|
|
| |
| 基于GPU并行的一种快速常规态近场动力学算法 |
| A state based peridynamic algorithm based GPU parallelism |
| 投稿时间：2024-10-22 修订日期：2024-11-22 |
| DOI： |
| 中文关键词: 常规态近场动力学 GPU并行 CUDA编程技术 |
| 英文关键词:Ordinary state based peridynamics GPU parallelism CUDA programming technology |
| 基金项目:国家自然科学基金项目（面上项目，重点项目，重大项目） |
|
| 摘要点击次数: 55 |
| 全文下载次数: 0 |
| 中文摘要: |
| 本文利用 GPU 并行的 CUDA 编程技术，对近场动力学的并行求解方法展开了研究，计了一个成本低、性能高的近场动力学模拟框架PD-General。该分析框架能够以高效的计算速率，准确地模拟常规态近场动力学问题。相比串行程序和 OpenMP 并行程序，PD-General 分别取得了最大 800 倍和最大 100 倍的加速比。在常规百万级别的粒子模拟中，执行 4000 步迭代，单精度下可以在 5分钟内完成计算，双精度下可以在 20 分钟内完成模拟。在最大规模的探索上，程序实现了在单精度下，基于常规态近场动力学解决 1000 步规模为 69,856,200 迭代的二维单轴拉伸问题的计算，仅需两分钟。 |
| 英文摘要: |
| This paper utilizes the parallel CUDA programming technology of GPU to study the parallel solution method of peridynamics. A cost-effective and high-performance peridynamics simulation framework, PD-General, has been designed. This analytical framework can accurately simulate ordinary state based peridynamics problems at an efficient computational rate. Compared with serial programs and OpenMP parallel programs, PD-General has achieved a maximum speedup of 800 times and 100 times, respectively. In conventional million-level particle simulations, performing 4000 iterations, the computation can be completed within 5 minutes in single precision and within 20 minutes in double precision. In the largest scale exploration, the program has achieved a computation of 69,856,200 iterations for a 1000-step two-dimensional uniaxial tension problem based on the ordinary state based peridynamics in single precision, taking only two minutes. |
| 查看/发表评论 下载PDF阅读器 |
