dolfin team mailing list archive

[Niclas Jansson <njansson@xxxxxx>]

dolfin@xxxxxxxxxx (DOLFIN) writes:

> One or more new changesets pushed to the primary dolfin repository.
> A short summary of the last three changesets is included below.
>
> changeset:   6768:948a25a36e169ac227ed2e5c912ee21515dcf12c
> tag:         tip
> user:        "Garth N. Wells <gnw20@xxxxxxxxx>"
> date:        Mon Aug 17 20:18:49 2009 +0100
> files:       demo/pde/mixed-poisson/cpp/main.cpp dolfin/common/Set.h dolfin/io/VTKFile.cpp
> description:
> Awful hack to make cell-based VTK outout work in parallel.
>
> The good news is that the mixed Poisson demo runs in parallel.
>
> For some reason P0 dofs do not always belong to the process on which their cell resides. Any ideas Niclas?
>
>

The renumbering algorithm doesn't know/care about the distribution/
ownership of mesh entities. I implemented it as general as possible,
it quickly becomes rather messy when you optimize for dofs which
are related to mesh entities.

In the old parallel code I used something like

if P1 
  use node numbering
if vector P1
  use node numbering + blocking of dofs 
else
  use the general renumbering algorithm (the one implemented in DOLFIN tip)



Niclas