New Multilevel Parallel Preconditioning Strategies of Sparse Matrix for Speeding up CAD System

Abstract

ABSTRACTIn the modeling and transformation research of 3D human body organs and tissues based on medical images, according to 3D scene characteristics of huge amount volume in spatial data processing, the parallel computing of large sparse matrices has important significance for performance calculation of speeding up CAD system. In this paper, to avoid the parallel implementations difficulty encountered in the independent set search strategy of CAD system, a new parallel multilevel MSP (Multistep Successive Preconditioning strategies) pre-conditioner is presented. In each level, we use a diagonal value based strategy to permute the matrix into a 2 by 2 block from. During the preconditioning phase, we do forward and backward preconditioning to improve the performance of the pre-conditioner. In our experiments, MMSP shows better algorithmic CPU time and scalability for solving an actual case of CAD matrix with n = 1003 and nnz = 6940000, and the better convergence behavior of MMSP confirms the scaling of pe...