Fast Givens goes slow in MATLAB

Abstract

Numerical results show that a straightforward implementation in MATLAB of an algorithm for the Fast Givens QR Factorization of an m x n matrix takes more time than an implementation of an algorithm that uses ordinary Givens transformations. Results of timing experiments for various operations in MATLAB are reported and models are constructed to explain why the standard theoretical analysis based on flop counts does not account for the actual times for the algorithms.