University of Houston
Department of Computer Science
In partial fulfillment of the Requirements for the Degree of
Master of Science
Rong Zhang
will defend her thesis
Study of Branch Behavior and Effective Branch Prediction
Schemes for Multimedia Applications
Abstract
Branch prediction is important for modern microprocessors to achieve high performance. There have been numerous studies of branch prediction in the context of general purpose computing workload. However relatively few studies are devoted to branch prediction for multimedia programs. As multimedia applications are becoming an important application domain, it is of great interest to study how branch prediction should be applied in this area. The purpose of this thesis is to study the branch behavior of multimedia programs, and explore effectivebranch prediction schemes to get high prediction accuracy and low hardware cost for multimedia applications.
This thesis attempts to answer the following questions for multimedia applications: What is the typical branch behavior of multimedia programs? How effective are conventional branch prediction techniques on these programs? What is the most cost-effective branch prediction scheme for multimedia applications? Using a sophisticated ILP simulator, we studied the branch behavior of a suite of typical multimedia programs. We discovered very interesting branch behavior of these multimedia applications. Based on our experimental results, we investigated novel branch prediction schemes using both static and dynamic predictions. We demonstrate the cost-effectiveness of the new prediction schemes and study the trade off between performance and hardware cost and complexity for the new schemes.
Date: Monday, April 9, 2001
Time: 12:30 PM
Place: 550-PGH
Faculty, students, and the general public are invited.
Thesis Advisor: Dr. Willis K. King