Unstructured Grid Data Management Approach Matching High Performance Computer Architecture

Abstract

This paper provides an unstructured grid data management method that matches the high-performance computer architecture. The method includes: determining the unstructured grid data structure according to the multi-physics coupling process; mapping the unstructured grid data structure to the high-performance computer system In the structure; the process of determining the data structure of the unstructured grid includes: constructing a federation layer composed of various unstructured grid units; Cloning is performed to obtain the grid layer and the adjacency relationship between the grid layers; the grid layer is subdivided layer by layer without overlapping to obtain the layer-by-layer nested subdivision layer and the adjacency relationship between the subdivision layers; for the final The subdivision layer obtained by one subdivision defines the physical quantities in the multi-physics coupling process for the grid unit. This article can support arbitrary unstructured grid cell types and complex topological structures, which can greatly improve the numerical calculation efficiency.

Description

technical field [0001] This paper relates to the field of scientific and engineering computing, especially an unstructured grid data management method matching the high-performance computer architecture. Background technique [0002] In the field of scientific and engineering computing, unstructured grid is a kind of discrete grid in the calculation area, which can effectively support complex numerical simulation. At present, the high-performance computer architecture is becoming more and more complex, presenting the typical characteristics of "computer-node-processor-core-vector component-pipeline six-layer nested parallelism" and "heterogeneous many-core acceleration". In order to fully adapt to the architecture and greatly improve the efficiency of parallel computing based on unstructured grid, it is necessary to design an unstructured grid data structure that matches the current high-performance computer architecture. [0003] Traditionally, numerical simulations use mu...

AI Technical Summary

Problems solved by technology

[0005] This article is used to solve the problem that the unstructured grid data structure in the prior art is not suitable for the existing high-performance computer architecture, and there are low access locality, a decrease in the cache hit rate, and it is not conducive to mining the instruction level of the multi-functional components in the CPU core. Parallelism and vector acceleration functions and problems that cannot be adapted to heterogeneous many-core acceleration

Images