Newsgroups: comp.parallel.mpi
From: Gary L. Hennigan <glhenni@cs.sandia.gov>
Reply-To: glhenni@cs.sandia.gov
Subject: Re: Simulating MPI
Organization: Sandia National Labs
Date: 11 May 1998 09:58:44 -0600
Mime-Version: 1.0 (generated by tm-edit 7.108)
Message-ID: <sa7d8dkztmz.fsf@chtorr.cs.sandia.gov>

Martial Michel <martial@nist.gov> writes:
| 	I am wondering if a library writting send/recv data into a file and
| working without having a demon should exist for simulating MPI. My
| problem is that I am trying to develop some programs using MPI but am
| willing to use a tool like purify to check memory problem, and it
| appears to me that they interfere with one another.

I use purify all the time for MPICH programs compiled on my SGI
workstation. As you can imagine it's pretty slow, since it's running
multiple copies of the purified code on a single processor, but, it's
nice to be able to find memory problems that only show up in the
parallel case.

On the SGI the sequence is something like:

mpicc -g <mycode.c> -o <myexe>.mpi
purify <myexe>.mpi
mpirun -np <n> <myexe>.mpi.pure

It's simpler on the SGI because purify doesn't have to be used during
the compilation. Otherwise you'd have to configure MPICH to use
"purify" during the compilation.

Gary Hennigan