Newsgroups: comp.parallel.mpi
From: lederman@b125.super.org (Steve Huss-Lederman)
Subject: Re: MPI_PROC_NULL question
Organization: Supercomputing Research Center, Bowie MD, USA
Date: Mon, 3 Jul 1995 15:08:35 GMT
Message-ID: <LEDERMAN.95Jul3110835@b125.super.org>

The question was raised about using MPI_PROC_NULL to take care of
fixed boundary conditions with MPI_Sendrecv().  This is exactly the
type of thing that the MPIF was thinking of when it added this to the
standard.  If your 2D boundary conditions are non-periodic, then you
can aviod lots of if statements by using MPI_PROC_NULL.

It was raised that a segmentation fault occurred.  The possibilities
seems to be MPI implemenation error or programming error or ?.  MPI
implementation errors go to each implementor.  Can the bug be reduced
to a few lines in a test code?  If so, can you try it on another MPI
implementation to see if it works?  For others to be likely to help,
you probably need a very simple code that shows the problem.

Steve

