A study of tree-based control flow prediction schemes

In order to fetch a large number of instructions per cycle from a sequential program, wide-issue superscalar processors have to predict the outcome of multiple branches in a cycle, and fetch instructions from non-contiguous portions of code. Past research has developed schemes that predict the outcome of multiple branches by performing a single prediction. That is, instead of predicting the outcome of every conditional branch, a tree-like subgraph of the control low graph of the executed program is considered as a single prediction unit and a path is predicted through the tree, thereby allowing the superscalar fetch mechanism to go past multiple branches per cycle. In this paper, we investigate the potential of using different extents of correlation to improve the prediction accuracy of control flow prediction. We also investigate the potential of increasing the tree depth to increase the fetch size. We measure the prediction accuracy of these schemes using the SPEC ´92 integer benchmarks and the MIPS-I instruction set