In order to perform localization and navigation over significant distances (up

Embree SAHOur PHR-Fast Our PHR-HQ Embree Fast-Spatial 769 ms 472 ms 977 ms 1950 ms Figure 1: The San Miguel scene rendered in the Embree path tracer [WWB * 14]. Visualization of the number of traversal steps for primary rays using BVHs from different builders (the red color corresponds to 100 traversal steps per ray). From left-to-right: Embree SAH, our PHR-Fast, our PHR-HQ, and Embree Fast-Spatial. The bottom row shows the build times. Our PHR-Fast method provides 1.6× lower build time than Embree SAH, while the PHR-HQ method has 2× lower build time than Embree Fast-Spatial. In both comparisons, the builders provide equivalent ray tracing performance. AbstractWe propose a novel algorithm for construction of bounding volume hierarchies (BVHs) for multi-core CPU architectures. The algorithm constructs the BVH by a divisive top-down approach using a progressively refined cut of an existing auxiliary BVH. We propose a new strategy for refining the cut that significantly reduces the workload of individual steps of BVH construction. Additionally, we propose a new method for integrating spatial splits into the BVH construction algorithm. The auxiliary BVH is constructed using a very fast method such as LBVH based on Morton codes. We show that the method provides a very good trade-off between the build time and ray tracing performance. We evaluated the method within the Embree ray tracing framework and show that it compares favorably with the Embree BVH builders regarding build time while maintaining comparable ray tracing speed.

For ray tracing based methods, traversing a hierarchical acceleration data structure takes up a substantial portion of the total rendering time. We propose an additional data structure which cuts off large parts of the hierarchical traversal. We use the idea of ray classification combined with the hierarchical scene representation provided by a bounding volume hierarchy. We precompute short arrays of indices to subtrees inside the hierarchy and use them to initiate the traversal for a given ray class. This arrangement is compact enough to be cache‐friendly, preventing the method from negating its traversal gains by excessive memory traffic. The method is easy to use with existing renderers which we demonstrate by integrating it to the PBRT renderer. The proposed technique reduces the number of traversal steps by 42% on average, saving around 15% of time of finding ray‐scene intersection on average.