AMBER Archive (2003)

Subject: RE: Amber7 and mpirun on Linux

From: Scott Brozell (sbrozell_at_scripps.edu)
Date: Fri Apr 11 2003 - 14:40:38 CDT

Hello,

Elaborating on yesterday's response,
one can examine src/sander/_mdfil.f for the p4pg flag skipping logic.
Briefly, the build process creates _mdfil.f from mdfil.f using
the C preprocessor; if MPI is not defined (absence of -DMPI in the
MACHINE file) then the MPI ifdef block in mdfil.f will not exist
in _mdfil.f; the compiler input is _mdfil.f.

Regarding MACHINE files for MPI, one should heed the notification
on page 11 of the manual. Depending on one's platform, one may need
to modify the MACHINE file to include the MPI include path in the
environment variable MACHINEFLAGS and to include the MPI library(ies)
path and name(s) in the environment variable LOADLIB. For example,
to change Machine.pgf90_mpi for explicit use of MPICH
from
setenv MACHINEFLAGS "-DMPI -DRLE -DROWAT -DMEM_ALLOC"
to
setenv MACHINEFLAGS "-DMPI -DRLE -DROWAT -DMEM_ALLOC -I/somepath/mpich-1.2.4/include"
and
from
setenv LOADLIB "-lm "
to
setenv LOADLIB "-L/samepath/mpich-1.2.4/lib -lmpich -lm "

Do a make clean before the make install.

The Amber Reflector Archive has search capabilities.
Most of the above has been stated before:
http://amber.ch.ic.ac.uk/archive/200106/0064.html

Scott

ps this will become a FAQ.

On Fri, 11 Apr 2003, Francesc Corcho wrote:

>
> Hello,
>
> We have also find the same problem when running sander with mpirun. We
> have installed AMBER 7 on a dual system with AMD processors running SUSE
> 8.1.
>
> > > unknown flag:
> > >-p4pg
>
> As it seems that the p4pg is skipped with mdfil.f, which other problem
> could be avoiding the execution of sander in parallel.
>
> Thank you in advance for your help.
>
> Francesc Corcho
>