论文部分内容阅读
编译器在静态分析方式下很难对程序的非线性规律访存操作进行正确的数据预取.但采用pro-filing技术可以得到程序运行时候的访存规律,利用这些信息可以精确地插入数据预取指令.基于strideprofiling技术,提出了新的信息收集类型strideiterative,更精确地反映程序执行时访存指令的实际行为,并结合别名分析的结果调整对同一cache行的数据预取,得到比普通数据预取更好的预取性能.安腾2上运行CPU2000的12个整型测试例子平均有8.54%的性能提升,其中mcf性能提升达到了77.87%.
Compiler in the static analysis of the program is very difficult to program the non-linear access to the correct data prefetching operations.But using pro-filing technology can get the program running when the access rules, the use of these information can be accurately inserted into the data pre Fetch instruction.A new information collection type strideiterative is proposed based on strideprofiling technology to more accurately reflect the actual behavior of the fetch instruction during program execution and to adjust the prefetching of the data in the same cache line with the result of alias analysis to obtain the data stricter than normal data Prefetch better prefetching performance. The 12 integer test cases running CPU2000 on Itanium 2 average 8.54% performance improvement, of which mcf performance improvement reached 77.87%.