Re: Can't compile Sander with parallel on Linux cluster

From: David Konerding <dek_at_cgl.ucsf.edu>
Date: Fri 16 Mar 2001 09:54:10 -0800

"Ross Walker" writes:
>Hi Chomsri,
>
>What has happened is that when you have installed and configured mpich it
>has looked for fortran compilers on your system and picked up the absoft
>compiles and is using these. The Absoft compilers do not work properly with
>Amber.

What??? I don't use the Absoft compiler myself, but I very much doubt if that's
the case. AMBER is standard fortran 77 and any decent compiler should generate
code for it. The error the specific user got:

> mpif77 -c -O3 -m486 -malign-double -ffast-math -fomit-frame-pointer
> -fno-globals -ff90 -funix-intrinsics-hide _sander_.f
> /usr/absoft/bin/f77: illegal option -- 3


just means that absoft doesn't recognize the compiler option "-O3" and it should
probably be changed to be compatile with absoft's compiler flags. If there's
a Machine.absoft it will list the appropriate flags.

Linking against MPI should also not be a problem (provided all the details
are handled properly). The header files in MPI specify the exact interface
between AMBER and MPI and so an MPI compiled with one compiler should be
compatible with AMBER compiled with another compiler. That's the whole point
of standard ABIs. But if Absoft doesn't follow the standard ABI, then
they've got a problem.

Also, it's important to take care and know which mpicc, mpif77 and mpirun scripts are
being called. On newer red hat systems LAM is included and is in /usr/bin where your
PATH will pick it up. If you're trying to build with MPICH's mpicc and you don't
specify the paht to MPICH's mpicc, you'll get the LAM one and things will act very
oddly...

Dave
Received on Fri Mar 16 2001 - 09:54:10 PST
Custom Search