An international conference connecting people
in CAD research, education and business
Copyright (C) CAD Solutions, LLC. All rights reserved.
Proceedings of CAD'14, 2014, 156-159
New Multilevel Parallel Preconditioning Strategies of Sparse Matrix for Speeding up CAD Systems
Kai Wang, UC Berkeley Lawrence Berkeley National Lab
Abstract. With the more progress of CG software and hardware technology and growing demand of CAD application, the 3D scene objects and the complexity of the model increases rapidly, the realistic demand of rendering was extremely increased, the display resolution appears exponential increase. According to above characteristics of huge amount volume in spatial data processing, researching the problem that applying high performance parallel computing especially non-traditional model to practical solving process of large sparse matrices equations, which will has important significance for performance calculation of speeding up CAD system. As we all know, the linear system may be solved by a direct solver based on a factorization of the sparse matrix (Gauss eliminations), which is known to be robust. However the Gauss eliminations lack of inherent parallelism, and their O(n2) complexity of memory cost and O(n3) complexity of computational cost make them very expensive for solving large problems. So people turn to other methods and try to deal with sparse linear systems by taking the advantage of the sparse structure in the coefficient matrices.
Keywords. Parallel preconditioning strategies, sparse matrix, MSP strategy, multilevel pre-conditioner, 3D spatial data processing, performance of CAD system