Fast, parallel, and asynchronous construction of BVHs for ray tracing animated scenes

Abstract

Recent developments have produced several techniques for interactive ray tracing of dynamic scenes. In particular, bounding volume hierarchies (BVHs) are efficient acceleration structures that handle complex triangle distributions and can accommodate deformable scenes by updating (refitting) the bounding primitive without restructuring the entire tree. Unfortunately, updating only the bounding primitive can result in a degradation of the quality of the BVH, and in some scenes will result in a dramatic deterioration of rendering performance. In this paper, we present three different orthogonal techniques to avoid that deterioration: (a) quickly rebuilding the BVH using a fast, binning-based approach; (b) a parallel variant of that build to better exploit the multi-core architecture of modern CPUs; (c) asynchronously rebuilding the BVH concurrently with rendering and animation, allowing it to scale to even larger models by stretching the (parallel) BVH build over one or more frames. Our approach is particularly targeted toward future many-core architectures, and allows for flexibly allocating how many cores are used for rebuilding vs. how many are used for rendering.