dolfin team mailing list archive

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

Anders Logg wrote:
> On Wed, Jan 09, 2008 at 05:50:42PM +0000, Garth N. Wells wrote:
> > Perhaps it's worth focusing on getting DofMap (with re-working of the 
> > mapping) and SparsityPattern sorted out for parallel assembly before 
> > worrying too much about the PETSc wrappers.
>
> It's not about the design of the PETSc wrappers, it's about getting
> the things we need to initialize a PETSc MPIAIJ matrix efficiently so
> we can get speedup for the parallel assembly.

The design on the PETSc wrappers (at least the init functions) and 
speeding up assembly go hand-in-hand. We need to deliver the right 
information to the init functions. This is not readily available at the 
moment.

> The current status is that it seems to work, but we don't see any
> speedup, presumably since we don't initialize the data efficiently.
>
> > Once you have the 
> > SparsityPattern (for each process) it should give you what you need to 
> > initialise parallel matrices.
>
> What's the status of pDofMap?

The pDofMap in the sandbox is pretty ad hoc. I would focus on DofMap 
such that given a ufc::dof_map and the mesh, it can re-arrange the map.

Garth