Newsgroups: comp.parallel.mpi
From: Ilja Heitlager <ilja@cs.utwente.nl>
Subject: MPI_PROC_NULL & LAM
Organization: University of Twente, Subdept. SETI
Date: Fri, 10 May 1996 09:47:29 +0200
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Message-ID: <3192F491.66FE@cs.utwente.nl>

I'm currently writting a simple 2D Poisson like solver with MPI/LAM. And
I want to make extensive use of the cartesian topology stuff in MPI. But
these calls make extensive use of MPI_PROC_NULL, as defined in the MPI
standard (1.1, sec 3.11), for the NEWS exchanges between iterations.
Unfortunately this is not consistently implemented in the LAM 6.0 RTS.
Is there anybody who already encountered this problem and has fixed it ? 

Second of all, I do remember a discussion on MPI_PROC_NULL and a not
well defined behaviour concerning invalid tags and so on. Seems to me
that even MPI_PROC_NULL is not unambiguously defined.

Ilja Heitlager
ilja@cs.utwente.nl

