Analyzing Hash Influence on Branch Prediction Accuracy

Branch prediction is important in high-performance microprocessor design. Among branch prediction mechanisms, dynamic branch predictors are the best because they can deliver accurate branch prediction regardless of input changes or the program behavior. However in all existing dynamic branch predictors, the interference effects due to aliasing in the prediction tables are the most important causes of branch mispredictions. In this paper some popular hash functions used in different computer applications are proposed and their effects on dynamic branch predictors are analyzed. We investigate hash influence on branch prediction accuracy in both analytical and experimental ways. Our experimental results suggest that hash can be incrementally effective on branch prediction accuracy and this increase is dependent on branch behavior.

Analyzing Hash Influence on Branch Prediction Accuracy Keywords:

Branch prediction , hash , uniformity , branch behavior