[Yt-dev] Speedup in ghost zones

[Matthew Turk]

Hi all,

I was trying to calculate DivV and Vorticity for some big runs and it
was taking too long, so after a bit of effort I pushed a change today
(97262a5a252b) that speeds up ghost zone calculation in my (admittedly
pathological) tests by about a factor of four.  It's now dominated by
IO and by the actual refinement process (FillRegion).  There are a few
other low-hanging optimizations I may attempt to apply.

The results are bitwise identical to the old results for all of my
tests, but if you see anything odd, please let me know.  It's not
clear to me that behavior at cell edges is correct during the
*process* of refinement, which I noticed for the first time when doing
this.  If somebody else wants to take a look at this -- Sam I know in
particular has looked at the ghost zone code in the past -- it'd be
appreciated.

-Matt