An unstructured grid many-core coarseness parallel computing method

Abstract

The invention discloses an unstructured grid many-core coarseness parallel computing method. According to the method, on the basis of region decomposition of a first-level non-structural grid, second-level thread level region decomposition is added, each slave core solves a respective independent calculation region, the data hit rate of a slave core calculation task is ensured, and coarse-grainedparallelism of an MPI process level and coarse-grained parallelism of a slave core thread level are achieved. According to the method, the problem of adaptability of a common non-structural grid applied to a polymorphic heterogeneous processor can be solved, secondary load balance and coarse-grained many-core parallelization of a calculation core are automatically completed according to the scaleof non-structural grid data, and the calculation efficiency and the parallelization efficiency of numerical simulation of the non-structural grid on a heterogeneous system structure are improved.

Description

technical field [0001] The invention relates to the field of unstructured grid flow field solution, in particular to a non-structured grid many-core coarse-grained parallel computing method. Background technique [0002] In recent years, in order to improve the computing power of the system, multi-core and many-core processors have become the main components of high-performance computers. Efficient parallel algorithms for matching are lacking. With the deepening of user research, the calculation area and topological structure of numerical simulation of practical problems are becoming more and more complex and refined. Due to its excellent flexibility and super adaptability to complex shapes, unstructured grids are increasingly used in engineering calculation software. more extensive. But on the other hand, because the unstructured grid discards the structural restrictions of the grid nodes, resulting in disorder in data storage, the bottleneck problem of memory access is m...

AI Technical Summary

Problems solved by technology

However, the memory access characteristics of most core computing segments of unstructured grids are discontinuous. If the previous many-core parallel method is used to directly perform many-core parallelization on the core cycle, the performance will be low due to discontinuous memory access.

[0004] In summary, most of the current unstructured grid applications based on the structural characteristics of heterogeneous many-core processors use process-level coarse-grained parallelism and thread-level fine-grained parallelism. However, due to the flexibility and disorder of unstructured grids, There are a large number of irregular discrete memory accesses in the calculation process, the efficiency of many-core parallelism is low, and performance problems are prominent

Images