Evaluation of Branch Predictors Using High-density-branch Programs

1
Evaluation of Branch Predictors Using
High-density-branch Programs

• Fang Pang MEng.
• Lei Zhu MEng.
• Electrical and Computer Engineering Department
• University of Alberta
• November, 2002

2
Introduction

• In this project, we are going to evaluate
  different branch predictors for
  high-density-branch programs. A C code for Binary
  (6,4) encoder and decoder is used as example
  program.
• Simplescalar tool will be used as the simulation
  tool.
• The effectiveness of different types of branch
  predictors is observed. And the effect of the
  size of predictor is observed.

3
Example Program

• The Binary (6,4) encoder takes 4-bit source words
  as inputs, reads though the lookup table stored
  in the memory, finds out the corresponding 6-bit
  code words and exports code words.
• All code words are added by random noise and sent
  to Binary (6,4) decoder.
• The Binary (6,4) decoder takes these 6-bit code
  words as inputs, reads through the lookup table,
  finds out the corresponding 4-bit source words
  and exports source words. And it treats it as an
  error if the corresponding source word cant be
  found in the lookup table. At the end of process,
  it calculates the number of errors and the error
  rate.
• This program contains high-density branches. Most
  of them are random conditon branches.

4
Project Description

• Taken branch, bimodal predictor, 2-level adaptive
  predictor and combined predictor are evaluated.
• Different size of predictors are compared.
• Multiple function units and multiple ports of
  memory are used to support wide issue.
• Different caches for data and instructions are
  used.

5

• Simulation
• Sim-Outorder will be used to evaluate performance
  of different branch predictors. Different size
  predictors will be tested and compared.

6
Goal

• Find out the best branch predictor for this
  example program.
• Find out the relationship between the size of
  predictor and its performance.

7
Question

• ???
• Thanks!