Method and device for singular value and feature value composition of matrix on processing system

Abstract

The invention provides a method and a device for the singular value and feature value composition of a matrix on a processing system. The method for singular value composition includes: the matrix is converted into a bidiagonal matrix, and a sandwich matrix which needs to be converted into a singular vector matrix is obtained at the same time; submatrixes of which the elements are not zero are found out and converted in an iterative way on the vice diagonals in the bidiagonal matrix so that the bidiagonal matrix can be converged into a diagonal matrix and the sandwich matrix can be converted into the singular vector matrix; in each iterative step: Givens rotation parameter groups generated in the QR iterative processes of the submatrixes and the positions of the left boundary and the right boundary of the corresponding submatrixes which need to use the Givens rotation parameter groups in the sandwich matrix are accumulated in a memory; and when the number of the accumulated Givens rotation parameter groups reaches the preset threshold value, the Givens rotation parameter groups are utilized to carry out the Givens rotation of the sandwich matrix, wherein the threshold value is an integer which is higher than or equal to 2.

Description

technical field [0001] The present invention relates to the field of data processing, in particular, to a method and device for singular value decomposition of a matrix on a processing system and a method and device for eigenvalue decomposition of a matrix. Background technique [0002] LAPACK (Linear Algebra PACKage, linear algebra package) is a very effective, powerful and widely used linear algebra function library jointly developed by Oak Ridge National Laboratory, University of California, Davis, and University of Illinois, which is used in different high-performance computing environments. Efficiently solve numerical linear algebra problems, which has effectively worked for HPC (High Performance Computing) and computational science organizations for over 20 years. For details about LAPACK, please refer to http: / / netlib.amss.ac.cn / lapack / index.html. [0003] As a professional linear algebra library, LAPACK provides various linear algebra subroutines, including routines...

AI Technical Summary

Problems solved by technology

[0047] In view of the above problems, the present invention provides a method and device for singular value decomposition of a matrix and a method and device for eigenvalue decomposition of a matrix on a processing system, so as to solve the problem caused by the memory bandwidth limitation in the processing system Calculation performance degradation problem

Images