Behaviour of Conjugate Gradient Based Algorithms on a Multi-Vector Processor with a Memory Hierarchy

摘要

The conjugate gradient algorithm is one of the most efficient solvers for elliptic partial differential equations and has been successfully implemented on various vector computers. In this paper, the behaviour of conjugate gradient based algorithms on a multi-vector processor with a two-level memory hierarchy is investigated. Several domain decomposition techniques are considered as well as preconditioners that are suitable to be parallelized such as polynomial preconditioners or a vectorized version of the ICCG. Experimental results for the Poisson equation performed on an Alliant FX/8 are presented. 16 refs., 17 figs., 13 tabs. (ERA citation 13:013846)