Solving 3D Block Bidiagonal Linear Systems on Vector Computers

摘要

The authors discuss different techniques for the vectorization of the solution of 3D-block bidiagonal systems. They report on actually observed performances for the ICCG algorithm, for which these bidiagonal systems have the reputation to spoil the overall performance, on different vector computers. The potentially most powerful of the vectorization techniques leads to long vector operations, at the post, however, of strides and indirect addressing. Since the CYBER 205 is generally believed to stay behind in performance under such circumstances, they have chosen that machine to show in detail how these vectorization techniques can be implemented with almost equal performance as in the contiguous vector case. The methods are directly applicable to the ETA-10 family of supercomputers and may be adapted to other vector computers as well. (Copyright (c) 1989 by Faculty of Technical Mathematics and Informatics, Delft, The Netherlands.)