An Accelerated Ray Tracing Algorithm For The Sunway Taihulight

This paper introduce the acceleration methods of the ray path tracing algorithm based on the sunway TaihuLight heterogeneous multi-core architecture,which proposes the acceleration method of ray-scene intersection and accelerats the rendering rate on all-chip share mode in the large-scale scene,and proposes the three-leve parallel path tracing acceleration structure.We make a modification plan for the problem of disappearance of shadows in multi-node parallel rendering in distributed scenes,and proposes the acceleration of the shadow rays' occlusion judgement in shadering process of path tracing algorithm in order to reduce the calculation time of path tracing process and improve the all rendering efficiency.The path tracing process mainly includes the intersection operation of ray-scene and the intersection point coloring operation,both of which occupy more than 90%of all time.Considering the characteristics of the path tracking process,we divide the algorithm into intersection and shadering parts,and make them independent but related with the other.For the sunway TaihuLight multi-core structure,we processes two acceleration methods of ray traversal the scene-tree and intersection test with the primitives,which one is based on the tree nodes stream and the other is distributed ray intersection.We analysis and compare these two methods to make it possible to make full use of the computing power of the slave core in order to improve the efficiency of ray intersection,and use the master-slave asynchronous mode of sunway TaihuLight system to conceal a large part of intersection task and get higher efficiency.For the rendering of the large scale scene,we ues the multi core-groups inside the nodes to future divide the rendering tasks in order to make the most of computing resources,and we propose the three-level parallel computing structure to improve the all rendering efficiency.Focus on the problem of the disappearance of shadows in rendering of distributed scene,we make a modified scheme for distributed scene rendering engine and move the occlusion judgment of shadow ray into slave cores in order to further improve the shadering efficiency.We make the tests for every stage in research,and analyzes and explains every test result.In the final test session,different scenes are tesed,and different test results are discussed to explain the specific reasons for the different result.