论文部分内容阅读
针对组竞争仅考虑访问请求序列的替换结果而没有考虑请求的访存特征这个问题,提出了基于堆栈距离频度的复杂加权法在线识别访问模式的方法以及自适应访问模式的缓存管理替换算法,基本思想是依据在线识别的访问请求模式特征自动调整其插入策略.在Simics中,对选自SPEC CPU2000/2006的18个测试程序及组合负载的实验结果表明,该算法的缺失率相对于DIP、RRIP、TADIP和PIPP都有显著降低.
Aiming at the problem that the group competition only considers the replacement result of the access request sequence without considering the requested access characteristics, this paper proposes a method for online identification of access patterns based on the stack distance frequency and the buffer management replacement algorithm of adaptive access patterns. The basic idea is to automatically adjust its insertion strategy according to the characteristics of access request patterns identified online.In the Simics, the experimental results of 18 test programs and combined loads selected from SPEC CPU2000 / 2006 show that the missing rate of the algorithm relative to DIP, RRIP, TADIP and PIPP are significantly reduced.