LU Z W, ZHANG H R, WAN Y Q, et al.GPU parallel computing and optimization of backward-facing step flow field based on flux vector splitting[J].Journal of Henan Polytechnic University(Natural Science) ,2026,45(3):69-76.

doi:10.16186/j.cnki.1673-9787.2024120025

Received:2024/12/11

Revised:2025/06/23

Published:2026/04/28

GPU parallel computing and optimization of backward-facing step flow field based on flux vector splitting

Lu Zhiwei, Zhang Haoru, Wan Yunqian, Wang Yadong, Zhang Zhuokai, Liu Xiyao, Zhang Jun’an

School of Mechatronic Engineering， Xi’an Technological University， Xi’an 710032， Shaanxi， China

Abstract: Objectives To further explore the advantages of the flux vector splitting method in improving computational efficiency，a GPU （graphics processing unit） parallel algorithm based on this method was designed.  Methods The algorithm fully utilized the powerful parallel processing capabilities of the GPU to parallelize each step of the flux vector splitting method.The process of finding the maximum and minimum values was optimized through parallel reduction. Combined with the physical model of a backward-facing step，the two-dimensional compressible Navier-Stokes equations were discretized and numerically solved using the finite difference method. To verify the accuracy and effectiveness of the algorithm， its calculation results were compared with the classic numerical calculation method，MacCormack，and the flow field characteristics were explored.  Results The results showed that under the condition of the same number of grids，the calculation results obtained by the flux vector splitting method and the MacCormack method exhibited a consistent trend，indicating that both methods could accurately simulate the flow field. The speedup ratio of the flux vector splitting  method reached 8.72 when the grid number was 128×128，while that of MacCormack method was 5.51. As the grid number increased，the advantage of the flux vector splitting method in terms of speedup ratio over the MacCormack method became more significant， indicating that the flux vector splitting method exhibits higher efficiency and better scalability when handling large-scale computing tasks，allowing computing tasks to  be allocated more efficiently to multiple CUDA cores and fully utilizing the large-scale parallel processing capabilities of the GPU.  Conclusions This study not only verified the superiority of the flux vector splitting scheme in high-performance computing but also provided an important basis for algorithm selection for future GPU-based scientific computing. It also demonstrated how to use modern hardware architectures to improve the solution speed of complex fluid mechanics problems，which is of great significance for applications requiring rapid and accurate simulations.

Key words:flux vector splitting;parallel computing;complex flow field;numerical simulation;performance optimization