Computing Colloquium Student Presentation featuring Damyn Chipman

A Coupled Elliptic/Hyperbolic Adaptive Finite Volume Solver for High-Performance Heterogeneous Architectures

Presenter: Damyn Chipman
Computing PhD Student, Computational Math Science and Engineering emphasis

Location: In person in CCP 259 or register to attend via Zoom

Abstract: Coupled elliptic/hyperbolic systems of partial differential equations (PDEs) arise in a variety of science and engineering applications, including fluid solvers. Solving such systems on adaptive meshes allows for increased memory and compute efficiency but introduces additional complexities in implementation. We present our work on coupling the Hierarchical Poincaré-Steklov (HPS) method, a fast and direct elliptic solver (Gillman and Martinsson, 2014), with the hyperbolic solvers implemented in the ForestClaw software (Calhoun and Burstedde, 2017). The underlying systems are solved on a dynamic quadtree mesh as implemented in the p4est library (Burstedde, et al., 2011). This includes targeting high performance, heterogeneous CPU/GPU architectures for distributed memory parallelism using our new workload sharing implementation of the HPS method. We will show scaling analysis on the Polaris petascale machine from Argonne National Laboratory and progress on our target application of a non-dispersive tsunami model.