In this assignment, we have three papers on branch prediction.

At first, people used only "self-history" prediction. Later on, branch correlation was considered. The first paper introduced the branch correlation into dynamic branch prediction and acheived a high prediction accuracy. Entry-dimension and correlation-dimension were discussed. Also, an efficient implementation was presented.

In the second paper, T. Y. Yeh and Y. N. Patt studied different configuration of two-level adaptive branch prediction. They found that the combination of a per-address branch history table and a global pattern history table is the most effective implementation.

P.Y. Chang, Eric Hao and Y. N. Patt went further. Based on the observation that different branches have different prediction accuracies, hybrid branch predictors were proposed. Ideally, this hybrid predictors can acheive a higher prediction accuracy than any of the single-scheme predictors. One important problem in this method was the design of predictor selection mechnisms. Like branch prediction, a two-level predictor selection machanism was used in this paper. No signle configuration of the selection machanism fitted all applications. A better selection machanism may improve the prediction accuracy greatly.

Zhijie SHI