基于 CUDA 的细分曲面阴影体算法

摘要

为了在虚拟现实、电脑游戏等图形应用中更快速生成和实时绘制细分曲面的阴影,提出采用CUDA架构的GPU阴影体生成算法.该算法采用基于CUDA的曲面细分算法,通过CUDA共享内存结构使表面细分过程更加高效.采用基于CUDA的阴影体算法产生阴影轮廓线以及拉伸出阴影体.通过基于CUDA的流式缩减算法对阴影体数组进行压缩.通过优化CUDA和OpenGL的互操作,将绘制过程从以往算法的3步减少为2步.该算法在具有CUDA硬件的标准PC上进行测试.实验结果表明,与之前的GPU的算法相比,该算法可以生成更复杂细分曲面的阴影体,阴影体数组占用显存空间降低到2％以下,并可获得高达4倍的绘制速度提升.%A new GPU based shadow volume generation algorithm based on CUDA structure was proposed for fast generation and real-time rendering of shadow of subdivision surfaces in computer games and virtual reality applications. The algorithm introduces CUDA-based surface subdivision algorithm. Generation of surface subdivisions can run faster by using shared memory structure. CUDA-based shadow volume algorithm was introduced to generate the shadow silhouette line and extrude the shadow volume. CUDA-based stream reduction algorithm was introduced to reduce the shadow volume array. An optimized interopera-tion between CUDA and OPENGL was introduced to simplify the rendering step of the algorithm from three steps to two steps. Implemented on a standard PC with CUDA hardware, experiments show that the algorithm can generate the shadow volume of more complex subdivision surfaces compared with former GPU-based ones. The algorithm needs smaller video memory for the shadow volume array to less than 2%, and the rendering performance can gain acceleration up to more than four times.