Premium
A vectorized version of a sparse matrix‐vector multiply
Author(s) -
Hayes Linda J.,
Devloo Phillippe
Publication year - 1986
Publication title -
international journal for numerical methods in engineering
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.421
H-Index - 168
eISSN - 1097-0207
pISSN - 0029-5981
DOI - 10.1002/nme.1620230605
Subject(s) - sparse matrix , matrix (chemical analysis) , band matrix , finite element method , algorithm , element (criminal law) , computer science , single entry matrix , symmetric matrix , computational science , mathematics , algebra over a field , square matrix , pure mathematics , eigenvalues and eigenvectors , physics , engineering , structural engineering , materials science , quantum mechanics , law , political science , composite material , gaussian
A fast vectorized algorithm is presented for a sparse matrix‐vector multiply. It can be used when the matrix, A , can be represented as a multiplitting, A = ∑ A e . In particular, it can be applied to a matrix‐vector multiply arising in finite element techniques where the matrices A e are associated with the individual element contributions to the global matrix A . The algorithm presented here uses a data structure which is based on the individual matrices A e and can be applied both to symmetric and to non‐symmetric matrices. This algorithm would be attractive for vector architecture similar to either the CYBER 205 or the CRAY and has been implemented for both regular and irregular finite element grids on the CYBER 205. Execution times and storage requirements are compared to standard sparse and band matrix‐vector multiply algorithms.