Inversion of symmetric matrices in a new block packed storage
by
Gabriela Georgieva
Center of Applied Mathematics and Infromatics, University of Rousse, 7017 Rousse, Bulgaria
Coauthors: Plamen Yalamov (University of Rousse), Fred Gustavson (IBM Research Center)

In LAPACK we have two types of subroutines for solving problems with symmetric matrices: with full and packed storage. The performance of the full storage scheme is much better because it allows the usage of BLAS Level 2 and 3, while the memory requirements for the packed scheme are about twice less. Recently a new storage scheme was proposed which combines the advantages of both schemes: it has a performance similar to that of full storage, and the memory requirements are a little bit higher than for packed storage. In this paper we apply the scheme for inversion of symmetric indefinite matrices.

Date received: February 8, 2000