A Proposal for User Level Sparse BLAS

I. Duff
Rutherford Appleton Laboratory
Oxfordshire, England and
Toulouse, France

M. Marrone and G. Radicati
IBM Semea
Cagliari, Italy


This paper describes a set of Level 3 Basic Linear Algebra Subprograms for sparse matrices. We discuss the design, implementation and use of subprograms for the multiplication of a full matrix by a sparse one and for the solution of triangular systems with one or more full right-hand sides. The routines are designed to accommodate many different data structures. We include routines for transforming sparse data structures and for permuting sparse and full matrices. The major aim of this exercise is to establish standards to enable effcient, and portable, implementations of iterative algorithms for sparse matrices on high performance computers. We have tried to keep the interface and suite of codes as simple as possible while at the same time including suffcient functionality to cover most of the requirements of iterative solvers, and sufficient flexibility to cover most sparse matrix data structures. This document constitutes a proposal for standards in the above areas and the authors urgently solicit comments and suggestions on this draft. It is intended to revise this draft and to supply Fortran code for the revision before October. This proposal is intended to be complementary to the paper Level 2 and Level 3 Sparse BLAS Toolkit by Michael Heroux.

Contributed January 8, 1993.