Hi Jason,
I do not think it is worth trying to compile AMBERTools on a Cray for the
moment. On things like Kraken it is almost impossible to get a single node
so it does not really help. All that is likely to be needed is sander.MPI
and PMEMD. The issue with PMEMD is just making sure the netcdf libs have
been built.
For sander one could just try aliasing mpif90 = ftn and running the
configure script with Portland group and then make parallel. Then repeat in
pmemd editing the config.h script to use ftn in place of pgf90.
I plan to overhaul the pmemd scripts shortly to get rid of the need to edit
the config.h script and instead just use the mpi wrappers to load the
correct mpi libraries.
All the best
Ross
> -----Original Message-----
> From: amber-developers-bounces.ambermd.org [mailto:amber-developers-
> bounces.ambermd.org] On Behalf Of Jason Swails
> Sent: Friday, December 11, 2009 10:24 PM
> To: AMBER Developers Mailing List
> Subject: Re: [AMBER-Developers] configure modified for Cray XT5
>
> Hi,
>
> So I guess the question becomes should I commit the changes for
> crayxt5 or not? I'm guessing not since it does not rigorously work
> for Cray XT5 and the Portland Group Compilers (or gcc if that's what
> they have set as their developer environment). The biggest problem I
> encountered was for pnetcdf, where mpicc is hard-coded as the C
> compiler and will not work at all. I had to manually change this to
> get it to work for kraken (but it works the same if -crayxt5 is not
> set), but I'm guessing that the user that wants to compile it on
> kraken (or some other cray system) is equally savvy to be able to
> finagle the configure script to suit them. It is a simple conditional
> statement that changes the default mpicc to cc if -crayxt5 is
> specified, but again I'll leave it up to the higher-ups to decide what
> goes into configure.
>
> Thanks!
> Jason
>
> On Fri, Dec 11, 2009 at 8:33 PM, Scott Brozell
> <sbrozell.rci.rutgers.edu> wrote:
> > Hi,
> >
> > On Fri, Dec 11, 2009 at 08:13:57PM -0500, Jason Swails wrote:
> >> On Fri, Dec 11, 2009 at 8:04 PM, Ross Walker <ross.rosswalker.co.uk>
> wrote:
> >> >
> >> > For years on Kraken I have just always configured AMBER with pgf90
> and then
> >> > just told pmemd to use pgf90 and mpi. Then just edit the config.h
> file for
> >> > pmemd and replace pgf90 with ftn since ftn is just a wrapper for
> pgf90. Of
> >> > course the problem now is that we no longer support Portland group
> in the
> >> > main AMBER configure script. Thus rather than making a specific
> crayxt5
> >>
> >> But the configure script allows for gnu, intel, and pgi... has the
> pgi
> >> simply fallen out of maintenance? If we don't support it, why's it
> >> still there? (./configure pgi works just fine)
> >
> > pgi is a recent re-addition that also works for me (albeit w/
> caveats;
> > see bug reports...).
> > I think Ross just hasnt kept up with the wild ride of changes
> > made to configure.
> >
> > Scott
> >
> >> > option it would make more sense to make Portland group option for
> the
> >> > standard amber configure script that uses pgf90. That way netcdf
> would be
> >> > built with pgf90 and under pmemd it is just a renaming of pgf90 to
> ftn which
> >> > we could add as a cray option.
> >>
> >> cc, CC, and ftn can also be wrappers for gcc, g++, and gfortran as
> >> well, as long as the environment is set to those (or even pathscale
> >> compilers, but I know those have fallen out of service for amber).
> At
> >> least that's what kraken's website leads me to believe (they say it
> >> looks at your environment to determine which compilers the wrappers
> >> invoke). In any case, if an application is to be built for running
> on
> >> compute nodes, wouldn't we want netcdf to be built with the wrapper
> >> rather than the "raw" compiler? I'm simply going based on the
> >> instructions given on
> >> http://www.nics.tennessee.edu/computing-resources/kraken/compiling .
> >> Of course I'm not an expert and may be missing something
> completely...
> >>
> >> Thanks!
> >
> > _______________________________________________
> > AMBER-Developers mailing list
> > AMBER-Developers.ambermd.org
> > http://lists.ambermd.org/mailman/listinfo/amber-developers
> >
>
>
>
> --
> ---------------------------------------
> Jason M. Swails
> Quantum Theory Project,
> University of Florida
> Ph.D. Graduate Student
> 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 Sat Dec 12 2009 - 10:00:02 PST
