摘要： 基于几何着色器的边遍历算法效率主要受色器硬件规格的影响。为此，提出一种基于CUDA的边遍历算法，使用CUDA模拟实现顶点着色器与几何着色器功能，避免硬件限制。同时针对CUDA内存高效读取的合并访问要求，提出二步骤的三角形网格顶点分块、重排序算法，优化几何着色器模拟时顶点数据的读取。实验结果证明，与现有最优算法相比，该算法可获得更高的速度提升。

关键词: 边遍历, CUDA并行算法, 着色器模拟

Abstract: The efficiency of the geometry shader based edge traversal algorithm is mainly limited by the shader’s hardware specification. In order to eliminate the hardware limitations, it uses CUDA to simulate the functions of the vertex shader and the geometry shader. According to the coalesced access demands of efficient CUDA memory gathering, this paper proposes a two step segment and re-index algorithm for triangle mesh vertices. The proposed algorithm optimizes the vertex data gathering procedure during the geometry shader simulation. It demonstrates times of speedups compared with the existed optimal algorithm.

Key words: edge traversal, CUDA parallel algorithm, shader simulation

中图分类号: 

• TP311.52