I'm curious how difficult it would be to implement a massively parallel BVH construction algorithm (on GPU for example). Many of the steps of BVH construction are embarrassingly parallel, such as calculating the bounding box of each primitive and computing the cost at lower levels of the BVH, but it's a lot more difficult to reason about parallelizing partition operations due to the local and spatial dependencies involved.
I'm curious how difficult it would be to implement a massively parallel BVH construction algorithm (on GPU for example). Many of the steps of BVH construction are embarrassingly parallel, such as calculating the bounding box of each primitive and computing the cost at lower levels of the BVH, but it's a lot more difficult to reason about parallelizing partition operations due to the local and spatial dependencies involved.