Batched computation of the singular value decompositions of order two by the AVX-512 vectorization

DOI10.1142/S0129626420500152zbMATH Open1490.65063arXiv2005.07403OpenAlexW3118611730MaRDI QIDQ5087080FDOQ5087080

Authors: Vedran Novaković 

Publication date: 8 July 2022

Published in: Parallel Processing Letters (Search for Journal in Brave)

Abstract: In this paper a vectorized algorithm for simultaneously computing up to eight singular value decompositions (SVDs, each of the form $A=USigma{V}^{ast}$) of real or complex matrices of order two is proposed. The algorithm extends to a batch of matrices of an arbitrary length $n$, that arises, for example, in the annihilation part of the parallel Kogbetliantz algorithm for the SVD of a square matrix of order $2n$. The SVD algorithm for a single matrix of order two is derived first. It scales, in most instances error-free, the input matrix $A$ such that its singular values $Sigm{a}_{ii}$ cannot overflow whenever its elements are finite, and then computes the URV factorization of the scaled matrix, followed by the SVD of a non-negative upper-triangular middle factor. A vector-friendly data layout for the batch is then introduced, where the same-indexed elements of each of the input and the output matrices form vectors, and the algorithm's steps over such vectors are described. The vectorized approach is then shown to be about three times faster than processing each matrix in isolation, while slightly improving accuracy over the straightforward method for the $2imes2$ SVD.

Full work available at URL: https://arxiv.org/abs/2005.07403

Recommendations

• Vectorization of a Thread-Parallel Jacobi Singular Value Decomposition Method
• A One-Sided Jacobi Algorithm for Computing the Singular Value Decomposition on a Vector Computer
• Computing the singular value decomposition on a distributed system of vector processors
• scientific article; zbMATH DE number 3963085
• Reducing the Computations of the Singular Value Decomposition Array Given by Brent and Luk

zbMATH Keywords

singular value decompositionAVX-512 vectorizationbatched computation

Mathematics Subject Classification ID

Parallel numerical computation (65Y05) Numerical computation of eigenvalues and eigenvectors of matrices (65F15) Numerical algorithms for specific classes of architectures (65Y10)

Cites Work

• LAPACK Users' Guide
• Block-oriented \(J\)-Jacobi methods for Hermitian matrices
• Dynamic ordering for a parallel block-Jacobi SVD algorithm
• Accuracy of two SVD algorithms for \(2\times 2\) triangular matrices
• On efficient implementations of Kogbetliantz's algorithm for computing the singular value decomposition
• Full block \(J\)-Jacobi method for Hermitian matrices
• Solution of linear equations by diagonalization of coefficients matrix
• On high relative accuracy of the Kogbetliantz method
• Implementation of Jacobi Rotations for Accurate Singular Value Computation in Floating Point Arithmetic
• Asymptotic quadratic convergence of the two-sided serial and parallel block-Jacobi SVD algorithm
• Globally convergent Jacobi methods for positive definite matrix pairs
• The LAPW method with eigendecomposition based on the Hari-Zimmermann generalized hyperbolic SVD
• On the global convergence of the complex HZ method

Cited In (2)

• A Kogbetliantz-type algorithm for the hyperbolic SVD
• Vectorization of a Thread-Parallel Jacobi Singular Value Decomposition Method

Uses Software

• LAPACK
• CUDA

This page was built for publication: Batched computation of the singular value decompositions of order two by the AVX-512 vectorization

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q5087080)