[mpich-discuss] valgrind is indicating problems in MPICH2 when using the MPE library

[Dave Goodell]

On Feb 15, 2013, at 2:49 PM CST, John Michael Alex Grime wrote:

> If I compile an MPI code using the "-llmpe" and "-lmpe" options to enable MPI tracing, running the resultant code through valgrind ( "mpiexec -np X valgrind --track-origins=yes ./program ..." ) produces several instances of the following type of message:
[…]
> Has anyone else noticed any problems with MPE using uninitialized stack variables?

It's probably a bug in MPE, unfortunately.  I don't think we have funding to work on MPE right now, though we would welcome patches if someone wants to dive in and fix the issue.

I've filed a ticket so that the issue isn't totally lost: https://trac.mpich.org/projects/mpich/ticket/1794

Thanks for letting us know,
-Dave