dolfin team mailing list archive

["Robert Kirby" <robert.c.kirby@xxxxxxxxx>]

Hi all, I've heard other things about Trilinos not faring well in these
kinds of benchmarks, but being engineered for very large-scale things and
for robustness.

One feature that might be worth experimenting with is the variable block row
matrix in Trilinos.  It allows you to block rows variably (suppose you have
3 dof per vertex, 6 per edge, 3 internal, etc).  You supposedly can
customize Epetra matrices to fit this, which has performance wins in doing
the matrix-vector (which is, quite frankly, more important than assembly).
 Whether it's worth reengineering the assembly for one particular back-end,
I don't know.

Rob

On Wed, Apr 30, 2008 at 3:48 AM, Kent-Andre Mardal <kent-and@xxxxxxxxx>
wrote:

>
> I did not get to test uBlas, but on my machine the assembly process
> (mass matrix) in the attached script took 5.3s with PETSc and 6.1s with
> Epetra.
>
> Kent
>
>
> ons, 30.04.2008 kl. 09.36 +0100, skrev Garth N. Wells:
> > It looks like the necessary functions for assembly using Trilinos as a
> > back end are in place, so I was wondering if anyone has had a chance to
> > compare the assembly performance of the various linear algebra back
> ends?
> >
> > Garth
> > _______________________________________________
> > DOLFIN-dev mailing list
> > DOLFIN-dev@xxxxxxxxxx
> > http://www.fenics.org/mailman/listinfo/dolfin-dev
>
> _______________________________________________
> DOLFIN-dev mailing list
> DOLFIN-dev@xxxxxxxxxx
> http://www.fenics.org/mailman/listinfo/dolfin-dev
>
>