Re: [AMBER] AMBER12 build on Linux x86_64

From: Jason Swails <jason.swails.gmail.com>
Date: Mon, 23 Apr 2012 18:27:50 -0400

True enough -- -xHost is set in the configure script and is passed to
several packages that use autotools (xblas, fftw3, and most likely NetCDF
and MTK++ as well). The only way to really avoid this is to strike -xHost
out of the configure script itself. If you use that sed command on
$AMBERHOME/AmberTools/src/configure, it will never use -xHost anywhere.
 You should probably try to replace it with some vectoring optimization if
you're really looking to milk the performance, but otherwise it's probably
not a very big deal.

After doing that, I suggest running "make clean uninstall" to get rid of
any remnant of the old compile, and then start new with a -xHost-less
configure script.

HTH,
Jason

On Mon, Apr 23, 2012 at 4:51 PM, Shaw, Sharon <sharon.shaw.hp.com> wrote:

> Hi Matthew and Jason,
> Thank you for your suggestions.
>
> I edited $AMBERHOME/config.h after configure ran,
> and then did a "make install". I have attached the gzip'd output.
> You will see that -xHost is somehow still in the compiles.
> If you can figure out where it is coming from, I would appreciate it.
> My goal is to make one build for multiple architectures or to at least
> give it a try!


> I normally do make a build script after I get it all figured out
> and use sed. I just was hoping there was an ARCH_FLAG or something
> similar where one could set the architecture code generation flags like
> -x, -ax
> for the entire application. No biggie.
>
> thanks
> Sharon
>
>
> -----Original Message-----
> From: Jason Swails [mailto:jason.swails.gmail.com]
> Sent: Monday, April 23, 2012 2:31 PM
> To: AMBER Mailing List
> Subject: Re: [AMBER] AMBER12 build on Linux x86_64
>
> Matthew's comments are accurate. I realize that -xHost is not ideal for a
> heterogeneous cluster, but the arguments against canning it were that it
> would disable vectorization completely and significantly hurt performance.
> For those that have heterogeneous clusters, I guess the idea is that they
> can figure out the vectorization flags they need.
>
> When I compile, I use sed on the config.h file that configure produces: sed
> -i "s.-xHost.Replacement Flag(s).g" config.h
>
> The only deviation from Amber 10 (which is what Matt's system was), is that
> there is only one config.h file ($AMBERHOME/config.h) that needs editing.
> (Note there may appear to be config.h files in $AMBERHOME, $AMBERHOME/src,
> and $AMBERHOME/AmberTools/src, but they are all the same file referenced
> via symlinks).
>
> HTH,
> Jason
>
> On Mon, Apr 23, 2012 at 3:16 PM, Matthew Zwier <mczwier.gmail.com> wrote:
>
> > Hi Sharon,
> >
> > I've not done this with Amber 12, but I have done it with Amber 10 and
> > 11. In those versions, it didn't seem to be possible with parameters
> > to the configure scripts, but changing both config.h files (one each
> > for Amber and AmberTools) was sufficient. I tended to do it with "sed
> > -i 's/-xHost/-xSSE3/g' config.h" or whatever (and I rolled it in with
> > a few other changes I had to make locally). Those changes had to be
> > made every time config.h was regenerated, which in the early days of
> > Amber 11 was up to four times if I recall (Amber vs. AmberTools,
> > serial vs. parallel). I wound up doing everything (configuration,
> > patching the config files, and building each toolset) in a shell
> > script to keep it all straight. Your mileage may vary slightly.
> >
> > If you're doing this for a single CPU microarchitecture or for
> > debugging reasons, that should be enough. If you're trying to get the
> > same executable to run on a small number of different CPU
> > microarchitectures, you may have better luck (i.e. greater performance
> > and/or stability) compiling separate copies of Amber for each,
> > particularly if you're targeting both Intel and AMD CPUs.
> >
> > Maybe someone else on the list can comment on what needs to change, if
> > anything, for Amber 12.
> >
> > Cheers,
> > Matt Z.
> >
> > On Mon, Apr 23, 2012 at 2:41 PM, Shaw, Sharon <sharon.shaw.hp.com>
> wrote:
> > > Hi,
> > > I would like to build Amber 12 with the Intel compilers but replace
> > "-xHost" with a different
> > > "-x" flag everywhere "-xHost" exists. Is there any easy way to do this?
> > > If not, could you please supply the filenames that I would need to
> > manually edit?
> > > It seems I am missing editing a few.
> > >
> > > thank you,
> > > Sharon
> > >
> > > Sharon Shaw
> > > Hewlett-Packard Co.
> > > High Performance Computing
> > > 972-605-4603
> > > sharon.shaw.hp.com
> > >
> > >
> > > _______________________________________________
> > > AMBER mailing list
> > > AMBER.ambermd.org
> > > http://lists.ambermd.org/mailman/listinfo/amber
> >
> > _______________________________________________
> > AMBER mailing list
> > AMBER.ambermd.org
> > http://lists.ambermd.org/mailman/listinfo/amber
> >
>
>
>
> --
> Jason M. Swails
> Quantum Theory Project,
> University of Florida
> Ph.D. Candidate
> 352-392-4032
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
>


-- 
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
352-392-4032
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Mon Apr 23 2012 - 15:30:02 PDT
Custom Search