可重构的光线追踪处理器架构模型及仿真

Reconfigurable RTGPU's Architecture and Simulation

虽然图形处理器发展迅速,但核心染色算法仍然沿用传统光栅化方法,对复杂光照的精确模拟效果较差。光线追踪算法模拟几何光学,能够实现较好的视觉效果,但运算量大。研究光线追踪算法的特点,对算法中各部分的计算量进行了定量分析,设计了新的流水线。提出了一种动态可重构的光线追踪图形处理器(RTGPU)架构模型,设计了将流水线映射到该架构的算法。开发了仿真平台,对流水线、硬件架构模型及映射算法进行了仿真。结果表明,在算法执行效率上,与CPU相比,RTGPU执行加速比达到了11,与GPU相比,执行效率显著提高。

Although GPU has developed rapidly, the core still uses rasterization and it is a hard problem for the complex scene. The ray tracing can simulate the geometrical optics and get the better effect, but it needs the amounts of computing power. Thus, the characteristic of the ray tracing was studied and the new pipeline was designed. A dynamic reconfigurable hardware architecture （RTGPU） was put forward mapping the pipeline to it. A simulation platform was designed. According to results of experiments, more than 11x on RTGPU could be gained achieving the better performance than GPU.