GPU-based simulation of seismic wave propagation with hybrid PSM/FDM method
-
摘要: 伪谱和高阶有限差分混合方法, 在垂直方向采用交错网格有限差分算子, 利用其并行程度高的特点, 在水平方向采用伪谱算子, 保留其高精度的优势, 是计算地震波场的有效方法. 图形处理器(graphic processing unit, 简写为GPU) 由于其高度并行性, 在计算此类问题中有显著的优势. 由英伟达(NVIDIA)公司推出的统一计算设备架构(compute unified device architecture, 简写为CUDA)平台极大地简化了GPU编程的难度. 为提高计算效率, 本文实现了基于CUDA 平台的混合方法二维地震波场模拟. 然后基于二维均匀介质模型将CPU与GPU版本的运行时间进行对比. 实际测试结果表明, 基于CUDA 的并行模拟方法在保证计算精度的同时显著地提高了计算速度, 为开展大规模非均匀地球介质地震波传播数值模拟提供了一种可选的方法.Abstract: Hybrid PSM/FDM method, approximating the horizontal spatial derivative with pseudospectral method (PSM) while using finite difference method (FDM) to approximate the vertical spatial derivative, is an efficient technique in seismic wave simulation because it makes usage of both the high level of parallelism in FDM and the great accuracy in PSM. With natural parallelism, graphic processing unit (GPU) is very suitable for calculating such kind of problem, and compute unified device architecture (CUDA) drastically simplifies GPU programming. In order to improve the computing efficiency, in this paper, we implemented the 2D hybrid method for seismic wave simulation based on CUDA. The GPU-based hybrid method brings about significant speedup according to numerical tests performed in this paper and proves to be a reliable way for simulating large scale seismic wave propagation.
-
-
表 1 测试环境
Table 1 Test environment
CPU 内存/GB GPU 显存/GB GCC版本 NVCC版本 Intel Core i5-4570 4 NVIDIA GeForce GTX 750 1 4.6.3 6 表 2 计算耗时表
Table 2 Elapsed time
网格数 B/T GPU耗时/s CPU耗时/s 加速比 128×128 2 0.84 16.03 190.8 256×256 4 1.96 67.10 34.23 512×512 8 6.42 266.91 41.57 1024×1024 16 22.79 1186.99 52.08 2048×2048 32 87.98 5223.65 59.37 注: B为线程块数,T为每个线程块中的线程数. -
郑亮, 张怀, 尹凤玲, 石耀霖. 2013. 基于GPU的大规模三维地震波传播有限元程序实现[C]//中国地球物理2013: 第七分会场论文集. 昆明, 2013年10月13日. Zheng L, Zhang H, Yin F L, Shi Y L. 2013. Finite element program implementation of large scale seismic wave propagation modeling based on GPU[C]//Chinese Geophysics 2013: Proceedings of the 7th Branch. Kunming, October 13, 2013.
Aoi S, Nishizawa N, Aoki T. 2012. Large scale simulation of seismic wave propagation using GPGPU[C]//15th World Conference on Earthquake Engineering. Lisbon, Portugal, September 24 to 28, 2012.
Cai C, Chen H Q, Deng Z, Chen D, Khan S U, Zeng K, Wu M X. 2012. GPGPU-aided 3D staggered-grid finite-difference seismic wave modeling[C/OL]//12th International Conference on Scalable Computing and Communications (ScalCom)/. [2014-05-12]. http://newtheme.hgpu.org/?p=8610.
NVIDIA. 2014. CUFFT Library User’s Guide[M/OL]. [2014-04-26]. http://docs.nvidia.com/cuda/cufft/.
Okamoto T, Takenaka H, Nakamura T, Aoki T. 2012. Large-scale simulation of seismic wave propagation of the 2011 Tohoku-Oki M9 earthquake[C]//Proceedings of the International Symposium on Engineering Lessons Learned from the 2011 Great East Japan Earthquake. Tokyo: Japan Association for Earthquake Engineering: 349-360.
-
期刊类型引用(0)
其他类型引用(2)