Data value prefetching method based on Markov model
Xue-Jun XIE, Mingyan Yu, Yizheng Ye · Annual Conference on Computers · 2006
Data value prefetching at the beginning of the pipeline is one approach to overcoming data dependences and improving instruction-level parallelism. In this paper, a Markov model used on data value prefetching is presented. The paper investigates behavior characteristic of Load instruction executing in program, then constructs a Markov model used for data value prefetching, and designs a Markov-based prefetchor. SPEC simulation indicates the constructed Markov model is suitable for data value prefetching. Compared with Two-level value prefetchor and Syntax-based prefetchor, the load instruction average of Markov-based is increased by 9.51% and 2.02%, the accuracy is increased by 12.9% and 8.2% respectively. Furthermore, the IPC is increased by 16.7% and 7.4% than other two predictors.