Past Event: Oden Institute Seminar

Graph Grammar-based Parallel Direct Solver for 1D and 2D Isogeometric Finite Element Method

Maciej Paszynski, Department of Computer Science, AGH University of Science and Technology, Krakow, Poland

Abstract

We present a graph grammar based implementation of a multi-frontal solver with NVIDIA CUDA GPU for isogeometric finite elements. The presentation will focus on numerical illustration and verification of computational cost estimates for a parallel shared memory implementation: O(p^2log(N/p)) for 1D , O(Np^2) for 2D, and O(N^1.33 p^2) for 3D problems. For comparison, the corresponding estimates for a standard sequential implementation are: O((N/p)p^2), O(N^1.5p^3) and O(N^2p^3), respectively. We conclude with a discussion on practical limitations of such GPU implementations. The work was done in collaboration with Krzysztof Kuznik (AGH), Victor Calo (KAUST) and David Pardo (U Basque Country).