Multi-Level Hashed Grid Construction Methods

Abstract

Ray tracing is an inherently parallel visualization algorithm. However to achieve good performance, at interactive frame rates, an acceleration structure to decrease the number of per ray primitive intersections is required. Grid acceleration structures have some of the fastest build times, with O(N) complexity, but traditionally achieved this at a high memory cost. Recent research has reduced the memory footprint by employing compression for one-level grids. Render time performance can be improved using multi-level grids. We describe two methods for building such multi-level grids. In the first method we employ a recursive compressed grid in which grid cells are adaptively subdivided in a variable fashion. The second method uses a finely divided compressed grid, with a lower resolution macrocell overlay to speed up traversal. We analyze the performance of these new algorithms, which enable improved render times, versus existing solutions.