Using Dataflow Based Context for Accurate Branch Prediction

Abstract

Contexts formed only from the outcomes of the last several instances of a static branch instruction or that of the last several dynamic branches do not always encapsulate all of the information required for correct prediction of the branch. Complex interactions between data flow and control flow change the context in ways that result in predictability loss for a significant number of dynamic branches. For improving the prediction accuracy, we use contexts derived from the predictable portions of the data flow graph. That is, the predictability of hard-to-predict branches can be improved by taking advantage of the predictability of the easy-to-predict instructions that precede it in the data flow graph. We propose and investigate a run-time scheme for producing such an improved context from the predicted values of preceding instructions. We also propose a novel branch predictor that uses dynamic dataflow-inherited speculative context (DDISC) for prediction. Simulation results verify that the use of dataflow-based contexts yields significant reduction in branch mispredictions, ranging up to 40%. This translates to an overall branch prediction accuracy of 89% to 99.5%.