A Processor for Two-Dimensional Symmetric Eigenvalue and Singular Value Arrays.

Abstract : A dedicated processor architecture is here introduced for multiprocessor array implementations of Jacobi methods. This new processing element factors arbitrary rotations into products of elementary rotations whose angles are exactly twice the CORDIC elementary angles. This characteristic permits complete concurrency between the evaluation of the Jacobi rotations and their application. Thus the processors in the resulting arrays are almost never idle. Keywords: Symmetric eigenvalues, Singular value decomposition, Signal processing.