Systolic Partitioned And Banded Linear Algebraic Computations

J. Greg Nash; Siegfried Hansen; K. Wojtek Przytula

doi:10.1117/12.976241

23 March 1986 Systolic Partitioned And Banded Linear Algebraic Computations

J. Greg Nash, Siegfried Hansen, K. Wojtek Przytula

Proceedings Volume 0698, Real-Time Signal Processing IX; (1986) https://doi.org/10.1117/12.976241
Event: 30th Annual Technical Symposium, 1986, San Diego, United States

Abstract

For general linear algebraic computations in science, engineering and mathematics, problem sizes typically grow until constraints imposed by computation rates, I/O or memory availability result in unreasonably long execution times. From this point of view such computing architectures as systolic arrays provide a potential for processing much larger arrays. However, in order to utilize this potential, the problem of how to efficiently decompose large arrays onto a fixed systolic array in a numerically stable way must be solved. In some cases this is most efficiently done by partitioning the algorithm so that it can be run as a sequence of operations on input subarrays that are of the same size as the underlying hardware array. For some cases, such as those associated with banded matrices, it is necessary only to process the non-zero data. In all cases, however, the algorithms must run on the same hardware array without an undue amount of extra control or communication overhead. In this paper we describe algorithmic and architectural approaches to providing a fairly broad range of linear algebraic operations on matrices that are much larger than the underlying hardware array. All of the techniques we describe here have been simulated in detail using APL in order to verify correctness.

Citation Download Citation

J. Greg Nash, Siegfried Hansen, and K. Wojtek Przytula "Systolic Partitioned And Banded Linear Algebraic Computations", Proc. SPIE 0698, Real-Time Signal Processing IX, (23 March 1986); https://doi.org/10.1117/12.976241

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available