Scalable Blas 2 and 3 Matrix Multiplication for Sparse Banded Matrices on Distributed Memory Mimd Machines

In this paper, we present two algorithms for sparse banded matrix-vector and sparse banded matrix-matrix product operations on distributed memory multiprocessor systems that support a mesh and ring interconnection topology. We aslo study the scalability of these two algorithms. We employ systolic type techniques to eliminate synchronization delay and minimize the communication overhead among processors. The performance of algorithms developed for the above operations depends on the bandwidth of the matrices involved and have been currently implemented on the NCUBE II with 64 processors. Our preliminary experimental data agree with the expected theoretical behavior.