high performance computing on graphics processing units: hgpu.org

The topic of this thesis is ray tracing dynamic scenes and doing that efficiently while harnessing the massive computational power of today’s graphics cards. It is motivated by the ever increasing interest in raytracing and global illumination for creating effects in movies, but also the increased usage of 2D and 3D ray tracing in modern computer games. The thesis explores different techniques for creating hierarchical acceleration structures for ray tracing, specifically binary kd-trees, and how to create these efficiently on graphics processing units. The quality of a kd-tree is defined as the speed with which it can be used to ray trace a scene and a lot of previous research has been focused on creating kd-trees of high quality, which usually results in an increased construction time. The goal of this thesis is to explore the relationship between construction speed and tree quality for kd-trees. I will focus on dynamic scenes, where the kd-tree must be rebuild before an image is rendered, and investigate if the overall time to ray trace a scene can be increased by producing acceleration structures faster, but at a lower quality. As part of the thesis I will develop an optimized ray tracer to evaluate the quality of the kd-trees produced. The total time spent constructing and ray tracing the kd-tree will be used to estimate if there is a potential performance gain to be had by quickly producing kd-trees of lower quality.