Ok - I'm almost done with that - I'll put up a patch soon...


On Mon, Jul 8, 2013 at 3:07 PM, Roy Stogner <roystgnr@ices.utexas.edu> wrote:

On Mon, 8 Jul 2013, Derek Gaston wrote:

It looks like you turned off all shape function caching in
fe_hermite_shape_*D.C back in November. Would you mind if I just
remove everything about caching there?

I ask because I went to go implement hermite caching for pthreads...
and then saw that it was disabled.  I'd rather not implement caching
for pthreads and then disable it ;-)

Yeah, go for it.  That caching gives a nice free win in performance
99.999...% of the time, but it gives *incorrect* results in corner
cases with a very unlikely combination of AMR and element renumbering.

Even disabling renumbering isn't a fix, and IIRC a real fix would have
been complicated and intrusive, so we might as well excise the code
that's there completely before someone naively reenables it.