A PPM-like, tag-based branch predictor

Volume 7International audienceThis paper describes cbp1.5, the tag-based, global-history predictor derived from PPM that was rank five at the first Championship Branch Prediction competition. This predictor is a particular instance of a family of predictors which we call GPPM. We introduce GPPMideal, an ideal GPPM predictor. It is possible to derive cbp1.5 from GPPM-ideal by introducing a series of degradations corresponding to real-life constraints. We characterize cbp1.5 by quantifying the impact of each degradation on the distributed CBP traces