| Line 2: | Line 2: | ||
==Abstract== | ==Abstract== | ||
| − | We present data-parallel approaches to solve radial-basis function interpolation problems in the context of partitioned multi-physics simulations, where interpolation | + | We present data-parallel approaches to solve radial-basis function interpolation problems in the context of partitioned multi-physics simulations, where interpolation methoods are required to transfer coupling data between non-matching vertex clouds. Data-parallel approaches are a key component for the efficient use of accelerator cards and thus for performance portability on modern compute platforms. The presented approach is integrated into the open-source coupling library preCICE. |
After discussing different implementation strategies, we introduce a solution based on thelinear algebra library Ginkgo, which provides a common abstraction layer for cross-platform performance with focus on solving sparse linear systems. The new implementation exploits accelerator cards for both, matrix assembly as well as solving the resulting linear system. The capability of the presented approach is compared to already existing implementations in preCICE using a turbine blade geometry | After discussing different implementation strategies, we introduce a solution based on thelinear algebra library Ginkgo, which provides a common abstraction layer for cross-platform performance with focus on solving sparse linear systems. The new implementation exploits accelerator cards for both, matrix assembly as well as solving the resulting linear system. The capability of the presented approach is compared to already existing implementations in preCICE using a turbine blade geometry | ||
== Full Paper == | == Full Paper == | ||
<pdf>Media:Draft_Sanchez Pinedo_447867399pap_148.pdf</pdf> | <pdf>Media:Draft_Sanchez Pinedo_447867399pap_148.pdf</pdf> | ||
Latest revision as of 13:05, 2 November 2023
Abstract
We present data-parallel approaches to solve radial-basis function interpolation problems in the context of partitioned multi-physics simulations, where interpolation methoods are required to transfer coupling data between non-matching vertex clouds. Data-parallel approaches are a key component for the efficient use of accelerator cards and thus for performance portability on modern compute platforms. The presented approach is integrated into the open-source coupling library preCICE. After discussing different implementation strategies, we introduce a solution based on thelinear algebra library Ginkgo, which provides a common abstraction layer for cross-platform performance with focus on solving sparse linear systems. The new implementation exploits accelerator cards for both, matrix assembly as well as solving the resulting linear system. The capability of the presented approach is compared to already existing implementations in preCICE using a turbine blade geometry
Full Paper
Document information
Published on 02/11/23
Submitted on 02/11/23
Volume Multi-Physics and Multi-Scale Simulations with the Coupling Library preCICE, 2023
DOI: 10.23967/c.coupled.2023.016
Licence: CC BY-NC-SA license
Share this document
claim authorship
Are you one of the authors of this document?