Multithreading in the PLASMA library

Jakub Kurzak, Piotr Luszczek, Asim YarKhan, Mathieu Faverge, Julien Langou, Henricus Bouwmeester, Jack Dongarra

Research output: Chapter in Book/Report/Conference proceedingChapterpeer-review

15 Scopus citations

Abstract

Parallel Linear Algebra Software for Multicore Architectures (PLASMA) is a numerical software library for solving problems in dense linear algebra on systems of multicore processors and multisocket systems of multicore processors [1]. PLASMA offers routines for solving a wide range of problems in dense linear algebra such as nonsymmetric, symmetric, and symmetric positive definite systems of linear equations, least square problems, singular value problems, and eigenvalue problems (currently only symmetric eigenvalue problems). PLASMA solves these problems in real and complex arithmetic and in single and double precision. PLASMA is designed to give high efficiency on homogeneous multicore processors and multisocket systems of multicore processors. As of today, the majority of such systems are on-chip symmetric multiprocessors with classic super-scalar processors as their building blocks (x86 and alike) augmented with short-vector SIMD extensions (SSE and alike). PLASMA has been designed to supercede LAPACK [2], principally by restructuring the software to achieve much greater efficiency on modern computers based on multicore processors.

Original languageEnglish
Title of host publicationMulticore Computing
Subtitle of host publicationAlgorithms, Architectures, and Applications
PublisherCRC Press
Pages119-142
Number of pages24
ISBN (Electronic)9781439854358
ISBN (Print)9781439854341
DOIs
StatePublished - Jan 1 2013
Externally publishedYes

Fingerprint

Dive into the research topics of 'Multithreading in the PLASMA library'. Together they form a unique fingerprint.

Cite this