Code review comment for lp:~jobh/dolfin/symmetric-assemble

On Thu, Feb 02, 2012 at 02:01:22PM -0000, Garth Wells wrote:
> On 2 February 2012 13:46, Joachim Haga <email address hidden> wrote:
> >>
> >> > I'm confused. I thought you said earlier that SystemAssembler was
> >> > overly complex (and hence not desireable to extend),
> >>
> >> I would like to see it simplified before being extended.
> >
> >
> > I see, ok!
> >
> >
> >> OpenMPAssembler has an outer loop over colours, and an inner loop over
> >> cells. The other assemblers have one loop over cells.
> >
> >
> > Oh, the outer loop won't be a problem, all cells can be set to the same
> > color in sequential runs.
> >
>
> Yes, but there are some subtle issues that need to be taken care off.
> Assembler uses a Mesh iterator, but we can't use the iterators in
> OpenMPAssembler, so we loop with an integer, get the cell index and
> then create a cell. We might want to integrate colouring more deeply
> in Mesh, which would make things easier.

We could either make CellIterator that takes an optional color
argument, or we could loop over integers also in the regular
assembler.

--
Anders

> >
> >> Another issue with SystemAssembler is that it is not robust in
> >> parallel with the faster bc methods.
> >>
> >
> > Noted! The only thing I'm missing now is... why, with all these problems,
> > do you still recommend SystemAssembler as the path forward? ;)
> >
>
> Symmetry!
>
> Garth
>
> > -j.
> >
>