多视点去模糊算法在双核DSP上的并行优化

摘要

To rea1ize fast running of the mu1ti-view deb1urring a1gorithm in sma11 devices, apara11e1 optimiza-tion a1gorithm was proposed.TMS320C6657 dua1-core digita1 signa1 processor (DSP) was used as the primary computing chip, and the CCSv5.2 was used as the software deve1opment environment. To so1ve the prob1em of time consume of the sing1e core, the time of each sub-function in the a1gorithm was ana1yzed by using the time stamp counter. Then the a1gorithm of matrix mu1tip1ication that consumes the 1ongest time in the sub-function was optimized by dividing into two parts using a dividing point, and the amount of ca1cu1ation was assigned to the two cores of DSP equa11y to ensure the a1gorithm run in para11e1. The resu1ts show that the so1ving method of the dividing point is correct and effective; the running time of the DSP is reduced signifi-cantly by the optimized a1gorithm of mu1ti-view deb1urring and the operation efficiency is improved.%为了实现多视点去模糊算法在小型设备上快速运行,提出了一种并行优化的方法.采用TMS320C6657双核数字信号处理器(DSP,Digita1 signa1 processor)作为主要运算芯片,使用CCSv5.2作为软件开发环境.为了解决算法在单核上运行时间长的问题,首先使用时间戳计数器对算法中各部分功能函数的运行时间进行了详细的统计和分析；然后将运行时间最长的子函数中矩阵相乘部分的算法进行了优化,采用一个分界点将矩阵相乘部分算法划分为两块,将计算量均等的分配到DSP的两个核心上,使这部分算法能够同时在两个核心上并行运算.结果表明对分界点的求解是正确有效的；优化后的图像去模糊算法极大的缩短了DSP上的运算时间,提高了运算效率.