Block-level prediction for wide-issue superscalar processors

Abstract

Changes in control flow, caused primarily by conditional branches, are a prime impediment to the performance of wide-issue superscalar processors. This paper investigates a block-level prediction scheme to mitigate the effects of control flow changes caused by conditional branches. Instead of predicting the outcome of each conditional branch individually, this scheme predicts the target of a sequential block of instructions, thereby allowing the superscalar processor to go past multiple branches per cycle. This approach is evaluated using the MIPS architecture, for 8-way and 12-way superscalar processors, and an improvement in effective fetch size of approximately 15% and 25%, respectively, over identical processors that use branch prediction is observed. No appreciable difference in the prediction accuracy was observed, although block-level prediction predicted one out of four outcomes.<>