Cache-oblivious spacetree traversals.

Abstract

In scientific computing and related fields, mathematical functions are often approximated on meshes where each mesh cell contains a local approximation (e.g., using polynomials) of the represented quantity (density functions, physical quantities such as temperature or pressure, etc.). The grid cells may adaptively refine within areas of high interest or where the applied numerical algorithms demand improved resolution. The resolution even may dynamically change throughout the computation. In this context, we consider tree-structured adaptive meshes, i.e., meshes that result from a recursive subdivision of grid cells. They can be represented via trees – quadtrees or octrees being the most prominent examples. In typical problem settings, quantities are stored on entities (vertices, edges, faces, cells) of the grid. The computation of these variables is usually characterized by local interaction rules and involves variables of adjacent grid cells only.