An Implementation of the Singular Value Decomposition on the Connection Machine CM-2
Abstract
In modern digital signal processing, the singular value decomposition is increasingly recognized as an important mathematical tool. The true measure of usefulness of such a tool is very much dependent on the ability to compute it at 'supercomputer' throughput rates. This report describes an implementation of the singular value decomposition (SVD) on the Connection Machine CM-2 using parallel Fortran. The algorithm is based on Hestenes's, which is a Jacobi iteration in which pairs of rows are rotated to become orthogonal. The Fortran implementation of this algorithm on a full CM-2 is comparable in execution speed to the Linpack implementation on a Convex C220 processor.
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 11, 1991
- Accession Number
- ADA234124
Entities
People
- Nhi-anh Chu
Organizations
- United States Naval Research Laboratory