t8code - scalable and modular adaptive mesh refinement

Abstract

t8code is a versatile open source library for parallel adaptive mesh refinement on hybrid meshes. [1] It is exascale-ready and capable of efficiently managing meshes with up to a trillion elements distributed on a million of cores as already shown in a peer-reviewed research paper. [2] On the top-level, t8code uses forests of trees to represent unstructured meshes with complex geometries. Space-filling curves index individual elements within a forest, which requires only minimal amounts of memory allowing for efficient and scalable algorithms of mesh management. In contrast to existing solutions, t8code has the capability to manage an arbitrary number of tetrahedra, hexahedra, prisms and pyramids within the same mesh. With this poster we want to present the first official release (v1.0) of our software and give a quick overview over its main features. Besides presenting the core algorithms of t8code, we give application scenarios on how our library integrates into major simulation frameworks for weather forecasting, climate modeling and engineering; and how they benefit from our approach to do AMR. [1] https://github.com/DLR-AMR/t8code [2] https://epubs.siam.org/doi/abs/10.1137/20M1383033