dolfin team mailing list archive

["Martin Sandve Alnæs" <martinal@xxxxxxxxx>]

On 12/8/06, Garth N. Wells <g.n.wells@xxxxxxxxxx> wrote:
> Martin Sandve Alnæs wrote:
> > On 12/7/06, Garth N. Wells <g.n.wells@xxxxxxxxxx> wrote:
> >>
> >>
> >> Martin Sandve Alnæs wrote:
> >
> > Yes, but that's a separate issue.
> >
> > We're probably just talking past each other here.
> >
> > After looking at your code again, I think the misunderstanding is in this:
> > (I don't know PetSC, so I might be wrong here)
> >
> > You call VecCreateMPI(..., n_local, n_global, &vec), which probably
> > distributes the global vector entries in contiguous chunks to
> > processes. If you didn't renumber, I now see your "problem".
> >
> > I've used an Epetra_Map(..., array_of_local_entries, n_local) which
> > distributes the global vector entries with a completely general
> > local-to-global mapping.
>
> I see now how Epetra works. Epetra_Map creates a mapping between global
> and local indices. Do you create this, or does Epetra create it for you
> on the fly as you insert terms? I'm building this map manually.

The algorithm I posted earlier builds "set<int> index_set" from a cell
partition and a ufc::dof_map. I insert the contents of this set (the
set of all global dofs touched by the cell partition) into an integer
array which I pass to the Epetra_Map constructor. Then Epetra_Map
figures out the rest.

> When you have things running with Epetra and we have things up and
> running with PETSc, it would be nice to compare the performance of the
> two. Looks like Epetra has some nice functionality.

We should do that.

martin