一种面向非规则数据的预取优化组合策略

The Optimized Grouping Strategy for Irregular Data Prefetching

在面向非规则数据的线程预取策略中,默认开启的硬件预取会加剧共享存储资源竞争,降低整个内存系统的有效性。为解决该问题,提出一种面向非规则数据的的线程预取与硬件预取优化组合策略。该策略测试和分析了L2硬件预取配置对非规则数据的线程预取性能与访存行为影响,并在此基础上为非规则数据的线程预取实现了L2硬件预取的配置优化。实验结果和分析表明,优化后预取组合策略减少了系统访存请求,提高了预取准确率和时效性,进一步改善了预取性能。

In threaded prefetching techniques for irregular data, the default opening hardware prefetchers often aggravate the shared storage resource competition and reduce the effectiveness of the entire memory system. To solve this problem, an optimized grouping strategy of hardware prefetchers with threaded prefetching for irregular data is presented. The strategy analyzes the effect of L2 prefetchers on performance and memory behavior to threaded prefetching for irregular data, and realizes the configuration optimization of L2 prefetchers. Experimental results show that the memory accesses are reduced obviously after optimizing the configuration of L2 prefetchers, leading to the growth of prefetching accuracy and timeliness, the performance improvement of threaded prefetching for irregular data.