From: "J.Henrichs" <zdv155@zam446.zam.kfa-juelich.de>
Newsgroups: comp.parallel.mpi
Subject: Re: Problems porting MPI-programs from Linux-cluster to a Cray
    Origin 2000. Please help!
Date: Fri, 04 Jun 1999 15:07:07 +0200
Organization: Forschungszentrum Juelich
Message-Id: <3757CF7B.F5EE8B3F@zam446.zam.kfa-juelich.de>
References: <37564D39.DD11171A@sofus.hiof.no>
Mime-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
Content-Transfer-Encoding: 8bit
Xref: ukc comp.parallel.mpi:5171


Hi

Ole-Martin Mørk wrote:
> 
> I have this problem:
> I have a little MPI program written in C which runs well on 16-nodes
> Linux RH cluster, but when I try to run it on a Cray Origin 2000 with
> IRIX64 6.5 I get this problem:
> 
> MPI: MPI_COMM_WORLD rank 0 has terminated without calling MPI_Finalize()
> MPI: aborting job
> 
> I am running MPI_Finalize() on all processes and even testing with
> MPI_Finalized(int *bla) if it is actually finalized before exiting.

I do not have experience on Origins, but maybe some thoughts can help
you:
1) I think that the problem is not the call to MPI_Finalize, but that
the process with rank 0 terminates without calling it - possibly at a
very different place in your code. Is there a core dump? Or du you get
any other error message/exception? Do you use an exit somewhere (without
doing a finalize)? Check, if all processes really enter MPI_Finalize.

2) MPI_Finalized is an MPI-2 call - so I am quite surprised that you can
use it (there is only one MPI-2 implementation available (iirc) - for
Fujitsu). Maybe this is a problem - you might be calling an internal
function.

3) Try inserting a barrier before the finalize call (there should be a
barrier within the finalize, but maybe there is a bug??), and check, if
all processes arrive there.

Good luck
    Joerg

