地震叠前深度偏移在CUDA平台上的实现
|
摘要
由于GPU(图形处理芯片)拥有强大的通用计算能力,在地球物理领域进行GPU计算的应用研究日益受到关注。基于CUDA软件开发环境,根据裂步法叠前深度偏移的算法特点,将偏移程序的波场延拓核心部分移植到GPU上进行并行计算,其余辅助计算在CPU上完成,实现了二维地震叠前深度偏移处理的GPU计算。在NVIDIA Tesla C870上的Marmousi模型测试结果表明,GPU处理速度是CPU(单核)的10倍左右。由于所用GPU仅支持单精度浮点运算,GPU和CPU计算结果之间存在一定的差异,但这种差异在偏移剖面上未产生视觉上可以识别的影响。
GPU (Graphics Processing Unit) has a strong universal computing power.The application of GPU computing in the field of geophysical prospecting has attracted increasing attention.According to the characteristics of split-step prestack depth migration,we transferred the core of wavefield continuation of the migration algorithm from CPU to the GPU in the CUDA software development environment,and the remaining calculations were still completed on CPU.In this way we achieved GPU computing of 2-D seismic prestack depth migration.Marmousi model tests on NVIDIA Tesla C870 show that processing on GPU is about 10 times faster than that on CPU (single core).Though the results from GPU and CPU calculation are different due to the GPU only supporting single precision floating-point operation,their difference cannot be visually identified on the migration profiles.
GPU (Graphics Processing Unit) has a strong universal computing power.The application of GPU computing in the field of geophysical prospecting has attracted increasing attention.According to the characteristics of split-step prestack depth migration,we transferred the core of wavefield continuation of the migration algorithm from CPU to the GPU in the CUDA software development environment,and the remaining calculations were still completed on CPU.In this way we achieved GPU computing of 2-D seismic prestack depth migration.Marmousi model tests on NVIDIA Tesla C870 show that processing on GPU is about 10 times faster than that on CPU (single core).Though the results from GPU and CPU calculation are different due to the GPU only supporting single precision floating-point operation,their difference cannot be visually identified on the migration profiles.
引文
1赵改善.地球物理高性能计算的新选择:GPU计算技术[J].勘探地球物理进展,2007,30(5):399~404
2 Headwave.Massively parallel architecture[EB/OL].[2008-09-08].http://www.headwave.com/article/ar-ticleview/31
3 NVIDIA.Tesla GPU computing technical brief,Ver-sion1.0.0[EB/OL].(2005-05-24)[2006-01-12].ht-tp://www.nvidia.com/
4 NVIDIA.CUDA compute unified device architectureprogramming guide,Version 1.1[EB/OL].(2007-08-31)[2008-01-30].http://www.nvidia.com/
5 NVIDIA.CUDA programming_guide_11-Ch[EB/OL].(2008-6-20)[2008-09-08].http://www.nvidia.cn/
6 NVIDIA.CUDA CUFFTlibrary[EB/OL].(2007-02-30)[2008-09-08].http://www.nvidia.com/
7董伟,徐兆涛.分步傅里叶波动方程三维叠前深度偏移算法及应用[J].油气地质与采收率,2003,10(S1):16~18
8吴恩华.图形处理器用于通用计算的技术、现状及其挑战[J].软件学报,2004,15(10):1 493~1 504
2 Headwave.Massively parallel architecture[EB/OL].[2008-09-08].http://www.headwave.com/article/ar-ticleview/31
3 NVIDIA.Tesla GPU computing technical brief,Ver-sion1.0.0[EB/OL].(2005-05-24)[2006-01-12].ht-tp://www.nvidia.com/
4 NVIDIA.CUDA compute unified device architectureprogramming guide,Version 1.1[EB/OL].(2007-08-31)[2008-01-30].http://www.nvidia.com/
5 NVIDIA.CUDA programming_guide_11-Ch[EB/OL].(2008-6-20)[2008-09-08].http://www.nvidia.cn/
6 NVIDIA.CUDA CUFFTlibrary[EB/OL].(2007-02-30)[2008-09-08].http://www.nvidia.com/
7董伟,徐兆涛.分步傅里叶波动方程三维叠前深度偏移算法及应用[J].油气地质与采收率,2003,10(S1):16~18
8吴恩华.图形处理器用于通用计算的技术、现状及其挑战[J].软件学报,2004,15(10):1 493~1 504
版权所有：© 2023 中国地质图书馆 中国地质调查局地学文献中心