A GPU-based heterogeneous acceleration method for fluid mechanical simulation programs

Abstract

The invention discloses a GPU (Graphics Processing Unit)-based fluid machinery simulation program heterogeneous acceleration method. The method comprises the following steps of: analyzing a hot spot,and finding a subprogram with an acceleration potential; Reducing and avoiding data transmission between the host devices; Through combination of memory access and utilization of multiple types of memories, the effective memory utilization rate is increased, and the memory access ratio is calculated; Code reconstruction exposes data parallelism, GPU kernel explicit global synchronization lossy data parallelism should be avoided as much as possible; GPU adaptation of the serial algorithm is carried out, and the serial algorithm is replaced with a parallel algorithm with the same function; Thread distribution parameters are adjusted, thread calculation delay is fully hidden by adjusting thread distribution, and calculation throughput is improved; If the ideal effect is achieved through the steps, acceleration is completed, and otherwise, new round of iteration is started from hot spot analysis until the satisfactory effect is achieved. The invention provides the GPU acceleration method aiming at the characteristics of the fluid machinery simulation program, and the modified program can achieve an ideal acceleration effect.

Description

technical field [0001] The invention belongs to the intersection field of fluid mechanics and high-performance computing, and in particular relates to a heterogeneous acceleration method for a fluid mechanical simulation program based on GPU. Background technique [0002] Computational fluid dynamics is one of the important technologies in the field of fluid mechanics. By using numerical methods to solve the control equations of fluid mechanics in the computer, the flow of the flow field can be predicted. With the improvement of computer computing power, more and more refined fluid dynamics models can be constructed. At the same time, in order to calculate the flow field changes of fluid mechanics more accurately and make the flow field movement more "real", the fluid mechanics model is becoming more and more refined, which also puts forward higher requirements for computing power. [0003] As the transistor density and frequency of single-core chips reach the bottleneck, m...

AI Technical Summary

Problems solved by technology

Traditional fluid machinery programs are optimized for CPU architectures. With the rise of heterogeneous computing, the differences between coprocessor architectures and CPU architectures are increasing. The traditional fluid machinery simulation program writing methods and optimization schemes can no longer be directly compared with New hardware architecture adaptation

Images