A Comparison And Research Of Acceleration Structures For GPU Based Ray Tracing

Recently, with the rapidly increasing processing power and the growing of the levels of the programmability of graphics processing units (GPUs), GPUs have become popular platforms for doing lots of computation-intensive jobs. More and more researcher began to introduce the GPU to do this kind of computation intensive work, to accelerate the computation processing. The ray tracing algorithm is an important method to generate photorealism pictures, in film and game industry, it use wildly. But the ray tracing algorithm is a classic computation intensive algorithm, use the original ray tracing which is based on CPU to render a picture while take a long time. So, if we can map the CPU Based ray tracing to the GPU, accelerate the rendering time, will get great benefit. As a result, the GPU based ray tracing become to be a hot field for researcher.In this paper, we will discuss how to map the CPU based ray tracing to the GPU. As the early GPU main to run graphics processing, use the GPU to do the general processing will take some changes. Current GPU have powerful programmability, we can take the GPU as a stream processor, based on stream model, using the fragment program, we implement a ray tracer on the GPU, and three different accelerating structures. We give the solution to the form of acceleration structures and implementation of them. Additional, the paper compares the three acceleration structures acceleration effect, to decide which structure fit the GPU to run. The paper is mainly research the following topics:1) Design and implement a GPU based ray tracer, to map the ray tracing to the GPU.2) We discuss the three acceleration structures GPU implement in detail, and the texture layout of the structures data. As the GPU didn't have stack structures, the paper talk about how to traverse the tree structures like the KD-tree and BVH on GPU.3) To these three structures on the GPU, we also use the same scenes to test the acceleration effect, to find out which structure fit the GPU.Otherwise, we optimize the structures on the base of early papers, as the current GPU has more powerful functions than before. And we conclude that, to the GPU, the BVH structure is the best, and the implementation and traverse is more simple.