Virtual systolic array for QR decomposition

Jakub Kurzak; Piotr Luszczek; Mark Gates; Ichitaro Yamazaki; Jack Dongarra

doi:10.1109/IPDPS.2013.119

Virtual systolic array for QR decomposition

Jakub Kurzak, Piotr Luszczek, Mark Gates, Ichitaro Yamazaki, Jack Dongarra

Computer Science and Mathematics Div

Research output: Contribution to conference › Paper › peer-review

4 Scopus citations

Abstract

Systolic arrays offer a very attractive, data centric, execution model as an alternative to the von Neumann architecture. Hardware implementations of systolic arrays turned out not to be viable solutions in the past. This article shows how the systolic design principles can be applied to a software solution to deliver an algorithm with unprecedented strong scaling capabilities. Systolic array for the QR decomposition is developed and a virtualization layer is used for mapping of the algorithm to a large distributed memory system. Strong scaling properties are discovered, superior to existing solutions.

Original language	English
Pages	251-260
Number of pages	10
DOIs	https://doi.org/10.1109/IPDPS.2013.119
State	Published - 2013
Event	27th IEEE International Parallel and Distributed Processing Symposium, IPDPS 2013 - Boston, MA, United States Duration: May 20 2013 → May 24 2013

Conference

Conference	27th IEEE International Parallel and Distributed Processing Symposium, IPDPS 2013
Country/Territory	United States
City	Boston, MA
Period	05/20/13 → 05/24/13

Keywords

QR decomposition
dataflow programming
message passing
multi-core
roofline model
systolic array

Access to Document

10.1109/IPDPS.2013.119

Cite this

@conference{fa6636ad20d74f4087d4d2c3aa5548b7,

title = "Virtual systolic array for QR decomposition",

abstract = "Systolic arrays offer a very attractive, data centric, execution model as an alternative to the von Neumann architecture. Hardware implementations of systolic arrays turned out not to be viable solutions in the past. This article shows how the systolic design principles can be applied to a software solution to deliver an algorithm with unprecedented strong scaling capabilities. Systolic array for the QR decomposition is developed and a virtualization layer is used for mapping of the algorithm to a large distributed memory system. Strong scaling properties are discovered, superior to existing solutions.",

keywords = "QR decomposition, dataflow programming, message passing, multi-core, roofline model, systolic array",

author = "Jakub Kurzak and Piotr Luszczek and Mark Gates and Ichitaro Yamazaki and Jack Dongarra",

year = "2013",

doi = "10.1109/IPDPS.2013.119",

language = "English",

pages = "251--260",

note = "27th IEEE International Parallel and Distributed Processing Symposium, IPDPS 2013 ; Conference date: 20-05-2013 Through 24-05-2013",

}

TY - CONF

T1 - Virtual systolic array for QR decomposition

AU - Kurzak, Jakub

AU - Luszczek, Piotr

AU - Gates, Mark

AU - Yamazaki, Ichitaro

AU - Dongarra, Jack

PY - 2013

Y1 - 2013

N2 - Systolic arrays offer a very attractive, data centric, execution model as an alternative to the von Neumann architecture. Hardware implementations of systolic arrays turned out not to be viable solutions in the past. This article shows how the systolic design principles can be applied to a software solution to deliver an algorithm with unprecedented strong scaling capabilities. Systolic array for the QR decomposition is developed and a virtualization layer is used for mapping of the algorithm to a large distributed memory system. Strong scaling properties are discovered, superior to existing solutions.

AB - Systolic arrays offer a very attractive, data centric, execution model as an alternative to the von Neumann architecture. Hardware implementations of systolic arrays turned out not to be viable solutions in the past. This article shows how the systolic design principles can be applied to a software solution to deliver an algorithm with unprecedented strong scaling capabilities. Systolic array for the QR decomposition is developed and a virtualization layer is used for mapping of the algorithm to a large distributed memory system. Strong scaling properties are discovered, superior to existing solutions.

KW - QR decomposition

KW - dataflow programming

KW - message passing

KW - multi-core

KW - roofline model

KW - systolic array

UR - http://www.scopus.com/inward/record.url?scp=84884899002&partnerID=8YFLogxK

U2 - 10.1109/IPDPS.2013.119

DO - 10.1109/IPDPS.2013.119

M3 - Paper

AN - SCOPUS:84884899002

SP - 251

EP - 260

T2 - 27th IEEE International Parallel and Distributed Processing Symposium, IPDPS 2013

Y2 - 20 May 2013 through 24 May 2013

ER -

Virtual systolic array for QR decomposition

Abstract

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Cite this