Analysis of incomplete matrix factorizations as multigrid smoothers for vector and parallel computers

摘要

The cost of smoothing is usually a major expense in multigrid codes. Efficient vectorizable and parallelizable versions of incomplete blockmatrix factorization methods used as smoothers for multigrid methods are discussed in this paper. The methods are particularly interesting for computers with parallel processors with pipelines, because both multitasking with little overhead and vectorization can be achieved.