dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Fri, Jan 02, 2009 at 11:19:25AM +0000, Garth N. Wells wrote:
> 
> 
> Anders Logg wrote:
> > On Thu, Jan 01, 2009 at 10:29:25PM +0100, kent-and@xxxxxxxxx wrote:
> >>>> We could ask the maintainer of the Debian package to add C++ support in
> >>>> future packages.
> >>>>
> >>>> PETSc 3 is out now. Should we move to it after the next release?
> >>>>
> >>>>> 3. Maybe fix assemble_system. It currently seems to be broken. To
> >>>>> test, set the "symmetric" option in the DG demo:
> >>>>>
> >>>>>   dolfin_set("symmetric", True)
> >>>>>
> >>>>> Is anyone going to look at assemble_system now?
> >>> Fixed (I hope :)).
> >>>
> >>> Garth
> >>>
> >> Sorry for getting back late on this issue, I've been away without e-mail.
> >> Is it fixed now ?
> >> I can have a look at it. I can also try to break it up. The only sane
> >> thing to do is
> >> to have functions 'assembleOneCell (...), assembleOneInteriorFacet(...) (with
> >> lots of arguments), I think.
> > 
> > I'd suggest
> > 
> >   assemble_cell(...)
> >   assemble_exterior_facet()
> >   assemble_interior_facet()
> > 
> > If you suspect it will be a quick fix, just go ahead. Otherwise (if
> > you suspect it will take a while), it can wait until 0.9.1.
> >
> 
> I would wait until 0.9.1 so we can do some testing. Also, last time it 
> was looked at, Assembler::assemble_system was quite a bit faster than 
> Assembler::assemble, so it's worth looking into why. It could be that 
> going to
> 
>     assemble_cell(...)
>     assemble_exterior_facet()
>     assemble_interior_facet()
> 
> will result in a performance will drop.
> 
> Garth

ok, let's wait and do a careful review of assemble_system after
the release, including timings.

Let's try and keep the buildbots green until Monday...

-- 
Anders

Attachment: signature.asc
Description: Digital signature