Multi-layer rapid multi-pole parallel grid fine cutting method based on an auxiliary tree

Abstract

The invention discloses a multi-layer rapid multi-pole parallel grid fine cutting method based on an auxiliary tree. According to the invention, an auxiliary tree is constructed based on a coarse gridtriangular unit center; a discrete mode with the same structure as that of a multi-pole sub-tree for calculation is adopted for the auxiliary tree; layered dispersion is carried out on the process number. A finest layer is traversed according to a box parallel first layer discrete mode. distribution mapping is established between the triangular units and the processes for calculation, and the triangular units, edges and vertexes are rearranged to form coarse grid files with continuous partitions. Secondly, the coarse grid files are read in parallel, the coarse grids are uniformly, consistently and finely sectioned in each process, newly-generated edges, points and triangles are renumbered to form complete fine grid information, the locality of data can be kept to the maximum extent, the communication data size is reduced, and therefore high parallel efficiency is achieved.

Description

technical field [0001] The invention belongs to the research field of computational electromagnetic calculation, and in particular relates to a multi-layer fast multi-pole parallel grid fine-section method based on an auxiliary tree. Background technique [0002] The method of moments is an exact algorithm in computational electromagnetism that solves integral equations. It is especially suitable for solving open-domain problems, such as scattering and radiation problems, because the radiation boundary conditions are automatically satisfied. The iterative solution of the final matrix equation of the method of moments can be accelerated using multilevel fast multipole techniques. The multi-layer fast multipole technology divides the target into multi-layer boxes, and realizes matrix-vector multiplication through aggregation, transfer, and divergence in a grouped and hierarchical manner. With the help of efficient parallel computing technology, the calculation scale of the m...

AI Technical Summary

Problems solved by technology

Considering the array of intermediate variables, the memory required for a single grid generation is close to or even exceeds 1TB. Whether it is hardware resources or computing time, serial programs are difficult to implement

[0010] (2) Difficulty reading

Restricted by the operation of rearranging the edges in multipoles, the implementation of mesh refinement cannot adopt the general decomposition mode based on the target geometry map partition (metis), otherwise the unreasonable initial mesh discretization will lead to a large number of Communication

Images