Transparent Gif

Department of Computer Science

University of California, Santa Barbara

Abstract

Accurate Indirect Branch Prediction

by: Karel Driesen and Urs Hï½lzle

Abstract:

Indirect branch prediction is likely to become increasingly important in thefuture because indirect branches occur more frequently in object-orientedprograms. With misprediction rates of around 25% on current processors,indirect branches can incur a significant fraction of branch mispredictionoverhead even though they remain less frequent than the more predictableconditional branches. We investigate a wide range of two-level predictorsdedicated exclusively to indirect branches. Starting with predictors that usefull-precision addresses and unlimited tables, we progressively introducehardware constraints and minimize the loss of predictor performance at eachstep. For programs from the SPECint95 suite as well as a suite of large C++applications, a two-level predictor achieves a misprediction rate of 9.8% witha 1K-entry table and 7.3% with an 8K-entry table, representing more than athreefold improvement over an ideal BTB. A hybrid predictor further reducesthe misprediction rates to 8.98% and 5.95%, respectively.

Keywords:

indirect branch prediction, Holzle, Hoelzle

Date:

December 1997

Document: 1997-19

XHTML Validation | CSS Validation
Updated 14-Nov-2005
Questions should be directed to: webmaster@cs.ucsb.edu