A class of communication-avoiding algorithms for solving general dense linear systems on CPU/GPU parallel machines

Marc Baboulin; Simplice Donfack; Jack Dongarra; Laura Grigori; Adrien Rémy; Stanimire Tomov

hal-00656457, version 3

A class of communication-avoiding algorithms for solving general dense linear systems on CPU/GPU parallel machines

Marc Baboulin ( Author to contact preferably , ) ¹, Simplice Donfack () ¹, Jack Dongarra () ², Laura Grigori () ¹, Adrien Rémy () ¹, Stanimire Tomov () ²

N° RR-7854 (2012)

Abstract: We study several solvers for the solution of general linear systems where the main objective is to reduce the communication overhead due to pivoting. We first describe two existing algorithms for the LU factorization on hybrid CPU/GPU architectures. The first one is based on partial pivoting and the second uses a random preconditioning of the original matrix to avoid pivoting. Then we introduce a solver where the panel factorization is performed using a communication-avoiding pivoting heuristic while the update of the trailing submatrix is performed by the GPU. We provide performance comparisons for these solvers on current hybrid multicore-GPU parallel machines.

1: Laboratoire de Recherche en Informatique (LRI)
CNRS : UMR8623 – Université Paris XI - Paris Sud
2: Innovative Computing Laboratory (ICL)
University of Tennessee

hal-00656457, version 3
http://hal.inria.fr/hal-00656457
oai:hal.inria.fr:hal-00656457
From: Marc Baboulin <>
Submitted on: Wednesday, 29 February 2012 11:27:57
Updated on: Wednesday, 29 February 2012 13:02:58

See detailed view

Associated documents

PDF :

Export

Bibtex EndNote TEI RefWorks

<?xml version="1.0" encoding="UTF-8"?>
<listBibl xmlns="http://www.tei-c.org/ns/1.0" xmlns:xlink="http://www.w3.org/1999/xlink" xmlns:mml="http://www.w3.org/1998/Math/MathML">
<listBibl type="report"><head>Research reports</head>
<biblStruct type="report" xml:id="hal-00656457"><monogr>
<author><persName><forename>Marc</forename><surname>Baboulin</surname></persName>
<affiliation>
<orgName type="laboratory">LRI</orgName>
<orgName type="team">Laboratoire de Recherche en Informatique</orgName>
<country>FR</country>
<orgName type="institution">CNRS : UMR8623</orgName>
<orgName type="institution">Université Paris XI - Paris Sud</orgName>
</affiliation>
</author>
<author><persName><forename>Simplice</forename><surname>Donfack</surname></persName>
<affiliation>
<orgName type="laboratory">LRI</orgName>
<orgName type="team">Laboratoire de Recherche en Informatique</orgName>
<country>FR</country>
<orgName type="institution">CNRS : UMR8623</orgName>
<orgName type="institution">Université Paris XI - Paris Sud</orgName>
</affiliation>
</author>
<author><persName><forename>Jack</forename><surname>Dongarra</surname></persName>
<affiliation>
<orgName type="laboratory">ICL</orgName>
<orgName type="team">Innovative Computing Laboratory</orgName>
<country>US</country>
<orgName type="institution">University of Tennessee</orgName>
</affiliation>
</author>
<author><persName><forename>Laura</forename><surname>Grigori</surname></persName>
<affiliation>
<orgName type="laboratory">LRI</orgName>
<orgName type="team">Laboratoire de Recherche en Informatique</orgName>
<country>FR</country>
<orgName type="institution">CNRS : UMR8623</orgName>
<orgName type="institution">Université Paris XI - Paris Sud</orgName>
</affiliation>
</author>
<author><persName><forename>Adrien</forename><surname>Rémy</surname></persName>
<affiliation>
<orgName type="laboratory">LRI</orgName>
<orgName type="team">Laboratoire de Recherche en Informatique</orgName>
<country>FR</country>
<orgName type="institution">CNRS : UMR8623</orgName>
<orgName type="institution">Université Paris XI - Paris Sud</orgName>
</affiliation>
</author>
<author><persName><forename>Stanimire</forename><surname>Tomov</surname></persName>
<affiliation>
<orgName type="laboratory">ICL</orgName>
<orgName type="team">Innovative Computing Laboratory</orgName>
<country>US</country>
<orgName type="institution">University of Tennessee</orgName>
</affiliation>
</author>
<title type="main" level="m">A class of communication-avoiding algorithms for solving general dense linear systems on CPU/GPU parallel machines</title><title type="main" level="m">A class of communication-avoiding algorithms for solving general dense linear systems on CPU/GPU parallel machines</title><imprint><biblScope type="publicationDate">2012-01-07</biblScope></imprint></monogr><idno type="HALid">http://hal.inria.fr/hal-00656457</idno><idno type="HALFile">http://hal.inria.fr/hal-00656457/PDF/RR-7854.pdf</idno></biblStruct>
</listBibl>
</listBibl>