Alloyed Global and Local Branch History : A Robust Solution to Wrong-History Mispredictions

The need for accurate conditional-branch prediction is well known: mispred ictions waste large numbers of cycles, inhibit out-of-order execution, and waste power on mis-sp eculated computation. Prior work on branch-predictor organization has focused mainly on how to reduce conflicts i n the branch-predictor structures, while relatively little work has explored other causes of mispred ictions. Some prior work has identified other categories of mispredictions, but this paper organizes these cat egories into a broad taxonomy of misprediction types. Using the taxonomy, this paper goes on to sho w t at other categories—especially wrong-history mispredictions —are oftenmoreimportant than conflicts. This is true even if just a very simple conflict-reduction technique is used. Wrong-history mispredictions arise because current two-level, hist ory-based predictors provide only global or only local history. Their contribution to the overall misp rediction rate is substantial because most programs have some branches that require global history and also some t hat r quire local history. If only one or the other type of history is available, many branches are therefore penal ized. For SPECint95 programs using a global-history predictor, wrong-history misprediction s account for 35–50% of the total misprediction rate. By comparison, conflicts only account for 15-20%. Hybrid predictor s are one proposed solution; they use both a global-history and a local-history component. Unfortun ately, hybrid predictors only work well with large hardware budgets, because they must subdivide the availabl e area into subcomponents. Based on these observations, this paper proposes alloying local and global history together in a twolevel branch predictor structure. This is a simple but superior techniq ue for making global and local history simultaneously available and eliminating wrong-history mispredict ions. Unlike hybrid prediction, alloying gives robust performance for branch-predictor hardware budgets ranging from very large to very small. Alloying also consistently outperforms other two-level organizatio ns. In fact, a small alloyed predictor often performs as well as a much larger global-history predictor.