Computational Science — ICCS 2001

Volume 2073 of the series Lecture Notes in Computer Science pp 117-126


Statistical Models for Automatic Performance Tuning

  • Richard VuducAffiliated withComputer Science Division, University of California at Berkeley
  • , James W. DemmelAffiliated withComputer Science Division and Dept. of Mathematics, University of California at Berkeley
  • , Jeff BilmesAffiliated withDept. of Electrical Engineering, University of Washington

* Final gross prices may vary according to local VAT.

Get Access


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.