dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Fri, May 11, 2012 at 09:24:12AM +0100, Garth N. Wells wrote:
> On 11 May 2012 09:11, Anders Logg <logg@xxxxxxxxx> wrote:
> > On Fri, May 11, 2012 at 09:08:47AM +0100, Garth N. Wells wrote:
> >> On 11 May 2012 08:44, Anders Logg <logg@xxxxxxxxx> wrote:
> >> > Any implications on the speed of iteration?
> >> >
> >>
> >> Waiting on the benchmarks to run. When do they launch?
> >
> > I think 3am.
> >
> >> If anything, it should be faster since the FooIterators are now just
> >> typedefs for MeshEntityIteratorsBase<Foo>, so one layer of function
> >> calls has been removed.
> >
> > Let's hope so. There have been a few regressions lately, mostly as a
> > result of more complex index mapping:
>
> What was the change in index mapping?

My mistake. I saw a variable I didn't recognize and assumed it was
added when we first added support for parallel assembly/solve in
0.9.3.

I looked more closely at it now and the only differences between 0.9.2
and 0.9.3 that should have any effect on this are:

- Moving the implementation of the constructor for MeshEntity to .cpp
  and adding some more checks there

- Changing the MeshConnectivity variable in MeshTopology from ** to ***

I'll run a few tests and see if I can spot the regression.

--
Anders