摘要：Load instructions occasionally incur very long latencies that can significantly affect system per-formance. Load value prediction alleviates this problem by allowing the CPU to speculativelycontinue processing without having to wait for the slow memory access to complete.Current load value predictors can only correctly predict about forty to seventy percent of thefetched load values. To avoid the cycle-penalty for mispredictions in the remaining cases, confi-dence estimators are employed. They inhibit all predictions that are not likely to be correct.In this paper we present a novel confidence estimator that is based on prediction outco me histo-ries. Profiles are used to identify the high-confidence history patterns. Our confidence estimatoris able to trade off coverage for accuracy and vice-versa with great flexibility and reaches an aver-age prediction accuracy over SPECint95 of as high as 99.3%. Cycle-accurate pipeline-levelsimulations show that a simple last value predictor combined with our confidence estimator out-performs other predictors, sometimes by over 100%. Furthermore, this predictor is one of twopredictors that yield a genuine speedup for all eight SPECint95 programs
