fiat team mailing list archive

[Robert C. Kirby <kirby@xxxxxxxxxxxx>]

Perhaps it would be good to make it consistent between the two, just in 
case we need it.

It strikes me that what we really want to do is optimize 
Cartesian-product vector-valued elements for FIAT and FFC eventually 
where I just wrap a scalar element (less data, fewer operations, etc).  
In FFC, making use of this will make inferring and exploiting block 
structure easier.  Also, for FErari, I can really focus on optimizing 
scalar-valued element forms, since I will be just fed in blocks.

Rob

Robert C. Kirby
Assistant Professor
Department of Computer Science
The University of Chicago
http://people.cs.uchicago.edu/~kirby
On May 11, 2005, at 8:47 PM, Anders Logg wrote:

> It doesn't really matter for the mapping of dofs other than resulting
> in different types of maps for for continuous and discontinuous
> Lagrange elements. (And yes, all components are there so it's not
> wrong, only different.)
>
> For linear Lagrange with two components, FFC generates the following
> map:
>
> dofs[0] = cell.nodeID(0)
> dofs[1] = cell.nodeID(1)
> dofs[2] = cell.nodeID(2)
> int offset = mesh.noNodes()
> dofs[3] = offset + cell.nodeID(0)
> dofs[4] = offset + cell.nodeID(1)
> dofs[5] = offset + cell.nodeID(2)
>
> And for discontinuous linear Lagrance with two components, FFC
> generates the following map:
>
> dofs[0] = 6*cell.id() + 0
> dofs[1] = 6*cell.id() + 1
> dofs[2] = 6*cell.id() + 2
> dofs[3] = 6*cell.id() + 3
> dofs[4] = 6*cell.id() + 4
> dofs[5] = 6*cell.id() + 5
>
> In the first case the dofs are blocked wrt components and in the
> second case they are interleaved.
>
> One place where it does matter is when I need to generate code for
> projections (see other post). Maybe this can be solved by adding a
> descriptor of the type of dof to DualBasis? In addition to entity ids
> associated with nodes, could you store some enum which tells me the
> type of node (for example point evaluation) and the corresponding
> coordinate. Or is that already in there?
>
> /Anders
>
> On Wed, May 11, 2005 at 06:26:25PM -0500, Robert C.Kirby wrote:
>
> > Doesn't matter -- all the nodes are internal -- did I only do one
> > component's worth on the inside?
> >
> > Rob
> >
> > Robert C. Kirby
> > Assistant Professor
> > Department of Computer Science
> > The University of Chicago
> > http://people.cs.uchicago.edu/~kirby
> > On May 11, 2005, at 5:29 PM, Anders Logg wrote:
> >
> > > Rob, could you update entity_ids for DiscontinuousVectorLagrange to
> > > the new format (same as for VectorLagrange)? Seems like num_reps is 
> > > 1.
> > >
> > > /Anders
> > >
> > > _______________________________________________
> > > FIAT-dev mailing list
> > > FIAT-dev@xxxxxxxxxx
> > > http://www.fenics.org/cgi-bin/mailman/listinfo/fiat-dev
>
> --
> Anders Logg
> Research Assistant Professor
> Toyota Technological Institute at Chicago
> http://www.tti-c.org/logg/
>
> _______________________________________________
> FIAT-dev mailing list
> FIAT-dev@xxxxxxxxxx
> http://www.fenics.org/cgi-bin/mailman/listinfo/fiat-dev