A framework for out of memory SVD algorithms

Khairul Kabir, Azzam Haidar, Stanimire Tomov, Aurelien Bouteiller, Jack Dongarra

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

9 Scopus citations

Abstract

Many important applications – from big data analytics to information retrieval, gene expression analysis, and numerical weather prediction – require the solution of large dense singular value decompositions (SVD). In many cases the problems are too large to fit into the computer’s main memory, and thus require specialized out-of-core algorithms that use disk storage. In this paper, we analyze the SVD communications, as related to hierarchical memories, and design a class of algorithms that minimizes them. This class includes out-of-core SVDs but can also be applied between other consecutive levels of the memory hierarchy, e.g., GPU SVD using the CPU memory for large problems. We call these out-of-memory (OOM) algorithms. To design OOM SVDs, we first study the communications for both classical one-stage blocked SVD and two-stage tiled SVD. We present the theoretical analysis and strategies to design, as well as implement, these communication avoiding OOM SVD algorithms. We show performance results for multicore architecture that illustrate our theoretical findings and match our performance models.

Original languageEnglish
Title of host publicationHigh Performance Computing - 32nd International Conference, ISC High Performance 2017, Proceedings
EditorsJulian M. Kunkel, Pavan Balaji, David Keyes, Rio Yokota
PublisherSpringer Verlag
Pages158-178
Number of pages21
ISBN (Print)9783319586663
DOIs
StatePublished - 2017
Event32nd International Conference, ISC High Performance, 2017 - Frankfurt, Germany
Duration: Jun 18 2017Jun 22 2017

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume10266 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference32nd International Conference, ISC High Performance, 2017
Country/TerritoryGermany
CityFrankfurt
Period06/18/1706/22/17

Funding

This research was supported by the Exascale Computing Project (17-SC-20-SC), a collaborative effort of the U.S. Department of Energy Office of Science and the National Nuclear Security Administration.

FundersFunder number
U.S. Department of Energy
National Nuclear Security Administration

    Fingerprint

    Dive into the research topics of 'A framework for out of memory SVD algorithms'. Together they form a unique fingerprint.

    Cite this