Research Of Granular Discrete Element’s Calculation Method Based On GPU

The discrete element method calculation object is discretized into a large number of independent units, interaction and define the unit. Discrete element method can solve many practical problems, such as analog mudslides, landslides and other geological disasters, solve DEM problem for traditional computer systems, there are limitations. Especially in large-scale DEM calculation, the slow development in recent years.This article is a GPU-based parallel computing of large-scale DEM. Currently, parallel computing is the main way to improve the performance of the algorithm, whether it is a large-scale cluster system or PC, or even a smart phone, performance improvements are in the hands of parallel computing. Mainly used in computer visualization and graphics, image processing, GPU chip, having a large number of parallel computing units, there is a natural parallel advantage.Parallel computing in large-scale discrete element method has an important significance. The realization of the entire system by the CPU and GPU parallel computation process in this article is based on the discrete element of the large particles is based on the GPU platform for CUDA programming tool to complete the parallel collision detection and profile data including particles with triangular section interface Analyzing to achieve, for the pre-calculated data preparation and data output is done by the CPU, the main contact judgments and mechanical calculations are done by the GPU. The number of particles can be reached one million level, calculated by comparison with the serial system speedup.Through a variety of methods has been optimized for parallel computing system through the use of the three main kernel function NVIDIA Profiler on parallel computing system analysis, using different optimization methods for the function of the type of structure, in the particle-particle contact determine the kernel function, the neighbor search is the key, the grid method of calculation to determine the optimal parallel mode meshing optimized data structure applicable to the consolidation of data memory access, and shared storage mode access conflicts to optimize. Mechanics algorithm optimized to reduce the internal branch, before optimization and optimization of results and analysis of the performance of the optimized ratio displayed on the computing performance of the entire process in an optimized stability and advantages With the increase in the number the more obvious characteristics, further evidence of the effectiveness of the GPU discrete element method, stability, efficiency.