论文信息 - Sparse matrix vector multiplication techniques on the IBM 3090 VF

Sparse matrix vector multiplication techniques on the IBM 3090 VF

This article demonstrates the benefits of applying Fortran programming techniques based on outer loop vectorization and the multiply-and-add compound instruction to the implementation of several sparse matrix vector multiplication algorithms on the IBM 3090 VF. The discussed algorithms use the scalar and vector ITPACK storage schemes and some variants of them. Their performance is analysed by comparing the execution rates and the storage requirements for several test matrices from the Harwell-Boeing collection.

A. Peters

[1] Y. Saad,et al. Krylov Subspace Methods on Supercomputers , 1989 .

[2] S. G. Tucker,et al. The IBM 3090 System: An Overview , 1986, IBM Syst. J..

[3] John G. Lewis,et al. Sparse matrix test problems , 1982, SGNM.

[4] Bowen Liu,et al. Programming in VS Fortran on the IBM 3090 for maximum vector performance , 1988, Computer.

[5] Rami G. Melhem,et al. Parallel solution of linear systems with striped sparse matrices , 1988, Parallel Comput..

[6] Randolph G. Scarborough,et al. A Vectorizing Fortran Compiler , 1986, IBM J. Res. Dev..

[7] P. Girdinio,et al. A new storage scheme for an efficient implementation of the sparse matrix-vector product , 1989, Parallel Comput..

[8] David M. Young,et al. ITPACK project: Past, present, and future , 1984 .

[9] Helmut Weberpals. Architectural approach to the IBM 3090E vector performance , 1990, Parallel Comput..

[10] G. V. Paolini,et al. Data structures to vectorize CG algorithms for general sparsity patterns , 1989 .

[11] Niel K. Madsen,et al. Matrix Multiplication by Diagonals on a Vector/Parallel Processor , 1976, Inf. Process. Lett..