Parallel Block Hessenberg Reduction using Algorithms-By-Tiles for Multicore Architectures Revisited

Ltaief, Hatem and Kurzak, Jakub and Dongarra, Jack (2009) Parallel Block Hessenberg Reduction using Algorithms-By-Tiles for Multicore Architectures Revisited. [MIMS Preprint]

PDF ltaief_kurzak_dongarra_070808.pdf Download (430kB)

Abstract

The objective of this paper is to extend and redesign the block matrix reduction applied for the family of two-sided factorizations, introduced by Dongarra et al. [9], to the context of multicore architec- tures using algorithms-by-tiles. In particular, the Block Hessenberg Re- duction is very often used as a pre-processing step in solving dense linear algebra problems, such as the standard eigenvalue problem. Although expensive, orthogonal transformations are commonly used for this re- duction because they guarantee stability, as opposed to Gaussian Elimi- nation. Two versions of the Block Hessenberg Reduction are presented in this paper, the rst one with Householder re ectors and the second one with Givens rotations. A short investigation on variants of Fast Givens Rotations is also mentioned. Furthermore, in the last Top500 list from June 2008, 98% of the fastest parallel systems in the world are based on multicores. The emerging petascale systems consisting of hundreds of thousands of cores have exacerbated the problem even more and it becomes judicious to eciently integrate existing or new numerical lin- ear algebra algorithms suitable for such hardwares. By exploiting the concepts of algorithms-by-tiles in the multicore environment (i.e., high level of parallelism with ne granularity and high performance data rep- resentation combined with a dynamic data driven execution), the Block Hessenberg Reduction presented here achieves 72% of the DGEMM peak on a 12000 12000 matrix with 16 Intel Tigerton 2:4 GHz processors.

Item Type:	MIMS Preprint
Additional Information:	Appears also as Technical Report UT-CS-08-631, Department of Computer Science, University of Tennessee, Knoxville, TN, USA, June 2008 and as LAPACK Working Note 209"
Subjects:	MSC 2010, the AMS's Mathematics Subject Classification > 65 Numerical analysis MSC 2010, the AMS's Mathematics Subject Classification > 68 Computer science
Depositing User:	Ms Lucy van Russelt
Date Deposited:	13 Jan 2009
Last Modified:	20 Oct 2017 14:12
URI:	https://eprints.maths.manchester.ac.uk/id/eprint/1212

Actions (login required)

View Item