Get access

Analyzing and addressing false interactions during compiler optimization phase ordering

Authors

  • Michael R. Jantz,

    Corresponding author
    1. Department of Electrical Engineering and Computer Science, University of Kansas, Lawrence, KS, U.S.A.
    • Correspondence to: Michael R. Jantz, Department of Electrical Engineering and Computer Science, University of Kansas, Lawrence, KS, U.S.A.

      E-mail: mjantz@ittc.ku.edu

    Search for more papers by this author
  • Prasad A. Kulkarni

    1. Department of Electrical Engineering and Computer Science, University of Kansas, Lawrence, KS, U.S.A.
    Search for more papers by this author

  • This work extends our earlier conference submission, titled Eliminating False Phase Interactions to Reduce Optimization Phase Order Search Space, published in the ACM Conference on Compilers, Architectures and Synthesis for Embedded Systems (CASES), 2010. We extend this earlier work by conducting a more thorough investigation with additional benchmarks, validation for many of our earlier experiments with SimpleScalar cycle-accurate simulation and native runs of the latest ARM hardware, more detailed analysis of the results, and re-evaluating and updating our description and conclusions. We also perform new studies (entire Section 7) in this paper that have never been reported.

SUMMARY

Compiler optimization phase ordering is a fundamental, pervasive, and long-standing problem for optimizing compilers. This problem is caused by interacting optimization phases producing different codes when applied in different orders. Producing the best phase ordering code is very important in performance-oriented and cost-constrained domains, such as embedded systems. In this work, we analyze the causes of the phase ordering problem in our compiler, Very Portable Optimizer (VPO), and report our observations. We devise new techniques to eliminate, what we call, false phase interactions in our compiler. We find that reducing such false phase interactions significantly prunes the phase order search space. We also develop and study algorithms to find the best average performance that can be delivered by a single phase sequence over our benchmark set and discuss the challenges in resolving this important problem. Our results show that there is no single sequence in VPO that can achieve the optimal phase ordering performance across all functions. Copyright © 2013 John Wiley & Sons, Ltd.

Ancillary