Research On Two-Dimensional Hydrodynamic Parallel Model Based On GPU

Two-dimensional hydrodynamic model has been applied in the dam-break flood, navigation flow computation and other fields of water simulation. With the increase of simulation tasks and precision requirement, long-time calculation has become the bottleneck of model application. The modern computer technology to improve the computing efficiency of hydrodynamic model has become urgent demand of model application. This paper researched two-dimensional hydrodynamic model parallel computation method based on GPU, took the Songhua River Pangtoupao flood storage area and the upper Chongqing channel of the Yangtze River flow computation as research case, the parallel model efficiency and key influence factors were discussed, the mainly research achievements and results were as follows:(1) Researched solving method of two-dimensional hydrodynamic model and established two-dimensional hydrodynamic model based on the finite volume method to adapt research area flow complexity. The principle and method of OpenACC API standard parallel method were summarized, expounded OpenACC parallel standard and parallel process, researched parallel implementation of different directives and tested multifunctional clause in parallel process. The two-dimensional hydrodynamic serial computation model was reformed to parallel computation model based on OpenACC mode, analyzed key technologies data environment control, data transmission and efficiency optimization in model establishment.(2) The two-dimensional hydrodynamic parallel computation model was applied to the upper Chongqing channel of the Yangtze River flow real-time computation, compared model computing results to three different measured instantaneous flows. The parallel model can simulate nicely navigation flow conditions of different water regime. The acceleration performance of parallel model was contrasted through different mesh precision. The computing results showed that the acceleration effect is more obviously with greater computing amount. As the amount of grid was 253161, a speedup factor of 27.7× can be achieved through data transmission optimization. Based on two-dimensional hydrodynamic parallel computation model, a scheme base of flow simulation results was established to stored channel flow simulation results of different flow conditions, combined parallel computation model to improve real-time performance of flow simulation.(3) The two-dimensional hydrodynamic parallel computation model was applied to the Songhua River Pangtoupao flood storage area flow simulation, tested parallel model on Nvidia Kepler K20c card through four different precision mesh schemes. The results showed the parallel model computing efficiency improved greatly on the premise of guaranteeing the accuracy, a speedup factor of 20.7x can be achieved. Based on the same computing tasks, compared OpenMP and OpenACC two parallel method and summarized realization characteristics, computing performance and applicable conditions of parallel method.