基于混合模式缓存优化的三角形条带化

Cache-Friendly Triangle Strip Generation Based on Hybrid Model

针对已有缓存优化的三角形条带化算法不能兼具较高顶点缓存命中率和适应多种顶点缓存的问题,提出了一种基于混合模式缓存优化的三角形条带化算法.采用优化求解传输代价方程的算法,通过精确地模拟缓存状态变化来获得较理想的缓存命中率;启用后进先用(LIFU)的数据引用方式重新定义了优化求解传输代价方程,使三角形条带同时兼顾顺时针和逆时针2种增长方向,极大地提高了三角形条带内部顶点的重用性,使之在任意顶点缓存中均可有效地提高顶点缓存命中率.实验结果表明,该算法兼具较高顶点缓存命中率和适应多种顶点缓存的优点,可有效地缓解GPU的处理速度不断提高而数据访问速度严重滞后的现象.

Most of the existing cache optimization triangle generation methods have a critical problem that high cache optimization and wide applicability cannot be achieved at the same time.To solve this problem,a new algorithm that employs hybrid model to generate triangle strips is proposed in this paper.Measured by optimal cost function,given a mesh,a sequence of triangle strips which have the minimal vertex cache miss rate is generated by this algorithm.It adopts last-in-first-used（LIFU）vertices cache replacement policy.Moreover,it redefines a novel primary cost function.Measured by the cost function,the triangle strip can grow in either counter-clockwise or clockwise direction,which is beneficial to the vertices re-use among one triangle strip and further ensures low cache miss rate even in any oblivious cache.Examples show both high cache optimization and wide applicability can be achieved at the same time.It can be an effective solution to the bus traffic between graphics subsystem and memory.