Two-dimensional cache-oblivious sparse matrix–vector multiplication

► Extended a one-dimensional cache-oblivious sparse matrix–vector multiplication scheme to two dimensions. ► Introduced the doubly separated block diagonal (DSBD) form for sparse matrices. ► Introduced a family of recursive sparse blocking schemes for matrices in DSBD form. ► Obtained speedups in sp...

Full description

Saved in:
Bibliographic Details
Published in:Parallel computing Vol. 37; no. 12; pp. 806 - 819
Main Authors: Yzelman, A.N., Bisseling, Rob H.
Format: Journal Article
Language:English
Published: Elsevier B.V 01.12.2011
Subjects:
Blocking
Cache-oblivious
Computation
Data structures
Fine-grain
Gain
Matrix–vector multiplication
Multiplication
Parallel computing
Parallel processing
Partitioning
Recursive bipartitioning
Sparse matrix
Two dimensional
Sparse matrix
Parallel computing
Cache-oblivious
Recursive bipartitioning
Matrix–vector multiplication
Fine-grain
ISSN:0167-8191, 1872-7336
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:► Extended a one-dimensional cache-oblivious sparse matrix–vector multiplication scheme to two dimensions. ► Introduced the doubly separated block diagonal (DSBD) form for sparse matrices. ► Introduced a family of recursive sparse blocking schemes for matrices in DSBD form. ► Obtained speedups in sparse matrix–vector multiplication of over a factor of three. In earlier work, we presented a one-dimensional cache-oblivious sparse matrix–vector (SpMV) multiplication scheme which has its roots in one-dimensional sparse matrix partitioning. Partitioning is often used in distributed-memory parallel computing for the SpMV multiplication, an important kernel in many applications. A logical extension is to move towards using a two-dimensional partitioning. In this paper, we present our research in this direction, extending the one-dimensional method for cache-oblivious SpMV multiplication to two dimensions, while still allowing only row and column permutations on the sparse input matrix. This extension requires a generalisation of the compressed row storage data structure to a block-based data structure, for which several variants are investigated. Experiments performed on three different architectures show further improvements of the two-dimensional method compared to the one-dimensional method, especially in those cases where the one-dimensional method already provided significant gains. The largest gain obtained by our new reordering is over a factor of 3 in SpMV speed, compared to the natural matrix ordering.
Bibliography:ObjectType-Article-2
SourceType-Scholarly Journals-1
ObjectType-Feature-1
content type line 23
ISSN:0167-8191
1872-7336
DOI:10.1016/j.parco.2011.08.004