Software Investigate

Exploiting Inter-Sequence Correlations for Program Behavior Prediction

Download now Free registration required

Executive Summary

Prediction of program dynamic behaviors is fundamental to program optimizations, resource management, and architecture reconfigurations. Most existing predictors are based on locality of program behaviors, subject to some inherent limitations. In this paper, the authors revisit the design philosophy and systematically explore a second source of clues: statistical correlations between the behavior sequences of different program entities. Concentrated on loops, they examine the correlations' existence, strength, and values in enhancing the design of program behavior predictors. They create the first taxonomy of program behavior sequence patterns.

  • Format: PDF
  • Size: 2611.2 KB