Efficient KD-Tree Based Mesh Redistribution for Data Remapping Algorithms
摘要
In this paper, we present a new mesh redistribution algorithm developed for the parallel data remapping library Portage. During distributed memory parallel remapping, source and target meshes are partitioned independently of each other, requiring a mesh redistribution so that all cells on the target mesh partition are covered by source mesh partition cells. Our new algorithm uses a KD-tree data structure to capture the general shape of the target mesh and find an improved overlap between source and target mesh partitions so as to redistribute fewer cells. We present numerical results showing that the KD-tree method reduces memory storage requirements for the redistributed mesh on each partition and is faster than the old bounding box method.