A GPU-accelerated Boussinesq Wave Model For Coastal Porous Structure

The Boussinesq-type nearshore wave propagation numerical model has superior numerical calculation ability in the nearshore area.With the development of related theories and numerical methods,this kind of model is widely used in coastal and marine engineering.Nowadays,infiltration structures are a very important part of coastal engineering design and construction,which can effectively reduce the erosion effect of waves on the coast.The momentum equation is introduced into the traditional Boussinesq-type water wave numerical model,which makes it suitable for simulating wave propagation and deformation on permeable structures composed of sand and gravel.By simulating the propagation of waves on porous structures,it has important reference value for the design and construction of coastal engineering.As a phase-analytical numerical model,Boussinesq-type offshore wave propagation numerical model often requires a long calculation time to ensure the calculation accuracy and time resolution when solving in the time domain.In order to improve the computational efficiency of such numerical models,some people have introduced graphics processing unit(GPU)parallel technology and parallel algorithms to such models,and have achieved considerable acceleration effects.With the development of GPU hardware technology and related acceleration algorithms,algorithms such as shared memory and multi-stream computing are applied to the water wave numerical model,which further accelerates the calculation of the model.In order to further improve the computational efficiency,it is necessary to explore and implement a multi-GPU parallel computing method for this kind of numerical model on the basis of the existing Boussinesq-type nearshore wave propagation numerical model calculated on a single GPU.Compared with the previous numerical model of single-GPU computing,multi-GPU computing can reduce the load of a single GPU and achieve a larger number of computing grids.This paper firstly verifies the Boussinesq water wave numerical model after introducing the momentum equation.By comparing the numerical simulation results of different dimensions and different test terrains with the experimental results,it is found that the model can better reflect the wave propagation on permeable structures.Transmission,reflection,and diffraction phenomena that occur.The computational efficiency is improved by optimizing the parallel algorithm of the existing Boussinesq numerical model for single-GPU computing.Use shared memory in GPU devices,use reduction algorithms in programs,and utilize different CUDA stream computations to speed up programs.Two different working conditions are used to verify the numerical model after using the above parallel acceleration algorithm,and the acceleration ratios under different working conditions are calculated.Through statistics,it is found that compared with the numerical model of the CPU version,the speedup ratios are 55.56 and 32.57 under the conditions of single precision and double precision,respectively.This paper then realizes the multi-GPU parallel computing of Boussinesq water wave model based on the multi-GPU parallel computing method.Using Open MP technology,multi-thread cooperative control of different GPUs is realized,and numerical calculations in different computing areas are realized on their respective GPUs,and the reasons for the delay of multi-GPU model programs are analyzed through performance analysis software.