Parallel blocked sparse matrix-vector multiplication with dynamic parameter selection method

Authors:
Makoto Kudo;Hisayasu Kuroda;Yasumasa Kanada
Affiliations:
Department of Computer Science, Graduate School of Information Science and Technology, The University of Tokyo;Super Computing Division, Information Technology Center, The University of Tokyo;Super Computing Division, Information Technology Center, The University of Tokyo
Venue:
ICCS'03 Proceedings of the 2003 international conference on Computational science: PartIII
Year:
2003

Citing 5
Cited 1

Improving the memory-system performance of sparse-matrix vector multiplication

IBM Journal of Research and Development
Improving performance of sparse matrix-vector multiplication

SC '99 Proceedings of the 1999 ACM/IEEE conference on Supercomputing
Performance optimizations and bounds for sparse matrix-vector multiply

Proceedings of the 2002 ACM/IEEE conference on Supercomputing
Optimizing the performance of sparse matrix-vector multiplication

Optimizing the performance of sparse matrix-vector multiplication
On Improving the Performance of Sparse Matrix-Vector Multiplication

HIPC '97 Proceedings of the Fourth International Conference on High-Performance Computing

An evaluation towards automatically tuned eigensolvers

LSSC'05 Proceedings of the 5th international conference on Large-Scale Scientific Computing

Quantified Score

Hi-index	0.00

Visualization

Abstract

A blocking method is a popular optimization technique for sparse matrix-vector multiplication (SpMxV). In this paper, a new blocking method which generalizes the conventional two blocking methods and its application to the parallel environment are proposed. This paper also proposes a dynamic parameter selection method for blocked parallel SpMxV which automatically selects the parameter set according to the characteristics of the target matrix and machine in order to achieve high performance on various computational environments. The performance with dynamically selected parameter set is compared with the performance with generally-used fixed parameter sets for 12 types of sparse matrices on four parallel machines: including PentiumIII, Sparc II, MIPS R12000 and Itanium. The result shows that the performance with dynamically selected parameter set is the best in most cases.