
>
> inner loops  ns/day
> C                 0.12
> fortran         0.29
> assembly    0.32
>
> This was with gromacs 3.3, where the assembly kernels still worked on
> that machine.


Hi Carsten,

Thanks for the message. Since you say --enable-fortran is the default on 
the SGI, I take it that it is already doing this even though I didn't 
specify it. Correct?  I guess it makes sense to just specify it to be 
sure.

Since there is a 10% speedup by going back to version 3.3, is there any 
reason *not* to use 3.3 instead on the Altix? I'm sure there are other bug 
fixes and features since then but I don't know much about MD so I don't 
know how important they are. I'm just getting it to run on our machine so 
the chemists can run the code.

Since the optimizations worked in 3.3, it *seems* like it might not be too 
bad to get them to work with the new version. Probably 4.0 was a major 
re-write though. Are there any plans for fixing the IA64 optimizations?

Thanks a lot.

Steve


> Carsten
>
>
> On Sep 17, 2009, at 5:19 PM, Steve Cousins wrote:
>
>>>
>>> I have installed Gromacs 4.0.5 on a 64 CPU Altix 3700 BX2 and when
>>> running mdrun it just seems to spin there. It uses 100% of one CPU
>>> but doesn't finish. I scanned through the list and found that I
>>> should try setting the NOASSEMBLYLOOPS environment variable. By
>>> doing this it runs now and the gmxtest.pl script comes out with most
>>> tests passing.
>>>
>>> Is there some way to compile gromacs to be able to use the assembly
>>> optimizations for the Itanium2 processor? I'm using the Intel 11.1
>>> compilers (icc and ifort). Am I missing much performance by not
>>> being able to take advantage of the optimizations?
>>>
>>> Thanks a lot,
>>>
>>> Steve
>>> _______________________________________________
>>> gmx-users mailing list    gmx-users at gromacs.org
>>> http://lists.gromacs.org/mailman/listinfo/gmx-users
>>> Please search the archive at http://www.gromacs.org/search before
>>> posting!
>>> Please don't post (un)subscribe requests to the list. Use the www
>>> interface or send it to gmx-users-request at gromacs.org.
>>> Can't post? Read http://www.gromacs.org/mailing_lists/users.php
>>
>>
> --
> Dr. Carsten Kutzner
> Max Planck Institute for Biophysical Chemistry
> Theoretical and Computational Biophysics
> Am Fassberg 11, 37077 Goettingen, Germany
> Tel. +49-551-2012313, Fax: +49-551-2012302
> http://www.mpibpc.mpg.de/home/grubmueller/ihp/ckutzne
>
>
>
>
>