home · mobile · calendar · defenses · 1997-1998 · 

Thesis Defense - Lindsay

Static Methods In Branch Prediction
Computer Science PhD Candidate

Recent microprocessors devote tens of thousands of transistors to the problem of performing branch prediction. This thesis is about several new classes of predictor that are profile-based.

A hybrid branch predictor is one which selects from among the predictions of multiple "component" predictors. Current hybrids have prediction selection hardware, which I eliminate by controlling selection from software. I call this a static hybrid predictor.

An important advantage of static selection is that each branch site is assigned to a single component predictor. This means that the branch site makes no demands on other components. As a result, my predictors show high capacity and low aliasing. The result is better absolute performance and better cost effectiveness than the predictors in the current literature.

Committee: Dirk Grunwald, Associate Professor (Chair)
James Martin, Associate Professor
Gary Nutt, Professor
Benjamin Zorn, Associate Professor
Andrew Pleszkun, Department of Electrical and Computer Engineering
Department of Computer Science
University of Colorado Boulder
Boulder, CO 80309-0430 USA
May 5, 2012 (14:20)