Fast parallel construction of stack-less complete LBVH trees with efficient bit-trail traversal for ray tracing

Abstract

This paper presents an efficient space partitioning approach for building high quality Linear Bounding Volume Hierarchy (LBVH) acceleration structures for ray tracing. This method produces more regular axis-aligned bounding boxes (AABB) into a complete binary tree. This structure is fully parallelized on GPU and the process to efficiently parallelize the construction is reviewed in detail in order to guarantee the fastest build times. We also describe the traversal algorithm that is based on a stack-less bit trail method to accelerate the frame rates of rendering of 3D models in graphics processing units (GPU). We analyze diverse performance metrics such as build times, frame rates, memory footprint and average intersections by AABB and by primitive, and we compare the results with a middle split and SAH (surface area heuristic) splitting method where we show that our structure provides a good balance between fast building times and efficient ray traversal performance. This partitioning approach improves the ray traversal efficiency of rigid objects resulting on an increase of frame rates performance of 30% SAH and 50% faster than middle split.