515 / 2024-06-24 16:40:27

A High-performance Hydrodynamic Model Based on Taichi Parallel Programming Language

Two-dimensional hydrodynamic model; Parallel computing; Cross-platform; Taichi programming language

Subtheme 1: Water-related Hazard and Risk Reduction > Topic 1: Water Hazard Under Climate Change

This paper elaborates on the high-performance two-dimensional hydrodynamic model developed by one of IWHR team based on the Chinese Taichi programming language—FRAS, FRAS possesses flexible parallel computing capabilities, is compatible with both homogeneous CPU-CPU and heterogeneous CPU-GPU computing architectures and supports a variety of parallel acceleration technologies including multi-core CPUs, CUDA, OpenGL, Metal, and Vulkan. Moreover, the model can operate smoothly on Nvidia GPUs as well as some GPUs produced by China, demonstrating its exceptional cross-platform performance. This research employs an unstructured grid to discretize two-dimensional space and utilizes the finite volume method for the numerical discretization of the continuity equation and momentum equation. By applying the FRAS model to the flood calculation of the Raoyang River in Liaoning Province, it was found that compared to the original serial code, parallel processing achieved an acceleration ratio of up to 14.7 times. The data structure AOS of cells and edges of cells can doubled the speed of calculation, up to 30 times compared to the original serial code.