Items tagged with sdmp sdmp Tagged Items Feed

Our previous article described the design of fast algorithms for multiplying and dividing sparse polynomials. We have integrated these algorithms into the expand and divide commands of Maple 14. In this post I want to talk a bit about what you might see when you try Maple 14. Keep in mind that the product isn't released yet and I don't work for Maplesoft, so general disclaimers apply. Nevertheless, one of the first things you may notice is this.

In our previous article we described a packed representation for sparse polynomials is designed for scalability and high performance. The expand and divide commands in Maple 14 use this representation internally to multiply and divide polynomials with integer coefficients, converting to and from Maple's generic data structure described here. In this post I want to show you how these algorithms work and why they are fast. It's a critical stepping stone for our next topic, which is parallelization.

Our first article introduced Maple's polynomial data structure and explained how Maple spends a lot of time working with monomials. To multiply polynomials having n and m terms, Maple must construct, simplify, hash, and sort all nm pairwise products to determine what monomials are equal. This work is performed even if the result has far fewer than nm terms, making it a rather inefficient way to multiply large multivariate polynomials. This article describes a new data structure for multivariate polynomials that is being added to Maple for a future release.

9xyz  -  4yz  -  6xyz  -  8x  -  5

This is the first in a series of short informal articles about our efforts to speed up polynomial arithmetic in Maple. We begin with an example of how polynomials are represented in Maple right now.

9xyz  -  4yz  -  6xyz  -  8x  -  5

When you enter a polynomial in Maple, it creates a generic data structure like the one above. In Maple's representation this polynomial is a sum of terms that is 11 words of memory long where each word is either 32 or 64 bits. For each term it stores a pointer to a monomial followed by a coefficient.