基于GPU加速的地震图像重建技术
|
摘要
针对目前地层层析成像算法中正演算法存在计算量大、计算速度慢的问题,以图像处理器(GPU)为核心,研究并实现了一种基于GPU平台的时域有限差分(FDTD)正演算法。CUDA是一种由NVIDIA推出的GPU通用并行计算架构,也是目前较为成熟的GPU并行运算架构。而FDTD正演算法本身在算法特性上满足并行的要求,二者的结合将极大地加速程序的计算速度。在基于标准Marmousi速度模型的正演模拟中,程序速度提升30倍,而GPU正演图像与CPU正演结果误差小于千分之一。算例表明CUDA可以大大加速目前的FDTD正演算法,并且随着GPU硬件自身的发展和计算架构的不断改进,加速效果还将进一步提升,这将有利于后续波形反演工作的进展。
In order to solve the problems of high computation and low speed in traditional waveform tomography algorithm,FDTD algorithm was proposed based on CUDA platform with GPU as the core processor.CUDA is a general parallel computing architecture introduced by NVIDIA and also one of the most popular architectures in GPU calculation.Combined the advantage of strong calculation ability of GPU with the natural parallel characteristic of FDTD algorithm,the performance of the program can be greatly increased.Forward modeling based on Marmousi model shows that the speed is increased by 30times but the error of the result between GPU and CPU is less than millesimal.Simulation results indicate CUDA can be used in acceleration of FDTD algorithm.As the development of hardware and computing architecture,the accelerating effect can be increased,which will contribute a lot for the research of wave inversion.
In order to solve the problems of high computation and low speed in traditional waveform tomography algorithm,FDTD algorithm was proposed based on CUDA platform with GPU as the core processor.CUDA is a general parallel computing architecture introduced by NVIDIA and also one of the most popular architectures in GPU calculation.Combined the advantage of strong calculation ability of GPU with the natural parallel characteristic of FDTD algorithm,the performance of the program can be greatly increased.Forward modeling based on Marmousi model shows that the speed is increased by 30times but the error of the result between GPU and CPU is less than millesimal.Simulation results indicate CUDA can be used in acceleration of FDTD algorithm.As the development of hardware and computing architecture,the accelerating effect can be increased,which will contribute a lot for the research of wave inversion.
引文
[1]Alford R M,Kelly K R,Boore D M.Accuracy of finite-difference modeling of the acoustic wave equation[J].Geophysics,1974,39(6):834-842.
[2]Kelly K R,Ward R W,Treitel S,et al.Synthetic seismograms:a finite-difference approach[J].Geophysics,1976,41(1):2-27.
[3]Igel H,Mora P,Roillet B.Anisotropic wave propagation through finite-difference grids[J].Geophysics,1995,60(4):1203-1216.
[4]Owens J D,Houston M,David L,et al.GPU computing[J].Proc.the IEEE,2008,96(5):879-899.
[5]Mu D,Chen P,Wang L Q.Accelerating the discontinuous Galerkin method for seismic wave propagation simulations using the graphic processing unit(GPU)-single-GPU implementation[J].Computers and Geosciences,2013,51:282-292.
[6]Sharp G C,Kandasamy N,Singh H,et al.GPU-based streaming architectures for fast cone-beam CT image reconstruction and demons deformable registration[J].Physics in Medicine and Biology,2007,52(19):5771-5783.
[7]Nvidia Corporation.CUDA getting started guide for microsoft windows[EB/OL].(2012-4).https://developer.nvidia.com/cuda-toolkit-archive.
[8]Nvidia Corporation.CUDA C programming guide[EB/OL].(2012-7-15).https://developer.nvidia.com/cuda-toolkit-archive.
[9]胡英,张东,陈琼,等.基于Marmousi模型的声波方程有限差分正演算法[J].武汉大学学报:理学版,2012,58(1):78-72.
[10]Krajuwsky S E,Turner L E,Okoniewski M M.Graphics processor unit(GPU)acceleration of finitedifference time-domain(FDTD)algorithm[C]//Proc.Inter.Symp.Circuits and Systems,2004:265-268.
[11]马安国,成玉,唐遇星,等.GPU异构系统中的存储层次和负载均衡的策略研究[J].国防科技大学学报,2009 31(5):38-43.
[12]张舒.GPU高性能运算之CUDA[M].北京:中国水利水电出版社,2009:141-156.
[13]丁科.PML吸收边界条件影响因素分析[J].物探与化探,2012,36(4):623-627.
[2]Kelly K R,Ward R W,Treitel S,et al.Synthetic seismograms:a finite-difference approach[J].Geophysics,1976,41(1):2-27.
[3]Igel H,Mora P,Roillet B.Anisotropic wave propagation through finite-difference grids[J].Geophysics,1995,60(4):1203-1216.
[4]Owens J D,Houston M,David L,et al.GPU computing[J].Proc.the IEEE,2008,96(5):879-899.
[5]Mu D,Chen P,Wang L Q.Accelerating the discontinuous Galerkin method for seismic wave propagation simulations using the graphic processing unit(GPU)-single-GPU implementation[J].Computers and Geosciences,2013,51:282-292.
[6]Sharp G C,Kandasamy N,Singh H,et al.GPU-based streaming architectures for fast cone-beam CT image reconstruction and demons deformable registration[J].Physics in Medicine and Biology,2007,52(19):5771-5783.
[7]Nvidia Corporation.CUDA getting started guide for microsoft windows[EB/OL].(2012-4).https://developer.nvidia.com/cuda-toolkit-archive.
[8]Nvidia Corporation.CUDA C programming guide[EB/OL].(2012-7-15).https://developer.nvidia.com/cuda-toolkit-archive.
[9]胡英,张东,陈琼,等.基于Marmousi模型的声波方程有限差分正演算法[J].武汉大学学报:理学版,2012,58(1):78-72.
[10]Krajuwsky S E,Turner L E,Okoniewski M M.Graphics processor unit(GPU)acceleration of finitedifference time-domain(FDTD)algorithm[C]//Proc.Inter.Symp.Circuits and Systems,2004:265-268.
[11]马安国,成玉,唐遇星,等.GPU异构系统中的存储层次和负载均衡的策略研究[J].国防科技大学学报,2009 31(5):38-43.
[12]张舒.GPU高性能运算之CUDA[M].北京:中国水利水电出版社,2009:141-156.
[13]丁科.PML吸收边界条件影响因素分析[J].物探与化探,2012,36(4):623-627.
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