dolfin team mailing list archive

["Garth N. Wells" <gnw20@xxxxxxxxx>]

Jed Brown wrote:
> Johan Hake wrote:
> > On Tuesday 22 September 2009 18:56:48 Garth N. Wells wrote:
> > > I've compiled DOLFIN again MPICH2 instead of OpenMPI. It doesn't lead to
> > > a flood of MPI-related Valgrind errors, so I would suggest that after
> > > the release that DOLFIN and PETSc be built against MPICH2 on one of the
> > > buildbots and that bot can perform the memory tests.
> > Sounds good!
> >
> > We should probably also run some demos in parallel, to test that there are no 
> > leaks in the parallel stuff too.
>
> For MPICH2, you may need to compile with --enable-g=meminit to avoid
> spurious warnings in parallel.  Open MPI has memchecker which can detect
> certain kinds of MPI misuse.
>
>   http://www.open-mpi.org/faq/?category=debugging#memchecker_errors
>
> IIRC, you shouldn't see the spurious warnings when Open MPI is built
> with memchecker (they have technical/performance reasons not to
> eliminate them otherwise).

I've tried. We still see them. The OpenMPI memchecker options are 
hopeless. There are a bunch of threads on this but not resolutions.

Garth


> Jed
>
>
>
> ------------------------------------------------------------------------
>
> _______________________________________________
> DOLFIN-dev mailing list
> DOLFIN-dev@xxxxxxxxxx
> http://www.fenics.org/mailman/listinfo/dolfin-dev