基于多核计算平台和高速缓存感知的Haar小波变换算法

摘要

针对目前多核计算平台的普及性以及多核平台的强大计算能力,通过充分利用高速缓存的工作原理以及多线程程序设计的优势,提出了一种在多核平台上高速缓存优化的并行Haar小波计算算法.通过测定算法运行过程中高速缓存缺失率以及算法在不同多核计算平台上的运算性能,反映了该算法极大地降低了缺失率和减少了计算时间.如在数据规模8192条件下缺失率从95%降低到8.37%,计算时间从4.35s减至0.89s.由此证明了该计算方法具有高速且可移植的特性.%For the widespread use of the multi-core computing platform and its powerful computing ability, through the full use of the cache as well as the principle of multi-threaded programming advantage, this paper proposed a cache-aware parallel Haar wavelet algorithm on the multi-core platform. By measuring the ration of cache read missing and the performance on different multi-core computing platforms, it is proved that the proposed calculation method has high computation speed and is transplantable on different multi-core platforms.