Implementação em OpenCL e Análise Comparativa de Três Algoritmos de Interseção Raio-Triângulo em FPGA

Abstract

The calculation of ray-triangle intersections represents nearly 85% of the compu tational cost of rendering methods based in ray tracing. This monograph introduces a hardware implementation and a comparative study of three algorithms to determine ray-triangle intersection for FPGA: M¨oller-Trumbore, Ingo Wald and Hanika algorithms. To reduce the size of the synthesized circuits, we proposed fixed point implementations of M¨oller-Trumbore and Ingo Wald algorithms, originally developed under floating point representation. We compare these implementations against Hanika’s algorithm, which uses the fixed point representation by default. To accelerate the design of the digital hardware and synthesize the kernel code in high-level, we resorted to the OpenCL framework. The results reported comprise: execution times on FPGA, CPU and GPU; comparison between rendered images and the statistics of synthesized designs.