Acceleration and Optimization of Thompson Cloud Microphysics Scheme Based on Domestic Accelerator
Year: 2025
Type: conference-paper
Abstract: Thompson scheme is a microphysical parameterization scheme used in Weather Research and forecast Model (WRF) to describe the cloud physical process and mainly simulates the evolution of cloud and precipitation; however, there are some problems in high-resolution meteorological simulation, such as complex calculation of physical processes, large amount of data processing and slow operation speed, which significantly reduce the computational efficiency and processing capacity of the model. In order to accelerate the calculation speed of the scheme on the domestic heterogeneous computing platform, a new Thompson acceleration version is designed according to the architecture characteristics of the domestic accelerator. Based on WRF V4.5.2, the OpenACC instruction is embedded in Thompson scheme to parallelize it, and optimization methods such as GPUFORT transcoding, kernel function decomposition and merging, and optimization of thread block size are adopted to improve the overall computing performance of the scheme on domestic heterogeneous computing platform. The experimental results show that under the premise that the simulation results of the accelerated version are consistent with the results of the original CPU version, the calculation speed of the Thompson accelerated version on the domestic accelerator is increased by 17.20 times, and the calculation speed on the NVIDIA GPU accelerator is increased by 19.28 times, which has a good acceleration effect. The accelerated version of the Thompson physical parameterization scheme for the domestic heterogeneous computing platform is realized in the research. The acceleration method and optimization strategy adopted provide a certain reference for the acceleration and optimization of other physical parameterization schemes in the WRF model on the domestic heterogeneous computing platform.
Authors: Ziyi Ma, Danqin Ren, Xiguo Xie
Institution: Zhengzhou University
Language: English
Subfield: Hardware and Architecture
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4413812410