Improving Branch Prediction by Dynamic Dataflow-based Identification of Correlated Branches from a Large Global History Renju Thomas, Manoj Franklin ECE Department University of Maryland, College Park Chris Wilkerson Desktop Platforms Group Intel

1
Improving Branch Prediction by Dynamic
Dataflow-based Identification ofCorrelated
Branches from a Large Global HistoryRenju
Thomas, Manoj FranklinECE DepartmentUniversity
of Maryland, College ParkChris WilkersonDesktop
Platforms GroupIntel Corporationchris.wilkerson_at_
intel.comJared StarkMicroprocessor
ResearchIntel Labsjared.w.stark_at_intel.com
2

• Basic Idea
• History-based predictors use a global history to
  predict a branch.
• Not all branches in the global history affects a
  branch e.x. two correlated branches separated by
  a function call.
• If use only the history of branches that
  actually affect a branch, the predictor might
  get more accurate.

3

• What is an affector ?
• A branch becomes an affector for a future branch
  if it can affect the outcome of the future branch
  by choosing whether or not certain instructions
  that directly affect the future branchs source
  operands are executed.

4
(No Transcript)
5
Affector Register File
6
(No Transcript)
7
(No Transcript)
8

• Implementation
• -SimpleScalar v3.0
• -12 benchmarks from SPEC95 and SPEC00
• -Perseptron or YAKS as base predictor

9
(No Transcript)
10
(No Transcript)