[Pkg-openmpi-maintainers] Bug#592326: Bug#592326: Failure of AZTEC test case run.

[Jeff Squyres]

On Sep 2, 2010, at 10:15 AM, Rachel Gordon wrote:

> 1).
> I run it on 4 processors and got:
> node           0 :Hello, world
> node           2 :Hello, world
> node           1 :Hello, world
> node           3 :Hello, world

Ok, good.

> 2). The AZTEC code was compiled with openmpi.
> 
> 3). Enclosed please fined the fortran routine:
>    az_tutorial_with_MPI.f
> 
> 4). The Aztec code is an open source code. Its last version was released in 2001. Now it is a part of a large package 'TRILINOS'.

I can't seem to download from http://trilinos.sandia.gov/download/trilinos-10.4.html without registering, which I don't really want to do.  :-)

> I didn't consult with AZTEC authors/maintainers about a known problem running AZTEC with Openmpi on  Debian.

It looks like the example code is passing MPI_COMM_WORLD down from Fortran to C.  I wonder if Aztec is assuming that the C handle will be an integer (which, if you've always used MPICH, works just fine).  Open MPI's C handles are not integers; they are pointers.  They must be converted via the MPI_Comm_f2c function call before the handle can be used in C.

You might want to check the Aztec source code and see if it is doing that properly, or if it is just using the handle passed from Fortran without conversion.  If so, this would be an Aztec problem -- nothing to do with Open MPI or Debian (i.e., the MPI standard says you must call MPI_Comm_f2c -- if they're not doing it, then their code is buggy).

-- 
Jeff Squyres
jsquyres at cisco.com
For corporate legal information go to:
http://www.cisco.com/web/about/doing_business/legal/cri/