University of Houston
Department of Computer Science
In partial fulfillment of the Requirements for the Degree of
Master of Science
Manhong Guo
will defend his thesis
Evaluation of A Hybrid Branch- Prediction Method on
Traces from SPEC Benchmarks
Abstract
A hybrid branch-prediction method was recently proposed to improve branch
prediction accuracy without consuming much more hardware resources. It
distinguishes four different instruction history patterns and uses different prediction method for each pattern. The prediction methods includes static
prediction, a new prediction method named "switch counter", and the traditional two-level prediction methods. The purpose of this thesis is to evaluate performance of this hybrid branch-prediction method.
Using a self-developed simulation tool, the branch prediction accuracy
of this hybrid method of various configurations, along with those of several traditional
methods, is measured on traces from SPEC95 benchmarks. The results show that the hybrid method is the most cost-effective for branch prediction.
Results also show that both static prediction and "switch counter" prediction
in the hybrid method contribute to the performance improvement. We also
observed that, designed to handling instructions with regular patterns,
the "switch counter" algorithm alone does not perform as good as traditional
two-level method for instructions with irregular patterns; therefore, all
the four classes in the proposed hybrid method are necessary for an optimal
prediction.
Date: Friday, April 13, 2001
Time: 2:00 PM
Place: 550-PGH
Faculty, students, and the general public are invited.
Thesis Advisor: Dr. Willis King