SENSE-REVERSING PROFILE BASED CONFIDENCE ESTIMATOR FOR LOAD VALUE PREDICTION

This work presents a sense-reversing profile based confidence estimator (SRP CE) for load value prediction. This sense-reversing profile along with a prediction bit serves as an efficient confidence estimator for the Stride2Delta predictor, which outperforms the existing estimators. The drawback of the existing profile based confidence estimators is that they tend to saturate for some history patterns. The basic idea behind the proposed SRP CE is to overcome the saturation of history bits. This is done by making the prediction of the load value evenly spread on all the history patterns. This spread enables the decision logic of the confidence estimator to be implemented with a simpler hardware. The results show that this proposed scheme improves performance significantly.