Presenters-0613

CSDMS 2023: Patterns and Processes Across Scales

Building solvers for sustainable performance

Jed Brown
University of Colorado, Boulder, United States
jed.brown@colorado.edu

Abstract
Developers of solvers for PDE-based models and other computationally intensive tasks are confronted with myriad complexity, from science requirements to algorithms and data structures to GPU programming models. We will share a fresh approach that has delivered order of magnitude speedups in computational mechanics workloads, minimizing incidental complexity while offering transparency and extensibility. In doing so, we'll examine the PETSc and libCEED libraries, validate performance models, and discuss sustainable architecture for community development. We'll also check out Enzyme, an LLVM-based automatic differentiation tool that can be used with legacy code and multi-language projects to provide adjoint (gradient) capabilities.

Please acknowledge the original contributors when you are using this material. If there are any copyright issues, please let us know (CSDMSweb@colorado.edu) and we will respond as soon as possible.

Of interest for:

Marine Working Group

Terrestrial Working Group

Coastal Working Group

Education and Knowledge Transfer (EKT) Working Group

Cyberinformatics and Numerics Working Group

Hydrology Focus Research Group

Chesapeake Focus Research Group

Critical Zone Focus Research Group

Human Dimensions Focus Research Group

Geodynamics Focus Research Group

Ecosystem Dynamics Focus Research Group

Coastal Vulnerability Initiative

Continental Margin Initiative

Artificial Intelligence & Machine Learning Initiative

Modeling Platform Interoperability Initiative

River Network Modeling Initiative