Make thing go fast

[rrruko]

1 sample at 540x540 resolution took roughly 40 seconds on two computers.
Profiling indicates that the average ray is intersected against 200 boxes in a scene with 6000 triangles. I don't have enough experience to know for sure, but assuming that the BIH algorithm has a time complexity of O(log n) (which is an unrealistically optimistic best-case), I would expect about 20.

To me, this suggests a bug in the BIH implementation. The only other explanation I can think of for such bad time performance is that the intersection test itself is inefficient.

[rrruko]

Here's some data from a profile of the latest build, run for a single sample.

COST CENTRE                                no.      entries  %time %alloc   %time %alloc
mollerTrumbore                             32878  215088801   28.9    0.1    59.9   40.4
intersectBIH'                              31902   77729854    3.8    3.5    96.8   96.5
intersectBIH'.intersectsLeft               32652   55831607    0.6    0.0     6.8    2.0
intersectBIH'.intersections [leaf case]    32864   21898247    6.0    9.6    76.7   85.8
intersectBIH'.intersections [branch case]  32932   20525546    0.4    1.1     0.4    1.1
intersectsBB                               32747   55831607    5.8    2.0     6.2    2.0
intersectBIH'.left                         32635   55831607    0.9    0.8     0.9    0.8
intersectBIH'.right                        32760   55831607    0.9    1.2     0.9    1.2

mollerTrumbore is being called almost 3 times as often as intersectBIH', which doesn't look right.