Proposes a statistical view of the problem of performance-tuning software automatically.
Abstract: Achieving peak performance from library subroutines usually requires extensive, machine-dependent tuning by hand. Automatic tuning systems have been developed in response which 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. In this paper, we show how statistical modeling of the performance feedback data collected during the search phase can be used in... (Update)
Similar documents (at the sentence level):
24.2%: On Statistical Models in Automatic Tuning - Vuduc, Demmel, Bilmes
(Correct)
14.4%: Statistical Models for Automatic Performance Tuning - Vuduc, Demmel, Bilmes (2001)
(Correct)
Active bibliography (related documents): More All
0.2: Automating the Modeling and Optimization of the Performance.. - Singer, Veloso (2003)
(Correct)
0.2: Cache Models for Iterative Compilation - Knijnenburg, Kisuki, Gallivan
(Correct)
0.2: Self-adapting Numerical Software for Next Generation.. - Dongarra, Eijkhout (2002)
(Correct)
Similar documents based on text: More All
0.6: The PHIPAC v1.0 Matrix-Multiply Distribution - Bilmes, Asanovic, Chin, Demmel (1998)
(Correct)
0.3: Necessary Intransitive Likelihood-Ratio Classifiers - Ji, Bilmes (2003)
(Correct)
0.3: Performance Optimizations and Bounds for Sparse.. - Vuduc, Demmel.. (2002)
(Correct)
BibTeX entry: (Update)
@inproceedings{ vuduc2000:statmodel,
author = {Richard Vuduc and James Demmel and Jeff Bilmes},
title = {Statistical Modeling of Feedback Data in an Automatic Tuning System}
,
booktitle = {Third ACM Workshop on Feedback-Directed Dynamic Optimi
zation},
location = {Monterey, CA},
month = {December},
year = {2000},
url = {citeseer.ist.psu.edu/vuduc00statistical.html} }
Citations (may not include all citations):
947
Statistical Learning Theory (context) - Vapnik - 1998 ACM
387
A set of level 3 basic linear algebra subprograms (context) - Dongarra, Croz et al. - 1990 ACM DBLP
376
The cache performance and optimizations of blocked algorithm.. (context) - Rothberg, Lam et al. - 1991
345
Basic linear algebra subprograms for Fortran usage (context) - Lawson, Hanson et al. - 1979
157
Automatically tuned linear algebra software
- Whaley, Dongarra - 1998 ACM DBLP
124
FFTW: An adaptive software architecture for the FFT
- Frigo, Johnson - 1998
123
Optimizing matrix multiply using PHiPAC: a Portable
- Bilmes, Asanovi et al. - 1997
51
An extended set of Fortran basic linear algebra subroutines (context) - Dongarra, Croz et al. - 1988
37
High-level optimization via automated statistical modeling (context) - Brewer - 1995 ACM DBLP
22
Iterative compilation in program optimization
- Kisuki, Knijnenburg et al. - 2000
13
Optimizing sparse matrix vector multiplication on SMPs
- Im, Yelick - 1999
7
An investigation of recursive FFT implementations (context) - Haentjens - 2000
6
Numerical tabulation of the distribution of Kolmogorov's sta.. (context) - Birnbaum - 1952
5
Learning to predict performance from formula modeling and tr..
- Singer, Veloso - 2000 ACM DBLP
3
Automatically tuned collective operations (context) - Vadhiyar, Fagg et al. - 2000
3
matrix-multiply distribution (context) - Bilmes, Asanovi et al. - 1998
1
Technical Report Report (context) - Jordan, logistic - 1995
Documents on the same site (http://ssli.ee.washington.edu/people/bilmes/pubs-frame.html): More
Hidden-Articulator Markov Models For Speech Recognition - Richardson, Bilmes, Diorio (2000)
(Correct)
A Novel Representation for Rhythmic Structure - Iyer, Bilmes, Wright, Wessel (1997)
(Correct)
Directed Graphical Models Of Classifier Combination.. - Bilmes, Kirchhoff (2000)
(Correct)
Online articles have much greater impact More about CiteSeer.IST Add search form to your site Submit documents Feedback
CiteSeer.IST - Copyright Penn State and NEC