If you use a Cartesian topology, you have to follow the ordering it uses and use MPI_Cart_shift to know who the neighbors are in each direction. If the application insists on a different ordering, you could use one of the Dist_graph topologies where you explicitly specify who the neighbors are. Rajeev On Nov 25, 2013, at 4:08 PM, "Kokron, Daniel S. (GSFC-610.1)[Computer Sciences Corporation]" <[email protected]> wrote:

I am attempting to convert an existing application to use the new MPI_Neighbor_* routines for halo updates. The trouble I'm having is that the application fixes the ranks in a right-handed cartesian grid while that produced by MPI_Cart_create appears to be left-handed.

MPI_Cart_create produces a rank ordering for a 4x4 grid 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

The application expects the ranks to be ordered like 12 13 14 15 8 9 10 11 4 5 6 7 0 1 2 3

I've read the MPI-3 manual at http://www.mpi-forum.org/docs/mpi-3.0/mpi30-report.pdf

What am I missing? Can I manipulate the rank placement produced by MPI_Cart_create to suite this application?

Daniel Kokron NASA Ames (ARC-TN) SciCon group 301-286-3959 _______________________________________________ discuss mailing list [email protected] To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss