syfi team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Fri, Nov 09, 2007 at 10:45:37AM +0100, Kent-Andre Mardal wrote:
> tor, 08.11.2007 kl. 23.47 +0100, skrev Anders Logg:
> > Which form is "source element vector"? Is it the linear form
> > 
> >   L = v*f*dx?
> > 
> > How is this computed by SyFi (tabulate_tensor)? It would be
> > interesting to compare with the FFC code.
> > 
> > I'm not sure what your timeframe is but maybe you should also include
> > timings for FErari-optimized FFC code (in which case you need to bug
> > me about activating the optimization again in FFC). We did a series of
> > benchmarks for comparing plain FFC code with FErari code a while back
> > and the local speedups were typically between a factor 1-4:
> > 
> >     http://www.femcenter.org/pub/preprints/phiprint-2007-04.pdf
> > 
> > /Anders
> 
> 
> Ok, I thought the speed-up was < 50%, I don't know where I got that
> from. I think you said something like that but it was probably some time
> ago. 

Yes, typically < 50% (but sometimes more) for the global assembly.
For the local computation of the element tensor, the speedups are
better.

> Anyway, the element matrix calculations only take a few prosent of the
> total time (grid + element calc. + matrix assembly). Is this consistent 
> with your timings. 

It varies. The best global speedup we got was a factor 2.8 so in that
case the local computation must have been a significant portion of the
overall time.

/Anders


> We will have a look at your preprint. 
> 
> Kent
> 
> _______________________________________________
> SyFi-dev mailing list
> SyFi-dev@xxxxxxxxxx
> http://fenics.org/mailman/listinfo/syfi-dev