CSX: An extended compression format for SpMV on shared memory systems

Kourtis, K; Karakasis, V; Goumas, G; Koziris, N

dc.contributor.author	Kourtis, K	en
dc.contributor.author	Karakasis, V	en
dc.contributor.author	Goumas, G	en
dc.contributor.author	Koziris, N	en
dc.date.accessioned	2014-03-01T02:47:19Z
dc.date.available	2014-03-01T02:47:19Z
dc.date.issued	2011	en
dc.identifier.issn	0362-1340	en
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/33074
dc.subject	Compression	en
dc.subject	Shared memory	en
dc.subject	SMP	en
dc.subject	Sparse matrix-vector multiplication	en
dc.subject	SpMV	en
dc.subject.classification	Computer Science, Software Engineering	en
dc.subject.other	Data volume	en
dc.subject.other	ITS data	en
dc.subject.other	matrix	en
dc.subject.other	Multiple processing	en
dc.subject.other	Offline	en
dc.subject.other	Performance Gain	en
dc.subject.other	Run-time code generation	en
dc.subject.other	Shared memories	en
dc.subject.other	Shared memory system	en
dc.subject.other	SMP	en
dc.subject.other	Sparse matrices	en
dc.subject.other	Sparse matrix-vector multiplication	en
dc.subject.other	SpMV	en
dc.subject.other	Storage formats	en
dc.subject.other	Metadata	en
dc.subject.other	Matrix algebra	en
dc.title	CSX: An extended compression format for SpMV on shared memory systems	en
heal.type	conferenceItem	en
heal.identifier.primary	10.1145/2038037.1941587	en
heal.identifier.secondary	http://dx.doi.org/10.1145/2038037.1941587	en
heal.language	English	en
heal.publicationDate	2011	en
heal.abstract	The Sparse Matrix-Vector multiplication (SpMV) kernel scales poorly on shared memory systems with multiple processing units due to the streaming nature of its data access pattern. Previous re- search has demonstrated that an effective strategy to improve the kernel's performance is to drastically reduce the data volume in- volved in the computations. Since the storage formats for sparse matrices include metadata describing the structure of non-zero el- ements within the matrix, we propose a generalized approach to compress metadata by exploiting substructures within the matrix. We call the proposed storage format Compressed Sparse eXtended (CSX). In our implementation we employ runtime code generation to construct specialized SpMV routines for each matrix. Experi- mental evaluation on two shared memory systems for 15 sparse matrices demonstrates significant performance gains as the number of participating cores increases. Regarding the cost of CSX con- struction, we propose several strategies which trade performance for preprocessing cost making CSX applicable both to online and offline preprocessing. Copyright © 2011 ACM.	en
heal.publisher	ASSOC COMPUTING MACHINERY	en
heal.journalName	ACM SIGPLAN Notices	en
dc.identifier.doi	10.1145/2038037.1941587	en
dc.identifier.isi	ISI:000296264900025	en
dc.identifier.volume	46	en
dc.identifier.issue	8	en
dc.identifier.spage	247	en
dc.identifier.epage	256	en