Statistical Models for Automatic Performance Tuning

Autor:	Jeff A. Bilmes, James Demmel, Rich Vuduc
Rok vydání:	2001
Předmět:	Computer engineering Computer science Heuristic (computer science) Subroutine Message Passing Interface Performance tuning Statistical model Decision rule computer.software_genre computer Algorithm Expert system
Zdroj:	Computational Science — ICCS 2001 ISBN: 9783540422327 International Conference on Computational Science (1)
DOI:	10.1007/3-540-45545-0_21
Popis:	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 compiletime 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.
Databáze:	OpenAIRE
Externí odkaz:	https://explore.openaire.eu/search/publication?articleId=doi_________::d63823c8e72cde737ad1c4af5b15bc32 https://doi.org/10.1007/3-540-45545-0_21 Zobrazit plný text záznamu