On the BMW scene, this gives roughly a 10% speedup overall with clang/gcc, and 30%
speedup with visual studio (2008). It turns out visual studio was optimizing the
existing code quite poorly compared to pretty good autovectorization by clang/gcc,
but hand written SSE code also gives a smaller speed boost there.
This code isn't enabled when using the hair minimum width feature yet, need to
make that work with the SSE code still.
Now there is a single BVH traversal code with #ifdefs for various features.
At runtime it will then select the appropriate variation to use depending if
instancing, hair or motion blur is in use.
This makes scenes without hair render a bit faster, especially after the
minimum width feature was added. It's not the most beautiful code, but we can't
use c++ templates and there were already 4 copies, adding 4 more to handle the
hair case separately would be too much.
