Yeah, I'm pushing up some changes that will disable the mdgx compilation
for now. I've taken the Intel-specific flags out of the makefile (had 'em
in there while I was optimizing a lot of stuff, but they didn't really do
much that I could see). Once it's all in there, I'm expecting mdgx to be
between 85% and 105% as fast as pmemd, depending on the system, cutoffs,
etc. It's been a good run this past week making everything as sharp as I
can, but I'll be glad to finish off the parallel / serial dual
implementations and let people pound on it some before the AT2.0 release.
Dave
> On Wed, Oct 26, 2011 at 10:41 PM, <dcerutti.rci.rutgers.edu> wrote:
>
>> Well, I would be off to a flying start on that prize--I pushed up some
>> mdgx commits and they're NOT AT ALL compatible with the serial build.
>> Mea
>> cupla. Once you get to a parallel build, mdgx is almost
>> compatible--there
>> was a change to PtrajMask.h (formerly ptrajmask.h) but that'll get fixed
>> in my next commit. Serial compilation has been a mess because I have a
>> lot of stuff that should be roped off with #ifdef MPI statements but am
>> just now finding out about (obviously haven't tried a serial build in
>> awhile).
>>
>
> Indeed, when trying to build on OS X 10.6 with GCC 4.4 just now, I hit a
> wall at mdgx. The first problem was that Intel compiler-specific flags
> were
> hard-coded into the COPTFLAGS definition in the mdgx Makefile. Is there a
> reason you opted against just using the COPTFLAGS defined in config.h (and
> just add to them in configure if need be)?
>
> Even after I removed those I saw the issues you were talking about, and it
> won't get very far in the serial build. Do you think we should disable
> mdgx
> by default in the configure script until these build issues get sorted
> out?
> It will probably help the Automated Build System do its job whilst you
> work
> out the kinks in your code.
>
> All the best,
> Jason
>
> --
> Jason M. Swails
> Quantum Theory Project,
> University of Florida
> Ph.D. Candidate
> 352-392-4032
> _______________________________________________
> AMBER-Developers mailing list
> AMBER-Developers.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber-developers
>
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Wed Oct 26 2011 - 21:30:02 PDT
