GPU optimization method for immersed boundary high-precision statistical dynamics fluid simulation

Abstract

The invention provides a GPU optimization method for immersed boundary high-precision statistical dynamics fluid simulation. The method is wide in application prospect, and mainly has the following aspects: firstly, relatively high-precision real-time simulation is realized in a low-resolution scene, and interactive preview design can be realized; and secondly, in a large scene with relatively high resolution, the expandability of multiple nodes and multiple GPUs is fully utilized, so that the simulation can also be completed very efficiently, and great help is provided for the popularizationof the method in practical application in multiple fields, such as the establishment of a large-scale high-performance industrial design platform and the like.

Description

technical field [0001] The present invention relates to a single-GPU and multi-node multi-GPU large-scale parallel computing acceleration optimization method based on the submerged boundary high-order adaptive statistical dynamics method of the lattice Boltzmann equation, so as to realize large-scale computing with both precision and efficiency. Turbulent flow simulation and fluid-solid coupling simulation belong to the interdisciplinary technical field of fluid dynamics simulation, computer graphics and massively parallel computing. Background technique [0002] Fluid simulation is very important for industrial product design, medical diagnosis, visual animation, environmental protection, energy and other fields. Traditional fluid simulations usually solve the incompressible Navier-Stokes equations (INSE) with different types of discretization schemes, such as finite difference, finite volume, and finite element methods, and have been improved in different ways. Although t...

AI Technical Summary

Problems solved by technology

[0004] However, a direct implementation on GPU of a suite of statistical dynamics solvers based on ACM-MRT models does not achieve optimal performance

In some previous works, there are some attempts to improve the parallel performance of the statistical dynamics solver. The optimization mainly focuses on the rearrangement of the memory data layout and the adjustment of the calculation order, but these optimization schemes are all aimed at the very old BGK-SRT model or traditional RM-MRT model, it is difficult to achieve higher accuracy and stability of fluid simulation

Furthermore, among statistical dynamics methods, there is no GPU optimization work to specifically improve the performance of the immersed boundary-lattice Boltzmann method to support coupled simulations of fluid interiors with various complex static or dynamic solids

Images