The Research And Implementation Of GPU-based Ray Tracing Algorithm

With the pushing of various applications ranging games, CAD, virtual reality, film and television special effects, ray tracing as an important realistic rendering tech-nique has been well studied and developed. Modern GPUs have been proven to be suitable for accelerating ray tracing task through their general computing power. This thesis focuses on GPU based ray tracing, the main contents are realizing ray tracing on GPU and accelerating GPU ray tracer. The main work is as follows:Firstly, through a thorough study of GPU rendering pipe line and BrookGPU, a stream processing model for ray tracing is proposed. The mapping of this model onto GPU is realized and verified.Secondly, a GPU based entry point search strategy is proposed. This strategy ac-celerates the traversal stage of ray tracing on GPU by preset the entry point of a ray into the scene. Entry point search can not only save the redundant traversal steps in-troduced by KD-restart effectively but also allow setting entry point lower by skipping empty voxels. This enhances the ability that traversal stage benefits from a well de-signed KD-tree construction scheme. Short stack is used to realize the entry point search, and masking is used to support packet tracing. Therefore, the algorithm can work efficiently on the data-parallel structure of GPU. Experiments are carried out under Brook environment. The results show that in most cases, entry point search can improve the rendering speed effectively. In scenes with high occlusion, the traversal time can be saved by more than 10%.Lastly, on the basis of entry point search algorithm, a further optimization of hi-erarchical entry point search algorithm is proposed. This algorithm support ray bundle traversal, can adopt the ray coherence to save traversal steps per ray, and also alleviate the restriction on entry point search made by short stack. So it is expected to further improve the rendering speed.