A Method and Apparatus for Dependence Prediction in a Memory System

Introduction

A dependence predictor is a hardware structure that seeks to learn the dependencies among individual load and store instructions to improve the accuracy of the speculation. Present technologies employ dependence predictors, but place them inside the processing unit rather than in a memory hierarchy. Dependencies between load and store instructions are learned based on instruction addresses. This approach is suitable to conventional processors but not applicable to architectures in which load and store instructions to the same address could come from different locations.

Benefits

• Feasibility for distributed architectures
• Speeds up load processing when a dependence between load and a prior store is resolved

Market Potential/Applications

High-performance, low-power data processing systems

For further information please contact

University of Texas,
Austin, USA
Website : www.otc.utexas.edu