Compute nonoverlapping boxes on demand. about ibamr HOT 2 CLOSED

Fixed by IBAMR/IBSAMRAI2#3 (this repo is not yet publicly available)

from ibamr.

A little more on this one, for the sake of putting my notes somewhere:

This function is unnecessarily expensive (it computes something we never use) but, for most of the stuff I profile, IBFE is a lot more expensive. However - for problems where the SAMRAI linear solves take up the majority of runtime, this function dominates total runtime (It's something like O(n_patches2), which means it is also O(n_procs2)). My valve benchmark (which contains several small elastic structures in a large domain) spent 60% of its time calling this function. I'd guess that this function made it this far into SAMRAI (it is still in SAMRAI today) because few (possibly zero outside our group!) SAMRAI users solve parabolic or elliptic problems at scale.

We knew about this back when Elijah D published his thesis but those patches never quite made it into IBAMR, and I didn't notice because I spend most of my time profiling IBFE.

from ibamr.