论文部分内容阅读
随着存储系统的访问速度与处理器的运算速度的差距越来越显著,访存性能已成为提高计算机系统性能的瓶颈.通过对指令Cache和数据Cache失效行为的分析,提出一种预取策略——结合访存失效队列状态的预取策略.该预取策略保持了指令和数据访问的次序,有利于预取流的提取.并将指令流和数据流的预取相分离,避免相互替换.在预取发起时机的选择上,不但考虑当前总线是否空闲,而且结合访存失效队列的状态,减小对处理器正常访存请求的影响.通过流过滤机制提高预取准确性,降低预取对访存带宽的需求.结果表明,采用结合访存失效队列状态的预取策略,处理器的平均访存延时减少30%,SPEC CPU2000程序的IPC值平均提高8.3%.
As the difference between the speed of access to the storage system and the speed of the processor becomes more and more significant, the performance of the storage system has become a bottleneck in improving the performance of the computer system. By analyzing the invalidation of instruction cache and data cache, a prefetch strategy is proposed, which is a prefetch strategy combining with the state of failed access queue. The prefetch strategy maintains the order of instruction and data access, which facilitates the extraction of prefetch stream. And the instruction stream and data stream prefetch phase separation, to avoid mutual replacement. Prefetch timing of the launch of choice, not only consider the current bus is free, but also combined with the memory access failure queue status, reduce the normal memory access request processor. Through the flow filtration mechanism to improve the pre-fetching accuracy, reduce the pre-fetching to memory bandwidth requirement. The results show that with the prefetch strategy combined with the state of failed access queue, the average memory latency of the processor is reduced by 30% and the average IPC value of SPEC CPU2000 is increased by 8.3%.