Performance search engine driven by prior knowledge of optimization

Youngsung Kim; Pavol Černý; John Dennis

doi:10.1145/2774959.2774963

Performance search engine driven by prior knowledge of optimization

Youngsung Kim, Pavol Černý, John Dennis

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

Abstract

For scientific array-based programs, optimization for a particular target platform is a hard problem. There are many optimization techniques such as (semantics-preserving) source code transformations, compiler directives, environment variables, and compiler flags that influence performance. Moreover, the performance impact of (combinations of) these factors is unpredictable. This paper focuses on providing a platform for automatically searching through search space consisting of such optimization techniques. We provide (i) a search-space description language, which enables the user to describe optimization options to be used; (ii) search engine that enables testing the performance impact of optimization options by executing optimized programs and checking their results; and (iii) an interface for implementing various search algorithms. We evaluate our platform by using two simple search algorithms - a random search and a casetree search that heuristically learns from the already examined parts of the search space. We show that such algorithms are easily implementable in our platform, and we empirically find that the framework can be used to find useful optimized algorithms.

Original language	English
Title of host publication	ARRAY 2015 - Proceedings of the 2nd ACM SIGPLAN International Workshop on Libraries, Languages, and Compilers for Array Programming, co-located with PLDI 2015
Editors	Jan Vitek, Hidehiko Masuhara, Mary Sheeran, Laurie Hendren
Publisher	Association for Computing Machinery, Inc
Pages	25-30
Number of pages	6
ISBN (Electronic)	9781450335843
DOIs	https://doi.org/10.1145/2774959.2774963
State	Published - Jun 13 2015
Externally published	Yes
Event	2nd ACM SIGPLAN International Workshop on Libraries, Languages, and Compilers for Array Programming, ARRAY 2015 - Portland, United States Duration: Jun 13 2015 → …

Publication series

Name	ARRAY 2015 - Proceedings of the 2nd ACM SIGPLAN International Workshop on Libraries, Languages, and Compilers for Array Programming, co-located with PLDI 2015

Conference

Conference	2nd ACM SIGPLAN International Workshop on Libraries, Languages, and Compilers for Array Programming, ARRAY 2015
Country/Territory	United States
City	Portland
Period	06/13/15 → …

Funding

This work was supported in part by a gift from the Intel Corporation, through NSF Cooperative Grant NSF01 which funds the National Center for Atmospheric Research (NCAR), and through an Intel Parallel Computing Center grant from the Intel Corporation.

Keywords

Automated optimization
Code generation
Performance
Scientific computing

Access to Document

10.1145/2774959.2774963

Cite this

Kim, Y., Černý, P., & Dennis, J. (2015). Performance search engine driven by prior knowledge of optimization. In J. Vitek, H. Masuhara, M. Sheeran, & L. Hendren (Eds.), ARRAY 2015 - Proceedings of the 2nd ACM SIGPLAN International Workshop on Libraries, Languages, and Compilers for Array Programming, co-located with PLDI 2015 (pp. 25-30). Article 2774963 (ARRAY 2015 - Proceedings of the 2nd ACM SIGPLAN International Workshop on Libraries, Languages, and Compilers for Array Programming, co-located with PLDI 2015). Association for Computing Machinery, Inc. https://doi.org/10.1145/2774959.2774963

Kim, Youngsung ; Černý, Pavol ; Dennis, John. / Performance search engine driven by prior knowledge of optimization. ARRAY 2015 - Proceedings of the 2nd ACM SIGPLAN International Workshop on Libraries, Languages, and Compilers for Array Programming, co-located with PLDI 2015. editor / Jan Vitek ; Hidehiko Masuhara ; Mary Sheeran ; Laurie Hendren. Association for Computing Machinery, Inc, 2015. pp. 25-30 (ARRAY 2015 - Proceedings of the 2nd ACM SIGPLAN International Workshop on Libraries, Languages, and Compilers for Array Programming, co-located with PLDI 2015).

@inproceedings{cc8416c4b49847dfa5617aa0b12570ea,

title = "Performance search engine driven by prior knowledge of optimization",

abstract = "For scientific array-based programs, optimization for a particular target platform is a hard problem. There are many optimization techniques such as (semantics-preserving) source code transformations, compiler directives, environment variables, and compiler flags that influence performance. Moreover, the performance impact of (combinations of) these factors is unpredictable. This paper focuses on providing a platform for automatically searching through search space consisting of such optimization techniques. We provide (i) a search-space description language, which enables the user to describe optimization options to be used; (ii) search engine that enables testing the performance impact of optimization options by executing optimized programs and checking their results; and (iii) an interface for implementing various search algorithms. We evaluate our platform by using two simple search algorithms - a random search and a casetree search that heuristically learns from the already examined parts of the search space. We show that such algorithms are easily implementable in our platform, and we empirically find that the framework can be used to find useful optimized algorithms.",

keywords = "Automated optimization, Code generation, Performance, Scientific computing",

author = "Youngsung Kim and Pavol {\v C}ern{\'y} and John Dennis",

note = "Publisher Copyright: {\textcopyright} 2015 ACM.; 2nd ACM SIGPLAN International Workshop on Libraries, Languages, and Compilers for Array Programming, ARRAY 2015 ; Conference date: 13-06-2015",

year = "2015",

month = jun,

day = "13",

doi = "10.1145/2774959.2774963",

language = "English",

series = "ARRAY 2015 - Proceedings of the 2nd ACM SIGPLAN International Workshop on Libraries, Languages, and Compilers for Array Programming, co-located with PLDI 2015",

publisher = "Association for Computing Machinery, Inc",

pages = "25--30",

editor = "Jan Vitek and Hidehiko Masuhara and Mary Sheeran and Laurie Hendren",

booktitle = "ARRAY 2015 - Proceedings of the 2nd ACM SIGPLAN International Workshop on Libraries, Languages, and Compilers for Array Programming, co-located with PLDI 2015",

}

Kim, Y, Černý, P & Dennis, J 2015, Performance search engine driven by prior knowledge of optimization. in J Vitek, H Masuhara, M Sheeran & L Hendren (eds), ARRAY 2015 - Proceedings of the 2nd ACM SIGPLAN International Workshop on Libraries, Languages, and Compilers for Array Programming, co-located with PLDI 2015., 2774963, ARRAY 2015 - Proceedings of the 2nd ACM SIGPLAN International Workshop on Libraries, Languages, and Compilers for Array Programming, co-located with PLDI 2015, Association for Computing Machinery, Inc, pp. 25-30, 2nd ACM SIGPLAN International Workshop on Libraries, Languages, and Compilers for Array Programming, ARRAY 2015, Portland, United States, 06/13/15. https://doi.org/10.1145/2774959.2774963

Performance search engine driven by prior knowledge of optimization. / Kim, Youngsung; Černý, Pavol; Dennis, John.
ARRAY 2015 - Proceedings of the 2nd ACM SIGPLAN International Workshop on Libraries, Languages, and Compilers for Array Programming, co-located with PLDI 2015. ed. / Jan Vitek; Hidehiko Masuhara; Mary Sheeran; Laurie Hendren. Association for Computing Machinery, Inc, 2015. p. 25-30 2774963 (ARRAY 2015 - Proceedings of the 2nd ACM SIGPLAN International Workshop on Libraries, Languages, and Compilers for Array Programming, co-located with PLDI 2015).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Performance search engine driven by prior knowledge of optimization

AU - Kim, Youngsung

AU - Černý, Pavol

AU - Dennis, John

PY - 2015/6/13

Y1 - 2015/6/13

N2 - For scientific array-based programs, optimization for a particular target platform is a hard problem. There are many optimization techniques such as (semantics-preserving) source code transformations, compiler directives, environment variables, and compiler flags that influence performance. Moreover, the performance impact of (combinations of) these factors is unpredictable. This paper focuses on providing a platform for automatically searching through search space consisting of such optimization techniques. We provide (i) a search-space description language, which enables the user to describe optimization options to be used; (ii) search engine that enables testing the performance impact of optimization options by executing optimized programs and checking their results; and (iii) an interface for implementing various search algorithms. We evaluate our platform by using two simple search algorithms - a random search and a casetree search that heuristically learns from the already examined parts of the search space. We show that such algorithms are easily implementable in our platform, and we empirically find that the framework can be used to find useful optimized algorithms.

AB - For scientific array-based programs, optimization for a particular target platform is a hard problem. There are many optimization techniques such as (semantics-preserving) source code transformations, compiler directives, environment variables, and compiler flags that influence performance. Moreover, the performance impact of (combinations of) these factors is unpredictable. This paper focuses on providing a platform for automatically searching through search space consisting of such optimization techniques. We provide (i) a search-space description language, which enables the user to describe optimization options to be used; (ii) search engine that enables testing the performance impact of optimization options by executing optimized programs and checking their results; and (iii) an interface for implementing various search algorithms. We evaluate our platform by using two simple search algorithms - a random search and a casetree search that heuristically learns from the already examined parts of the search space. We show that such algorithms are easily implementable in our platform, and we empirically find that the framework can be used to find useful optimized algorithms.

KW - Automated optimization

KW - Code generation

KW - Performance

KW - Scientific computing

UR - http://www.scopus.com/inward/record.url?scp=84979747362&partnerID=8YFLogxK

U2 - 10.1145/2774959.2774963

DO - 10.1145/2774959.2774963

M3 - Conference contribution

AN - SCOPUS:84979747362

T3 - ARRAY 2015 - Proceedings of the 2nd ACM SIGPLAN International Workshop on Libraries, Languages, and Compilers for Array Programming, co-located with PLDI 2015

SP - 25

EP - 30

BT - ARRAY 2015 - Proceedings of the 2nd ACM SIGPLAN International Workshop on Libraries, Languages, and Compilers for Array Programming, co-located with PLDI 2015

A2 - Vitek, Jan

A2 - Masuhara, Hidehiko

A2 - Sheeran, Mary

A2 - Hendren, Laurie

PB - Association for Computing Machinery, Inc

T2 - 2nd ACM SIGPLAN International Workshop on Libraries, Languages, and Compilers for Array Programming, ARRAY 2015

Y2 - 13 June 2015

ER -

Kim Y, Černý P, Dennis J. Performance search engine driven by prior knowledge of optimization. In Vitek J, Masuhara H, Sheeran M, Hendren L, editors, ARRAY 2015 - Proceedings of the 2nd ACM SIGPLAN International Workshop on Libraries, Languages, and Compilers for Array Programming, co-located with PLDI 2015. Association for Computing Machinery, Inc. 2015. p. 25-30. 2774963. (ARRAY 2015 - Proceedings of the 2nd ACM SIGPLAN International Workshop on Libraries, Languages, and Compilers for Array Programming, co-located with PLDI 2015). doi: 10.1145/2774959.2774963

Performance search engine driven by prior knowledge of optimization

Abstract

Publication series

Conference

Funding

Keywords

Access to Document

Other files and links

Fingerprint

Cite this