[petsc-users] Regarding global indexing

[Jed Brown]

On Thu, Dec 29, 2011 at 01:27, Rahul Praghanmor <praghanmor at gmail.com>wrote:

> I successfully implement PETSc to solve on a single zone unstructured mesh
> i.e. serial implementation has done and tested.
> Now I want to implement a PETSc on a multi-zone mesh.The mesh partitioning
> in multi-zone is done using a separate tool, Metis. The multi-zone mesh is
> read in solver.I don't understand how to do the global indexing for
> multi-zone mesh to form a global matrix required for PETSc.
>

Start with a non-overlapping partition (usually what is provided by Metis).
Distribute the mesh according to this partition, then compute the local
size and use MPI_Scan() to compute the offset of your piece in the global
ordering. You can learn the global index of ghosted points by sending it
from owner to ghoster through the overlap.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20111229/c3c7331a/attachment.htm>