Research On The Acceleration Method Of Ray Tracing Algorithm On Gpu And Fpga Platform Based On Directionality

Electromagnetic simulation is an effective method to solve the problem of electromagnetic wave propagation in complex environments.As a high frequency approximation method,ray tracing algorithm has been widely used in electromagnetic simulation.However,the traditional ray-tracing algorithm inevitably has the problems of low efficiency and long time-consuming because it has to deal with a lot of line intersection calculation,so it needs to speed up the traditional ray-tracing algorithm.Generally,there are two kinds of methods to accelerate ray tracing:one is to speed up the algorithm by reducing the number of unnecessary intersections;the other is to speed up the algorithm by the hardware,such as using CPU multi-core,CUDA core on GPU or the logic operation unit of FPGA,to achieve parallel algorithm on the basis of traditional software acceleration algorithm.This paper has carried out research from these two methods.At the software acceleration level,this paper proposes an acceleration method based on ray directionality and regular terrain network.The ray is grouped by the projection direction of the ray on the horizontal plane,and the possible intersecting terrain is selected according to the ray direction of each group.The set of possible intersecting terrain and building surfaces are screened according to the ray direction of each group.The selected triangular faces are stored in the triangular face set according to the ray propagation direction and a certain regular order.The tracking process of the ray is carried out according to the method of intersection of the ray group and the triangular surface group,which greatly reduces the number of times the line surface is intersected.The simulation results show that the speed-up effect of the proposed method is 4-6 times faster than that of the KD-tree method.At the hardware acceleration level,this paper implements the parallel optimization of the acceleration algorithm on the GPU and FPGA.On the GPU platform,the parallel acceleration from the data preprocessing to the ray tracing process to the intersection judgment is realized on CUDA programming framework.The algorithm is completed by 4 kernel functions,and the memory allocation and thread of each kernel function.The allocation was discussed in detail,and the optimization effect was optimized,and the actual acceleration effect was 28 times that of the CPU platform.In the FPGA platform,a 5-level data parallel plus 14-stage pipeline parallel acceleration module is designed and implemented,and the acceleration effect reaches 10.3 times of the CPU platform.