dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Thu, Apr 10, 2008 at 07:53:23PM +0200, Anders Logg wrote:
> On Thu, Apr 10, 2008 at 07:07:25PM +0200, Ola Skavhaug wrote:
> > Martin Sandve Alnæs skrev den 10/04-2008 følgende:
> > > 2008/4/10, Ola Skavhaug <skavhaug@xxxxxxxxx>:
> > > > Anders Logg skrev den 10/04-2008 følgende:
> > > >
> > > > > On Thu, Apr 10, 2008 at 06:19:59PM +0200, Martin Sandve Alnæs wrote:
> > > >  > > I've been debugging some boundary condition trouble with the
> > > >  > > syfi/pydolfin combination and vector elements, and narrowed it down to
> > > >  > > the interpolate function which is used prior to plotting.
> > > >  > >
> > > >  > > The bug is in the assumed numbering of the vertex_values array as
> > > >  > > returned by ufc::finite_element::interpolate_vertex_values:
> > > >  > >
> > > >  > > void DiscreteFunction::interpolate(real* values) const
> > > >  > > {
> > > >  > > ...
> > > >  > >
> > > >  > >     // Copy values to array of vertex values
> > > >  > >     for (VertexIterator vertex(*cell); !vertex.end(); ++vertex)
> > > >  > >       for (uint i = 0; i < scratch->size; ++i)
> > > >  > >         values[i*mesh.numVertices() + vertex->index()] =
> > > >  > > vertex_values[i*num_cell_vertices + vertex.pos()];
> > > >  > >
> > > >  > >
> > > >  > > By the UFC definition this should've been:
> > > >  > >
> > > >  > > vertex_values[vertex.pos()*scratch->size + i];
> > > >  > >
> > > >  > > i.e. the vector components are collected together in memory for each
> > > >  > > vertex. This bug must be mirrored in FFC, and should be fixed both
> > > >  > > places at once.
> > > >  > >
> > > >  > > I'll let someone with write access to FFC and DOLFIN fix it.
> > > >  >
> > > >  > Then should we also change the order Viper expects to receive the
> > > >  > values?
> > > >
> > > >
> > > > Sure. You update stuff, I'll make it work with Viper later.
> > > >
> > > >
> > > >  Ola
> > > 
> > > What does VTK expect?
> >  
> > In general, it is described here:
> > www.vtk.org/pdf/file-formats.pdf 
> > 
> > Ola
> 
> Do you do anything with the values you get from u.interpolate() (which
> I assume you use) before sending the array to VTK?
> 
> Would changing the order mean you have to copy and reorder the values,
> or do you do that already?

ok, this has been fixed now. The order of values w.r.t. Viper is the
same so no need to fix that (now).

Note that if you use FFC, you must also update FFC to the latest
changeset (or get crazy plots for vectors). You must also run ffc-clean.

-- 
Anders