A Method and Apparatus for Control-Flow Prediction Using Multiple Independent Predictors
Introduction
Modern, high-performance processors all use dynamic branch predictors to increase speculation, permitting many computations to be in flight even in the face of unknown branches. Branch predictors are essential in the performance of modern processors and are now quite complex, consisting of multiple distinct predictors that all collaborate to result in a single accurate prediction.
In future distributed architectures, it may be the case that a variable number of processors collaborate to accelerate single programs. In that case, one problem that will need to be solved is how predictions are made to keep many instructions in flight among all of the participating processors.
Benefits
- Permits a larger logical predictor, with improved prediction accuracy, where each physical predictor is smaller than the sum total and can therefore be accessed at a fast desired speed.
- Improves microprocessor performance while improving physical scalability.
- Solves the problem of how to make multiple independent predictors or processors with predictors collaborate to predict a single flow of control.
Market Potential/Applications
High-performance, low-power data processing systems
For further information please contact
University of Texas,
Austin, USA
Website : www.otc.utexas.edu
