[Bf-committers] BIH progress update

[Yves Poissant]

From: Jean-Luc Peurière

> I think thin primitives are likely to break almost any accel structure.
> But when such one rise, it can be easily detected, no ?

It is not so obvious. A thin primitive may not be a problem is the scene is 
composed of all thin primitives and they are all oriented along the axises. 
It can get hairy. And detecting the troublesome ones will require more CPU 
which, again, delays render.

> Cant we cut it in 2 parts (macro-polys ?) ?

That is the strategy used with space partitioning structures such as 
kd-trees. The problem with that is that the tree can grow wildly. And 
spliting polygons introduces potential precision issues. There are other 
cheaper heuristics that may be used before spliting the triangles. And here 
again, detecting cases that require splitting delays render.

There are solutions. I just need to spend time working on that. It would not 
be possible to release an alternate or a replacing acceleration structure 
for v2.46 anyway. That would be for next release (2.50?). I still have time 
for that. As to deciding ti keep octree along BIH/BVH, That can be decided 
from the benchmark performances.

Yves