Chisel: Reliability-and accuracy-aware optimization of approximate computational kernels

Sasa Misailovic, Michael Carbin, Sara Achour, Zichao Qi, Martin Rinard

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

The accuracy of an approximate computation is the distance between the result that the computation produces and the corresponding fully accurate result. The reliability of the computation is the probability that it will produce an acceptably accurate result. Emerging approximate hardware platforms provide approximate operations that, in return for reduced energy consumption and/or increased performance, exhibit reduced reliability and/or accuracy.

We present Chisel, a system for reliability-and accuracyaware optimization of approximate computational kernels that run on approximate hardware platforms. Given a combined reliability and/or accuracy specification, Chisel automatically selects approximate kernel operations to synthesize an approximate computation that minimizes energy consumption while satisfying its reliability and accuracy specification.

We evaluate Chisel on five applications from the image processing, scientific computing, and financial analysis domains. The experimental results showthat our implemented optimization algorithm enables Chisel to optimize our set of benchmark kernels to obtain energy savings from 8.7% to 19.8% compared to the original (exact) kernel implementations while preserving important reliability guarantees.

Original languageEnglish (US)
Title of host publicationProceedings of the Conference on Object-Oriented Programming Systems, Languages, and Applications, OOPSLA
PublisherAssociation for Computing Machinery
Pages309-328
Number of pages20
ISBN (Electronic)9781450325851
DOIs
StatePublished - Oct 15 2014
Event2014 ACM SIGPLAN Conference on Object-Oriented Programming, Systems, Languages, and Applications, OOPSLA 2014 - Portland, United States
Duration: Oct 20 2014Oct 24 2014

Publication series

NameProceedings of the Conference on Object-Oriented Programming Systems, Languages, and Applications, OOPSLA

Other

Other2014 ACM SIGPLAN Conference on Object-Oriented Programming, Systems, Languages, and Applications, OOPSLA 2014
CountryUnited States
CityPortland
Period10/20/1410/24/14

Fingerprint

Energy utilization
Specifications
Hardware
Natural sciences computing
Energy conservation
Image processing

Keywords

  • Approximate Computing

ASJC Scopus subject areas

  • Software

Cite this

Misailovic, S., Carbin, M., Achour, S., Qi, Z., & Rinard, M. (2014). Chisel: Reliability-and accuracy-aware optimization of approximate computational kernels. In Proceedings of the Conference on Object-Oriented Programming Systems, Languages, and Applications, OOPSLA (pp. 309-328). (Proceedings of the Conference on Object-Oriented Programming Systems, Languages, and Applications, OOPSLA). Association for Computing Machinery. https://doi.org/10.1145/10.1145/2660193.2660231

Chisel : Reliability-and accuracy-aware optimization of approximate computational kernels. / Misailovic, Sasa; Carbin, Michael; Achour, Sara; Qi, Zichao; Rinard, Martin.

Proceedings of the Conference on Object-Oriented Programming Systems, Languages, and Applications, OOPSLA. Association for Computing Machinery, 2014. p. 309-328 (Proceedings of the Conference on Object-Oriented Programming Systems, Languages, and Applications, OOPSLA).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Misailovic, S, Carbin, M, Achour, S, Qi, Z & Rinard, M 2014, Chisel: Reliability-and accuracy-aware optimization of approximate computational kernels. in Proceedings of the Conference on Object-Oriented Programming Systems, Languages, and Applications, OOPSLA. Proceedings of the Conference on Object-Oriented Programming Systems, Languages, and Applications, OOPSLA, Association for Computing Machinery, pp. 309-328, 2014 ACM SIGPLAN Conference on Object-Oriented Programming, Systems, Languages, and Applications, OOPSLA 2014, Portland, United States, 10/20/14. https://doi.org/10.1145/10.1145/2660193.2660231
Misailovic S, Carbin M, Achour S, Qi Z, Rinard M. Chisel: Reliability-and accuracy-aware optimization of approximate computational kernels. In Proceedings of the Conference on Object-Oriented Programming Systems, Languages, and Applications, OOPSLA. Association for Computing Machinery. 2014. p. 309-328. (Proceedings of the Conference on Object-Oriented Programming Systems, Languages, and Applications, OOPSLA). https://doi.org/10.1145/10.1145/2660193.2660231
Misailovic, Sasa ; Carbin, Michael ; Achour, Sara ; Qi, Zichao ; Rinard, Martin. / Chisel : Reliability-and accuracy-aware optimization of approximate computational kernels. Proceedings of the Conference on Object-Oriented Programming Systems, Languages, and Applications, OOPSLA. Association for Computing Machinery, 2014. pp. 309-328 (Proceedings of the Conference on Object-Oriented Programming Systems, Languages, and Applications, OOPSLA).
@inproceedings{ebfc109b9f8343c2a8452160625c75d8,
title = "Chisel: Reliability-and accuracy-aware optimization of approximate computational kernels",
abstract = "The accuracy of an approximate computation is the distance between the result that the computation produces and the corresponding fully accurate result. The reliability of the computation is the probability that it will produce an acceptably accurate result. Emerging approximate hardware platforms provide approximate operations that, in return for reduced energy consumption and/or increased performance, exhibit reduced reliability and/or accuracy.We present Chisel, a system for reliability-and accuracyaware optimization of approximate computational kernels that run on approximate hardware platforms. Given a combined reliability and/or accuracy specification, Chisel automatically selects approximate kernel operations to synthesize an approximate computation that minimizes energy consumption while satisfying its reliability and accuracy specification.We evaluate Chisel on five applications from the image processing, scientific computing, and financial analysis domains. The experimental results showthat our implemented optimization algorithm enables Chisel to optimize our set of benchmark kernels to obtain energy savings from 8.7{\%} to 19.8{\%} compared to the original (exact) kernel implementations while preserving important reliability guarantees.",
keywords = "Approximate Computing",
author = "Sasa Misailovic and Michael Carbin and Sara Achour and Zichao Qi and Martin Rinard",
year = "2014",
month = "10",
day = "15",
doi = "10.1145/10.1145/2660193.2660231",
language = "English (US)",
series = "Proceedings of the Conference on Object-Oriented Programming Systems, Languages, and Applications, OOPSLA",
publisher = "Association for Computing Machinery",
pages = "309--328",
booktitle = "Proceedings of the Conference on Object-Oriented Programming Systems, Languages, and Applications, OOPSLA",

}

TY - GEN

T1 - Chisel

T2 - Reliability-and accuracy-aware optimization of approximate computational kernels

AU - Misailovic, Sasa

AU - Carbin, Michael

AU - Achour, Sara

AU - Qi, Zichao

AU - Rinard, Martin

PY - 2014/10/15

Y1 - 2014/10/15

N2 - The accuracy of an approximate computation is the distance between the result that the computation produces and the corresponding fully accurate result. The reliability of the computation is the probability that it will produce an acceptably accurate result. Emerging approximate hardware platforms provide approximate operations that, in return for reduced energy consumption and/or increased performance, exhibit reduced reliability and/or accuracy.We present Chisel, a system for reliability-and accuracyaware optimization of approximate computational kernels that run on approximate hardware platforms. Given a combined reliability and/or accuracy specification, Chisel automatically selects approximate kernel operations to synthesize an approximate computation that minimizes energy consumption while satisfying its reliability and accuracy specification.We evaluate Chisel on five applications from the image processing, scientific computing, and financial analysis domains. The experimental results showthat our implemented optimization algorithm enables Chisel to optimize our set of benchmark kernels to obtain energy savings from 8.7% to 19.8% compared to the original (exact) kernel implementations while preserving important reliability guarantees.

AB - The accuracy of an approximate computation is the distance between the result that the computation produces and the corresponding fully accurate result. The reliability of the computation is the probability that it will produce an acceptably accurate result. Emerging approximate hardware platforms provide approximate operations that, in return for reduced energy consumption and/or increased performance, exhibit reduced reliability and/or accuracy.We present Chisel, a system for reliability-and accuracyaware optimization of approximate computational kernels that run on approximate hardware platforms. Given a combined reliability and/or accuracy specification, Chisel automatically selects approximate kernel operations to synthesize an approximate computation that minimizes energy consumption while satisfying its reliability and accuracy specification.We evaluate Chisel on five applications from the image processing, scientific computing, and financial analysis domains. The experimental results showthat our implemented optimization algorithm enables Chisel to optimize our set of benchmark kernels to obtain energy savings from 8.7% to 19.8% compared to the original (exact) kernel implementations while preserving important reliability guarantees.

KW - Approximate Computing

UR - http://www.scopus.com/inward/record.url?scp=84908285161&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84908285161&partnerID=8YFLogxK

U2 - 10.1145/10.1145/2660193.2660231

DO - 10.1145/10.1145/2660193.2660231

M3 - Conference contribution

AN - SCOPUS:84908285161

T3 - Proceedings of the Conference on Object-Oriented Programming Systems, Languages, and Applications, OOPSLA

SP - 309

EP - 328

BT - Proceedings of the Conference on Object-Oriented Programming Systems, Languages, and Applications, OOPSLA

PB - Association for Computing Machinery

ER -