Optimizing general purpose compiler optimization

Abstract

The problem of defining optimal optimization strategies for compilers is well known and has been studied extensively over the last years. The problem arises from the fact that the sheer number of possible combinations of optimizations, their order, and their setting creates a search space which cannot adequately be searched. Although it has been shown that compiler settings can be found that outperform standard -Ox switches for a single application, it is not known how to find such settings that work well for sets of applicationsIn this paper, we introduce a statistical technique to derive a methodology which trims down the search space considerably, thereby allowing a feasible and flexible solution for defining high performance optimization strategies. We show that our technique finds a single compiler setting for a collection of programs (SPECint95) that performs better than the standard -Ox settings of <i>gcc</i>. 3.3.1.

DOI: 10.1145/1062261.1062293

7 Figures and Tables

Cite this paper

@inproceedings{Haneda2005OptimizingGP, title={Optimizing general purpose compiler optimization}, author={Masayo Haneda and Peter M. W. Knijnenburg and Harry A. G. Wijshoff}, booktitle={Conf. Computing Frontiers}, year={2005} }