dolfin team mailing list archive

["Garth N. Wells" <g.n.wells@xxxxxxxxxx>]

Anders Logg wrote:
> On Sun, Dec 03, 2006 at 09:54:18PM +0100, Garth N. Wells wrote:
> > kent-and@xxxxxxxxx wrote:
> > > > > > One thing to consider is more sophisticated dof mapping. It is worth
> > > > > > considering a special class to take care of this. This would be useful
> > > > > > for meshes with mixed cell/element types, parallel assembly and
> > > > > > computing sparsity patterns.
> > > > > Yes, we could create a separate class in src/fem/ that takes care of
> > > > > this. Name suggestions: NodeMap, NodeMapping, Nodes, Dofs, DofHandler,
> > > > > ...
> > > > What about DofHandler or DofManager? I see the class handling the dof
> > > > mapping, as well as generating sparsity patterns. I'll add something,
> > > > and start with a function for computing the number of non-zeroes per
> > > > matrix row (we're overestimating this at the moment substantially which
> > > > is giving me some memory problems).
> > > >
> > > > Garth
> > > You can consider the dof-handler used in SyFi/PyCC. It is not as fast as the
> > > nodemap generated by FFC, but it is very flexible. It works for any dofs
> > > I know.
> > >
> > > It is also only about 100 lines of code, using a STL map. You can find it
> > > in the file DofT.h
> > I'm working on this right now so I'll take a look at the SyFi handler. 
> > I'm planning to offer different possibilities for the mapping, and also 
> > to allow user to provide something for special problems if they wish. 
> > The current FFC mapping is about as fast as it gets for scalar valued 
> > equations. For vector-valued equations, it's not good for sparse assembly.
> >
> > Garth
>
> My suggestion would be to let FFC compute the mapping and then reorder
> what FFC produces:
>
>   1. Iterate over cells, call FFC code and build mapping
>
>   2. Assembly using the generated mapping
>
> Is this what you have in mind?

Yes, this is the first step I had in mind. At this stage, the new class 
just calls the FFC map function and returns it. Second implementation 
step will be to generate a new mapping from the FFC mapping. The class 
has/will have a number of functions to help with sparse assembly 
(nonzeros per row, sparsity pattern).

I'd also like to get some simple renumbering in place quickly for 
different mesh partitions so I can continue with parallel assembly.

> Or will the new class know about how to distribute dofs for different
> elements?

This will be needed later.

Garth

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