Thönnes D, Kohl N, Drzisga D, Bartuschat D, Rüde U (2019)
Publication Language: English
Publication Type: Conference contribution, Conference Contribution
Publication year: 2019
Event location: Spokane, Washington
URI: https://www10.cs.fau.de/publications/talks/2019/Thoennes_Spokane_SIAMCSE19_2019-02-25.pdf
Multigrid methods are an important class of iterative solvers and preconditioners as they provide fast convergence rates with O(n) complexity on specific classes of problems such as elliptic problems with uniform coefficients. While their implementation on distributed memory system has been shown to scale well both weakly and strongly, the advent of new mixed architectures requires a redesign of some core kernels of these methods. This minisymposium aims at exploring methods leveraging fully or partially structured grids: fully structured grids, block-structured grids, mixed structured/unstructured grids or nested grids. Exploiting the regularity and the predictable indexing of the data leads to new more parallel and scalable algorithms. This minisymposium also focuses on the parallel shared memory implementations of these algorithms for mixed architectures as it is an increasingly important aspect of high performance computing on recent architectures aiming at scalability at exascale.
APA:
Thönnes, D., Kohl, N., Drzisga, D., Bartuschat, D., & Rüde, U. (2019). HyTeG: A High Performance Multigrid Framework on Hybrid Meshes. In Proceedings of the SIAM Conference on Computational Science and Engineering (CSE19). Spokane, Washington, US.
MLA:
Thönnes, Dominik, et al. "HyTeG: A High Performance Multigrid Framework on Hybrid Meshes." Proceedings of the SIAM Conference on Computational Science and Engineering (CSE19), Spokane, Washington 2019.
BibTeX: Download