Vol.10, No.2 (1997) 205-213
Abstract: This paper presents a better way of reformulating matrix multiplication as polynomial multiplication and convolution. If the elements of the product matrix are known to be bounded, matrix multiplication and convolution can be done, by using a scaling factor, with a single large-number multiplication. The degrees of polynomials and size of numbers being multiplied are smaller than earlier. Also, adaptation a size of scaling factor and multiplier's word-length with matrix dimensions, makes a number of multiplication smaller, because we know beforehand that some products will be zero. With increasing multiplier's word-length decreases number of multiplication, so we can easily adapt this algorithm to various architectures. The algorithm is especially suitable for implementing with parallel multipliers.
Key words: Matrix multiplication, convolution, polynomial multiplication, parallel multipliers.