ScaLAPACK tutorial

Jack Dongarra, Antoine Petitet

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

2 Scopus citations

Abstract

This ScaLAPACK tutorial begins with a brief description of the LAPACK library. The importance of block-partitioned algorithms in reducing the frequency of data movement between different levels of hierarchical memory is stressed. By relying on the Basic Linear Algebra Subprograms (BLAS) it is possible to develop portable and efficient implementations of these algorithms across a wide range of architectures, with emphasis on workstations, vector-processors and shared-memory computers, as has been done in LAPACK. The ScaLAPACK library, which is a distributed memory version of LAPACK is then presented. A key idea in our approach is the use of Basic Linear Algebra Communication Subprograms (BLACS) as communication building blocks and the use of a distributed version of the BLAS, the Parallel Basic Linear Algebra Subprograms (PBLAS) as computational building blocks. The BLACS and PBLAS features are in turn outlined and it is shown how these building blocks can be used to construct higher-level algorithms, and hide many details of the parallelism from the application developer. Performance results of ScaLAPACK routines are presented validating the adoption of the block-cyclic decomposition scheme as a way of distributing block-partitioned matrices yielding to well balanced computations and scalable implementations. Finally, future directions for the ScaLAPACK library are described and alternative approaches to mathematical libraries are suggested that could integrate ScaLAPACK into efficient and user-friendly distributed systems.

Original languageEnglish
Title of host publicationApplied Parallel Computing
Subtitle of host publicationComputations in Physics, Chemistry and Engineering Science - 2nd International Workshop, PARA 1995, Proceedings
EditorsJack Dongarra, Kaj Madsen, Jerzy Wasniewśki
PublisherSpringer Verlag
Pages166-176
Number of pages11
ISBN (Print)3540609024, 9783540609025
DOIs
StatePublished - 1996
Event2nd International Workshop on Applied Parallel Computing in Computations in Physics, Chemistry and Engineering Science, PARA 1995 - Lyngby, Denmark
Duration: Aug 21 1995Aug 24 1995

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume1041
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference2nd International Workshop on Applied Parallel Computing in Computations in Physics, Chemistry and Engineering Science, PARA 1995
Country/TerritoryDenmark
CityLyngby
Period08/21/9508/24/95

Bibliographical note

Publisher Copyright:
© Springer-Verlag Berlin Heidelberg 1996.

Funding

FundersFunder number
National Science FoundationASC-9005933

    Fingerprint

    Dive into the research topics of 'ScaLAPACK tutorial'. Together they form a unique fingerprint.

    Cite this