GPU-native Dynamic Octree-based Grid Adaptation to Moving Bodies

The present paper is devoted to the development of algorithms for dynamic grid adaptation to the geometry of an immersed moving rigid body and tracking its position on the grid. Hierarchical octree-based mesh adaptation is dynamically performed for a Cartesian base grid so that 2:1 grid cell balancing is strictly maintained in calculations and the body surface is always located in a cloud of grid cells of the lowest adaptation level. The latter means that all cut cells and all their neighbors at each time step belong to the lowest adaptation level. The details of corresponding implementations on multi-core CPUs using OpenMP and GPUs using CUDA, performance results are also provided. According to performed estimation, proposed algorithms brings only 3 \(\%\) overhead to CFD solver runtime. To demonstrate the ability of our implementation to process complex geometry, a DLR F6 aircraft model is used as rigid body in the tests conducted.