dolfin team mailing list archive

[Johan Jansson <johanjan@xxxxxxxxxxxxxxxx>]

On Sat, Jun 17, 2006 at 03:42:25PM +0200, Anders Logg wrote:
> Will be interesting to see. I haven't implemented the computation of
> boundaries yet, so maybe you won't be able to run all tests.
> 
> I'll add boundaries in a day or two.
> 
> /Anders
> 
> 
> On Sat, Jun 17, 2006 at 03:18:29PM +0200, kent-and@xxxxxxxxx wrote:
> > 
> > Nice!!
> > 
> > When I come back I will run the PyCC tests. With these speedups
> > we probably need to re-evaluate the efficiency of matrix making
> > in PyCC :-)
> > 
> > I will post you the results.
> > 
> > Kent
> > 
> > > Here are some results from a simple benchmark of the new mesh library:
> > >
> > > Test                                  | Old mesh  New Mesh
> > > --------------------------------------------------------------
> > > Reading and initializing 1000 times   |  0.90 s    0.21 s
> > > Refining mesh uniformly 8 times       |  27.2 s    2.14 s
> > > Iterating over connectivity 100 times |  18.2 s    1.86 s
> > > Memory usage                          |  281 MB     43 MB
> > >
> > > Initial mesh: 8 vertices and 2 cells
> > > Final mesh:   263169 vertices and 524288 cells
> > >
> > > The speedup is around a factor 10. :-)
> > >
> > > The memory usage has also been reduced. It's now at a minimum (give or
> > > take a few bytes).
> > >
> > > /Anders
> > >
> > >
> > 
> > 
> > 
> 

The profiling for the Ko (elasticity) benchmark showed that at least
15% of the time is spent in the mesh functions. It looks like that
overhead is gone now, it will be nice to test it.

  Johan