OS-Aware Branch Prediction: Improving Microprocessor Control Flow Prediction for Operating Systems
IEEE Transactions on Computers
CGO '11 Proceedings of the 9th Annual IEEE/ACM International Symposium on Code Generation and Optimization
Hi-index | 0.00 |
Embedded processors like Intel's XScale use dynamic branchprediction to improve performance. Due to the presence of contextswitches, the accuracy of these predictors is reduced because theyend up storing prediction histories for several processes. Thispaper shows that the loss in accuracy can be significant anddepends on predictor type and size. Several new schemes areproposed to save and restore the predictor state on contextswitches in order to improve prediction accuracy. The schemesdiffer in the amount of information they save and vary in theiraccuracy improvement. It is shown that even for a small 128 entryskew predictor, 2 - 6% improvement in prediction rate can beachieved (for an average context interval of 100K instructions) fordifferent embedded applications while saving and restoring aminimal amount of state information (less than 32bits) on a contextswitch.