A maximally pipelined tridiagonal linear equation solver

Journal Article · Sun Jun 01 04:00:00 EDT 1986 · J. Parallel Distrib. Comput.; (United States)

DOI:https://doi.org/10.1016/0743-7315(86)90005-5· OSTI ID:6471987

Gao, G R

A maximally pipelined approach for solving tridiagonal systems of linear equations is proposed and a complete machine code structure for the algorithm is constructed. The new scheme has several important advantages over other existing parallel algorithms. Although the primary target machine used in this paper is a static data flow computer, the author expects that the principle can also be applied to other data flow computers, such as the dynamic data flow machine; however, a different perspective of pipelining may be required. It is the author's belief that the basic ideas may also be useful for a conventional parallel machine architecture. Future research is needed to implement such an algorithm on a real static data flow machine yet to be built. An automatic program transformation for both multiple backup and multiple copies of code presents an interesting challenge to the compiler construction for such computers.

• Research Organization: Lab. for Computer Science, Massachusetts Institute of Technology, Cambridge, MA 02139
• OSTI ID: 6471987
• Journal Information: J. Parallel Distrib. Comput.; (United States), Journal Name: J. Parallel Distrib. Comput.; (United States) Vol. 3:2; ISSN JPDCE
• Country of Publication: United States
• Language: English

Related Subjects

99 GENERAL AND MISCELLANEOUS
990210* -- Supercomputers-- (1987-1989)
990230 -- Mathematics & Mathematical Models-- (1987-1989)
ALGORITHMS
ARRAY PROCESSORS
COMPUTER ARCHITECTURE
COMPUTER CODES
COMPUTERS
DATA-FLOW PROCESSING
DIGITAL COMPUTERS
EQUATIONS
EXECUTIVE CODES
MACHINE TRANSLATIONS
MATHEMATICAL LOGIC
MEMORY DEVICES
NUMERICAL SOLUTION
PARALLEL PROCESSING
PROGRAMMING
SUPERCOMPUTERS
TENSORS
TRANSLATORS
VECTORS