基于国产编程语言的并行水动力模型开发及初步调优OA
Parallel Hydrodynamic Model and Preliminary Performance Tuning Based on a Domestic Programming Language
基于国产编程语言Taichi开发了具有跨平台并行计算能力的高性能二维水动力模型FRAS.FRAS 拥有良好的并行计算灵活性,能与同构CPU-CPU和异构CPU-GPU计算架构良好兼容,还支持多核CPU、CUDA、OpenGL、Metal、Vulkan等多种并行加速技术,具有良好的跨平台性能.采用非结构化网格离散二维空间,运用有限体积法对连续性方程和动量方程进行数值离散处理,将FRAS模型应用于辽宁省绕阳河的洪水计算,相较于原始串行代码,并行化处理后加速比为 14.7.通过优化变量存储结构,计算性能因访存优化而提升约 2 倍,初步优化后程序加速比达 30.1.
A high-performance two-dimensional hydrodynamic model FRAS was developed using the domestic programming language Taichi,featuring cross-platform parallel computing capability.FRAS provides excellent parallelization flexibility and is well compatible with both hom-ogeneous CPU-CPU and heterogeneous CPU-GPU computing architectures.It supports multiple parallel acceleration backends,including multi-core CPUs,CUDA,OpenGL,Metal,and Vulkan,demonstrating outstanding cross-platform performance.In this study,the two-dimensional domain is discretized with an unstructured mesh,and the continuity and momentum equations are numerically discretized using the finite volume method.FRAS was applied to flood simulations of the Raoyang River in Liaoning Province.Compared with the original serial code,the parallel implementation achieves a speedup of up to 14.7 times.By optimizing the data layout of variables,computational perform-ance is further improved by approximately 2 times due to enhanced memory locality,resulting in an overall speedup of about 30.1 times after preliminary optimizations.
王明阳;王静;李娜;俞茜;宫啸天
中国水利水电科学研究院,北京 100038||水利部防洪抗旱减灾工程技术研究中心,北京 100038中国水利水电科学研究院,北京 100038||水利部防洪抗旱减灾工程技术研究中心,北京 100038中国水利水电科学研究院,北京 100038||水利部防洪抗旱减灾工程技术研究中心,北京 100038中国水利水电科学研究院,北京 100038||水利部防洪抗旱减灾工程技术研究中心,北京 100038通辽水文水资源分中心,内蒙古 通辽 028000
建筑与水利
二维水动力模型并行计算跨平台Taichi编程语言
two-dimensional hydrodynamic modelparallel computingcross-platformTaichi programming language
《人民黄河》 2026 (2)
41-46,6
国家重点研发计划项目(2022YFC3006400,2022YFC3006403)国家自然科学基金资助项目(U2340225)
评论