Analysis of Combined Bimodal and GShare Branch Prediction Schemes

Taking advantage of instruction level parallelism is one of the key factors in improving computer performance. The main performance inhibitor in a pipelined processor is the branch penalty. Many techniques have been developed to reduce this penalty. The technique of branch prediction was explored further. Branch predictors are crucial in today's modern, superscalar processors for achieving high performance. They allow processors to fetch and execute instructions without waiting for a branch to be resolved. As a result the performance of the processor improves. Two branch prediction schemes were implemented and both resulted in an increase in performance. Combining these predictors and providing a selector to choose the most appropriate prediction scheme was implemented. A performance increase was observed with the combined prediction scheme versus the individual schemes alone. A local and global prediction scheme was utilized, the bimodal prediction served as the local predictor and the gshare prediction served as the global predictor. When combined, the prediction is as accurate at the local prediction but also takes into account dependencies on previous branches.