基于流式路径追踪的实时真实感渲染技术

摘要：从图形处理器（GPU）的线程调度和内存访问两个角度出发，提出了一种基于流式路径追踪的实时真实感渲染方案。它将传统的路径追踪算法分解成多个独立的逻辑模块，使得算法的实现更加贴合GPU硬件的调度模式；并且使用数组结构体（SoA）风格的内存布局重新排列数据，使得算法在GPU中运行可以减少对动态随机存取存储器（DRAM）的访问次数，从而提升算法的运行性能。该方案对GPU友好，相比于无优化的路径追踪算法，在GPU中的运行时间降低了83%~88%。

关键词：实时真实感渲染；路径追踪；GPU友好；流式；SoA

Abstract: A streaming path tracing for real-time realistic rendering by examining the graphics processing unit (GPU) thread scheduling and memory access is presented. Traditional path tracing algorithms are decomposed into several independent logic modules, making the implementation more in line with GPU hardware's scheduling pattern. Furthermore, using a structure of arrays (SoA) style memory layout rearranges data to reduce the number of accesses to dynamic random access memory (DRAM) when algorithms run on the GPU, thereby improving performance. The proposed scheme is GPU-friendly, showing a decrease in the runtime by approximately 83% to 88% on the GPU compared to the unoptimized path tracing algorithm.

Keywords: real-time realistic rendering; path tracing; GPU-friendly; streaming; SoA