Fast overlapped scattered array storage schemes for sparse matrices

Abstract

Several heuristic schemes for storing large sparse matrices in memory are presented. These heuristics exploit the distribution of the zero and non-zero elements of the matrix rather than just the number of non-zeros. Their performance ranges from very high packing density and acceptable processing time to extremely fast but acceptable packing density. The best packing density is comparable to the Ziegler scheme but at only 20% of the CPU time. The fastest scheme is about two orders of magnitude faster than the Ziegler method but achieves only about 60% of its packing density. Example matrices from circuit simulation data illustrate the superiority of the authors' schemes.<>