摘要
波动方程叠前深度偏移适用于强横向变速介质,是一种高精度成像方法,但其巨大的计算量阻碍了该技术的应用。Xeon Phi是一种全新的高性能计算设备,为波动方程叠前深度偏移方法的推广应用提供了新的技术支持。以裂步傅里叶算子为例,介绍了面向Xeon Phi平台的偏移算法移植和优化方法,即采用offload模式将计算核函数加载到Xeon Phi设备上,在Xeon Phi协处理器上采用多线程方式,并且调整程序结构,充分利用SIMD矢量引擎提高向量化处理效率。扩展负载动态均衡的并行框架,形成了一套适用于大规模异构系统、基于Xeon Phi平台的波动方程叠前深度偏移软件。实际数据测试表明Xeon Phi平台可以极大地提高地震偏移处理效率,具有良好的可扩展性。
Wave equation prestack depth migration is a high-precision seismic imaging method,suitable for complex media with sharp lateral velocity variation,however,its huge computation hinders its application.Xeon Phi is a new product for throughput computing which provides a better choice for these methods.Taking the Split Step Fourier algorithm for example,we introduce the code migration and performance optimization on Xeon Phi platform.We offload the kernel which is fully threaded to Xeon Phi by predefined compiler pragmas so as to take full advantage of SIMD engine to improve the efficiency of the parallelization processing;and we change the code structure to ensure the vectorization.Besides,by extending the parallel processing framework with the dynamic workload-balance mechanism,we develop a seismic imaging system based on Xeon Phi platform which is suitable for large-scale,full hybrid heterogeneous clusters.Finally,we conduct a real workload test and the results show that Xeon Phi platform can significantly improve the seismic imaging efficiency,and has good core scalability.
引文
[1]Li Ke-chen,Huang Liu-sheng.Load balance in depth migration parallel computation of 3D wave equation prestack[J].Computer Applications,2003,23(1):18-20.(in Chinese)
[2]Li Ken-li,Yang Jin,Peng Cheng-bin,et al.Research of the seismic data processing support library based on the hybrid model MPI+OpenMP[J].Computer Engineering&Science,2007,29(12):136-140.(in Chinese)
[3]Araya-Polo M,Rubio F,dela Cruz,et al.High-performance seismic acoustic imaging by reverse-time migration on the Cell/B.E[R].Kaleidoscope Project Report of Barcelona Supercomputing Center,2008.
[4]Ma Zhao-gui,Zhao Gai-shan,Wu Gang-shan,et al.Multi-level parallel optimization technique for prestack time migration from rugged topography[J].Geophysical Prospecting for Petroleum,2013,52(3):280-287.(in Chinese)
[5]Peng Jun-jie.Research of processing seismic data on CUDA architecture[D].Changsha:Hunan University,2009.(in Chinese)
[6]Liu Qi-lin,Huang Yue,Tang Jian-ming,et al.Wave equation pre-stack depth migration with GPU[J].Computing Techniques for Geophysical and Geochemical Exploration,2010,32(4):386-391.(in Chinese)
[7]Kuang Bin,Du Ji-xiu,Wang Hua-zhong,et al.3D wave equation prestack depth migration on Graphic Processing Unit[J].Oil Geophysical Prospecting,2011,46(5):705-709.(in Chinese)
[8]Cabezas J,Araya-Polo M,Gelado I,et al.High-performance reverse time migration on GPU[C]∥Proc of International Conference of the Chilean Computer Science Society(SCCC),2009:77-86.
[9]Abdelkhalek R,Calandra H,Coulaud O,et al.Fast seismic modeling and reverse time migration on a GPU cluster[C]∥Proc of the 2009 High Performance Computing&Simulation,2009:1.
[10]Ma Zhao-gui,Zhao Gai-shan,Wu Gang-shan,et al.Kirchhoff pre-stack time migration of complex surface data on CPU+GPU[J].Oil Geophysical Prospecting,2014,49(3):530-537.(in Chinese)
[11]Claerbout J F.Coarse grid calculation of waves in inhomogeneous media with application to delineation of complicated seismic structure[J].Geophysics,1970,35(3):407.
[12]Stolt R H.Migration by Fourier transforms[J].Geophysics,1978,44(3):1454.
[13]Gazdag J.Wave equation migration with the phase-shift method[J].Geophysics,1978,43(7):1342.
[14]Stoffa P L,Fokkema J T,Freire R M,et al.Split-step Fourier migration[J].Geophysics,1990,55(4):410.
[15]Ristow D,Rühl T.Fourier finite-difference migration[J].Geophysics,1994,59(12):1882.
[16]de Hoop M V,le Rousseau J H,Wu R S.Generalization of the phase-screen approximation for the scattering of acoustic waves[J].Wave Motion,2000,31(1):43-70.
[17]Liu Li-nong,Chen Shu-min,Gao Liang,et al.The parallel compution of 3D wave-equation prestack depth migration[J].Chinese Journal of Geophysics,2002,45(Suppl.):298-306.(in Chinese)
[18]Shan Lian-yu,Xu Zhao-tao.Parallel algorithm of one-way equation prestack depth migration[J].Oil Geophysical Prospection,2008,43(6):636-640.(in Chinese)
[1]李克臣,黄刘生.PC Cluster实现三维叠前深度偏移并行计算的负载平衡策略[J].计算机应用,2003,23(1):18-20.
[2]李肯立,杨进,彭成斌,等.基于MPI+OpenMP混合模型的并行地震数据处理支撑库的研究[J].计算机工程与科学,2007,29(12):136-140.
[4]马召贵,赵改善,武港山,等.起伏地表叠前时间偏移的多级并行优化技术[J].石油物探,2013,52(3):280-287.
[5]彭俊杰.基于CUDA平台地震数据处理研究[D].长沙:湖南大学,2009.
[6]刘奇琳,黄跃,唐建明,等.波动方程叠前深度偏移的GPU技术[J].物探化探计算技术,2010,32(4):386-391.
[7]匡斌,杜继修,王华忠,等.基于GPU计算平台的三维波动方程叠前深度偏移[J].石油地球物理勘探,2011,46(5):705-709.
[10]马召贵,赵改善,武港山,等.基于CPU+GPU联合计算真地表叠前时间偏移实用化研究[J].石油地球物理勘探,2014,49(3):530-537.
[17]刘礼农,陈树民,高亮,等.波动方程三维叠前深度偏移并行计算流程探索[J].地球物理学报,2002,45(增):298-306.
[18]单联瑜,徐兆涛.单程波动方程叠前深度偏移并行算法[J].石油地球物理勘探,2008,43(6):636-640.