Hierarchical blocking and data flow analysis for numerical linear algebra

Author

Chen, Dingju

Author_Institution

Center for Appl. Math., Cornell Univ., Ithaca, NY, USA

fYear

1990

fDate

12-16 Nov 1990

Firstpage

12

Lastpage

19

Abstract

The optimization of BLAS2 and BLAS3 for linear algebra on computers with hierarchical memory systems is discussed. A novel blocking strategy called hierarchical blocking and data-flow analysis is proposed, and its applications are given. Numerical results on IBM RISC System/6000 Model 530 are given. It is shown that data flow direction and leading dimensions are crucial factors in optimizing linear algebra programs. Since the proposed method follows the data flow of a program, it can reveal the effect of pipelining between data loading units and computational units

Keywords

linear algebra; matrix algebra; pipeline processing; BLAS2; BLAS3; IBM RISC System/6000 Model 530; data flow analysis; data flow direction; hierarchical blocking; hierarchical memory systems; leading dimensions; numerical linear algebra; optimization; pipelining; Computer aided instruction; Computer architecture; Data analysis; Distributed computing; Linear algebra; Mathematical model; Mathematics; Reduced instruction set computing; Registers; Workstations;

fLanguage

English

Publisher

ieee

Conference_Titel

Supercomputing '90., Proceedings of

Conference_Location

New York, NY

Print_ISBN

0-8186-2056-0

Type

conf

DOI

10.1109/SUPERC.1990.129996

Filename

129996