dolfin team mailing list archive

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

On Fri, 2006-03-31 at 16:04 +0200, Johan Jansson wrote:
> On Fri, Mar 31, 2006 at 01:50:13PM +0200, Garth N. Wells wrote:
> 
> ...
> 
> > > For a full 6x6 matrix which is assembled 1e7 times, the normalised
> > > timings were
> > > 
> > > PETSc*:                   25
> > > Boost multi_array class:  8
> > > Boost ublas::matrix:      1.7
> > > std::vector based matrix: 4.8
> > > standard array:           1
> > > 
> > 
> > Small correction:
> > Blitz++: 1.7
> > ublas:   1.95
> > 
> > ublas (http://www.boost.org/libs/numeric/ublas/doc/index.htm) looks
> > really nice and provides everything I would like to have for a dense
> > matrix. It's a template library, we don't have to worry about linking to
> > it. If we use it, we could avoid a dependency by including the header
> > files in DOLFIN somewhere, something like dolfin/contrib?
> > 
> > Garth
> > 
> 
> I evaluated ublas when we were looking at different sparse linear
> algebra packages. I also liked the design of ublas, but the sparse
> performance was quite poor. But if we only use the dense
> representation I guess it would be a great candidate.
> 

Johan, do you remember which compiler flags you used? I found that using
-DNDEBUG gives a speed-up factor of around 10 when using uBlas.

Garth 

> A dense linear algebra representation could perhaps also be used for
> the affine geometry mapping, which is currently implemented without
> abstractions. If there's no significant performance penalty that is.
> 
>   Johan
> 
> _______________________________________________
> DOLFIN-dev mailing list
> DOLFIN-dev@xxxxxxxxxx
> http://www.fenics.org/cgi-bin/mailman/listinfo/dolfin-dev