SAH Based KD Tree Construction on Hybrid Architecture

KD tree is one of the most widely used acceleration structure for ray tracing. High quality KD tree has the ability to avoid many unnecessary ray traversal and ray-triangle intersection calculations. The best method to construct a high quality KD tree is by using Surface Area Heuristic (SAH). However, the SAH computation is expensive. In this paper, we present a parallel SAH KD tree construction algorithm. We propose a architecture for parallel constructing process which can minimize the communication among working threads, and a load balance strategy for improving the parallel speedup. In our system, GPU is employed to accelerate the time consuming part of the building process. Our algorithm can achieve 4-5x speed up on a hybrid architecture of CPU and GPU.