Statistical Models for Automatic Performance Tuning

  • Richard Vuduc
  • James W. Demmel
  • Jeff Bilmes
Conference paper

DOI: 10.1007/3-540-45545-0_21

Part of the Lecture Notes in Computer Science book series (LNCS, volume 2073)
Cite this paper as:
Vuduc R., Demmel J.W., Bilmes J. (2001) Statistical Models for Automatic Performance Tuning. In: Alexandrov V.N., Dongarra J.J., Juliano B.A., Renner R.S., Tan C.J.K. (eds) Computational Science — ICCS 2001. ICCS 2001. Lecture Notes in Computer Science, vol 2073. Springer, Berlin, Heidelberg


Achieving peak performance from library subroutines usually requires extensive, machine-dependent tuning by hand. Automatic tuning systems have emerged in response, and they typically operate, at compile-time, by (1) generating a large number of possible implementations of a subroutine, and (2) selecting a fast implementation by an exhaustive, empirical search. This paper applies statistical techniques to exploit the large amount of performance data collected during the search. First, we develop a heuristic for stopping an exhaustive compile-time search early if a near-optimal implementation is found. Second, we show how to construct run-time decision rules, based on run-time inputs, for selecting from among a subset of the best implementations. We apply our methods to actual performance data collected by the PHiPAC tuning system for matrix multiply on a variety of hardware platforms.


Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.

Copyright information

© Springer-Verlag Berlin Heidelberg 2001

Authors and Affiliations

  • Richard Vuduc
    • 1
  • James W. Demmel
    • 2
  • Jeff Bilmes
    • 3
  1. 1.Computer Science DivisionUniversity of California at BerkeleyBerkeleyUSA
  2. 2.Computer Science Division and Dept. of MathematicsUniversity of California at BerkeleyBerkeleyUSA
  3. 3.Dept. of Electrical EngineeringUniversity of WashingtonSeattleUSA

Personalised recommendations