Re: [AMBER-Developers] [ross.rosswalker.co.uk: Re: [AMBER] Serial Version of libFpbsa.a]

From: Jason Swails <jason.swails.gmail.com>
Date: Sun, 18 Sep 2011 05:43:39 -0400

On Sat, Sep 17, 2011 at 9:23 AM, case <case.biomaps.rutgers.edu> wrote:

> ----- Forwarded message from Ross Walker <ross.rosswalker.co.uk> -----
>
>
> Did you build AMBERTools and AMBER in serial first? You need to do this
> before building the parallel version of the code.
>
> > Error: a serial version of libFpbsa.a must be built before parallel
>
> Just a note to developers: it would be great if this requirement could be
> relaxed. I've tried a little, but failed, in tracking down why the
> parallel
> build can't just do the right thing (tm).
>

So I've made a little headway here. I've implemented a "pmemd_only" rule
that will build serial, parallel, cuda, or cuda_parallel (based on your
configure command) with *nothing* built previously (in a "git clean -f -x
-d"ed tree). So you can do something like

./configure -mpi intel/gnu

cd ../../src && make pmemd_only

and it'll build pmemd.MPI (likewise for pmemd, pmemd.cuda, and
pmemd.cuda.MPI).

However, for sander it's a bit more complicated. libFpbsa.a must be built
in serial first, and there's no convenient way (that I know of) to remove
the parallel token from the FPPFLAGS before building libFpbsa.a. If this
requirement really still exists, then I don't see a way of enabling "make
parallel" in sander before "make serial" without creating a second copy of
the pbsa Makefile that uses a new FPPFLAGS token that doesn't have the -DMPI
included. But then this is another Makefile that the PBSA folks would have
to track...

However, when I just built libFpbsa.a with the parallel FPPFLAGS and mpif90,
it built just fine and the resulting sander passed the tests. Can the PBSA
folks comment on whether this should be allowed now?

Thanks!
Jason


> ...thx...dac
>
>
> _______________________________________________
> 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. Candidate
352-392-4032
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Sun Sep 18 2011 - 03:00:03 PDT
Custom Search