Profile-guided composition

  • Authors:
  • Jesper Andersson;Morgan Ericsson;Christoph Kessler;Welf Löwe

  • Affiliations:
  • Software Technology Group, MSI, Växjö University, Sweden;Software Technology Group, MSI, Växjö University, Sweden;Programming Environments Laboratory, IDA, Linköping University, Sweden;Software Technology Group, MSI, Växjö University, Sweden

  • Venue:
  • SC'08 Proceedings of the 7th international conference on Software composition
  • Year:
  • 2008

Quantified Score

Hi-index 0.00

Visualization

Abstract

We present an approach that generates context-aware, optimized libraries of algorithms and data structures. The search space contains all combinations of implementation variants of algorithms and data structures including dynamically switching and converting between them. Based on profiling, the best implementation for a certain context is precomputed at deployment time and selected at runtime. In our experiments, the profile-guided composition outperforms the individual variants in almost all cases.