Merge branch 'master' into cycles_hair_bvh

Merge branch 'master' into cycles_hair_bvh

Merge branch 'master' into cycles_hair_bvh

Merge branch 'master' into cycles_hair_bvh

Cycles: Support oriented bounding boxes for spatial split BVH

Very quick implementation, which will re-calculate bounding box
for primitives several times, but that we will optimize shortly.

This is already good enough to run further tests.

Cycles: De-duplicate binning builder functions

Use NULL-=pointer check to see if it builds aligned or unaligned split.

Shouldn't give big time penalty.

Cycles: Use proper heuristic when deciding whether we can create unaligned leaf right away

Cycles; Cleanup API in the unaligned heuristic

There are cases when we've got vector of references and sometimes we
only have an array. Having duplicated functions in heuristic is a bit
annoying and since it's quite low-level anyway let's just use pure
pointers there.

Cycles: Fix aligned range boundbox used to scale binning when doing unaligned split

Merge branch 'master' into cycles_hair_bvh

Merge branch 'master' into cycles_hair_bvh

Merge branch 'master' into cycles_hair_bvh

Cycles: Implement re-fitting of curve BVH

Internally re-fitting will fall back to AABB nodes which is not fully
efficient but is fast enough to rebuild BVH.

Probably for viewport we can use curve BVH but disable unaligned
heuristic to reduce preparation time.

Cycles: Optimize shadow rays intersection

Cycles: Solve crashes and wrong hair rendering with transparent shadows

Merge branch 'master' into cycles_hair_bvh

Cycles: Initial implementation of QBVH traversal for unaligned nodes

Implements both QBVH packing and traversal on SSE2+ processors.

With a test render scene render time goes from 93sec (in master) down
to 55sec. Kind of impressive, let's hope it's not because some bug and
that we can keep such a nice speedup.

Also finished some non-SIMD binary BVH code. On a test scene got about
20% of speedup comparing to 2.77a.

Well, let's verify everything, finish some remaining TODOs and make
the branch ready for master.

Cycles: Bring child visibility check

It is slower for simple cases, but could be faster in cases like complex
ray visibility settings.

Need some investigation about how to do visiblity check without time penalty.

Merge branch 'master' into cycles_hair_bvh

Cycles: SIMD optimization of unaligned BVH traversal

Some movement towards SIMD-optimized traversal of unaligned nodes,
not fully optimal yet, but now the branch is about 4% faster on
a test scene with hairy monkey with 40K hair.

Should still be possible to optimize matrix multiplication and
maybe some parts could be re-formulated a bit to become more
friendly for vectorization.

But likely next step would be to support QBVH for hair, that should
be easier to vectorze with similar to current formulation.