基于CUDA的超声二维声场EFIT仿真
|
摘要
随着图形处理器(GPU)的快速发展,基于计算设备统一构架(CUDA)可以方便地将并行计算技术应用于超声声场数值仿真计算,极大地提升计算效率。阐述了弹性动力学有限积分算法(EFIT)的原理,在采用CPU实现带吸收边界的钢材料二维点源激励声场仿真的基础上,基于GPU实现了仿真模型的并行计算,介绍了GPU程序的设计流程和参数优化方法,包括纹理内存使用、吸收边界优化和数据传输优化。对比了相同条件下CPU和GPU仿真计算的耗时和平均计算效率,定量分析了GPU对于EFIT模型效率的提升。比对结果表明,EFIT具有良好的并行计算条件,采用并行计算方法能够有效提升模型计算速度,对于复杂声场仿真应用具有广阔的应用前景。
With the rapid development of graphic processing unit( GPU),the parallel computing technology could be easily applied in the numerical simulation of ultrasonic sound field based on compute unified device architecture( CUDA). The calculating efficiency could be greatly promoted by using the parallel computing technology. The theory of elastodynamic finite integration technology( EFIT) is illustrated in this article. An EFIT 2 D ultrasonic sound field model with point source and absorption boundary in steel material is established by CPU,and on the basis of CPU code,the GPU model is built with parallel computing technology. The flow design procedure and parameter optimization method of GPU model are introduced,including the texture memory use,absorption boundary optimization and data transmission optimization. Based on the comparison of time consumption and average calculating efficiency,the efficiency promotion of EFIT model of CPU and GPU version are quantitatively analyzed. The result reveal that the EFIT model with GPU has much higher calculating efficiency. According to the comparison result,the calculation speed of EFIT model is promoted significantly with the parallel computing technology. And it has broad application prospects in complicated acoustic field simulation.
With the rapid development of graphic processing unit( GPU),the parallel computing technology could be easily applied in the numerical simulation of ultrasonic sound field based on compute unified device architecture( CUDA). The calculating efficiency could be greatly promoted by using the parallel computing technology. The theory of elastodynamic finite integration technology( EFIT) is illustrated in this article. An EFIT 2 D ultrasonic sound field model with point source and absorption boundary in steel material is established by CPU,and on the basis of CPU code,the GPU model is built with parallel computing technology. The flow design procedure and parameter optimization method of GPU model are introduced,including the texture memory use,absorption boundary optimization and data transmission optimization. Based on the comparison of time consumption and average calculating efficiency,the efficiency promotion of EFIT model of CPU and GPU version are quantitatively analyzed. The result reveal that the EFIT model with GPU has much higher calculating efficiency. According to the comparison result,the calculation speed of EFIT model is promoted significantly with the parallel computing technology. And it has broad application prospects in complicated acoustic field simulation.
引文
[1]徐娜,李洋,周正干,等. FDTD方法的改进及在超声波声场计算中的应用[J].北京航空航天大学学报,2013,39(1):78-82.XU N,LI Y,ZHOU Z G,et al. Improvement of finite difference time domain method and its application to calculation of ultrasonic sound fields[J]. Journal of Beijing University of Aeronautics and Astronautics,2013,39(1):78-82(in Chinese).
[2] FELLINGER P,MARKLEIN R,LANGENBERG K J,et al. Numerical modeling of elastic wave propagation and scattering with EFIT—Elastodynamic finite integration technique[J]. Wave Motion,1995,21:47-66.
[3]张霞,何兴无. CUDA平台下的超声弹性成像并行处理算法[J].计算机与数字工程,2012,40(9):113-116.ZHANG X,HE X W. A Parallel algorithm of ultrasound strainimaging based on CUDA[J]. Computer&Digital Engineering,2012,40(9):113-116(in Chinese).
[4]贾春刚,郭立新,刘伟.基于GPU的并行FDTD方法在二维粗糙面散射中的应用[J].电波科学学报,2016,31(4):683-687.JIA C G,GUO L X,LIU W. GPU-based FDTD method for analysis of electromagnetic scattering from a 2D rough surface[J].Chinese Journal of Radio Science,2016,31(4):683-687(in Chinese).
[5]付小波,马中高,余嘉顺,等.基于多图形处理单元加速的各向异性弹性波正演模拟[J].科学技术与工程,2018,18(11):16-22.FU X B,MA Z G,YU J S,et al. Anisotropic elastic wave forward modeling based on multiple graphics processing unit[J].Science Technology and Engineering,2018,18(11):16-22(in Chinese).
[6]杨尚琴.地震正演数值模拟仿真计算的并行优化设计方法[J].地球物理学进展,2017,32(3):1290-1296.YANG S Q. Parallel optimization design method for seismic forward modeling numerical simulation calculation[J]. Progress in Geophysics,2017,32(3):1290-1296(in Chinese).
[7] FELLINGER F,LANGENBERG K J. Numerical techniques for elastic wave propagation and scattering[C]∥Proceedings of the IUTAM Symposium on Elastic Wave Propagation and Ultrasonic Evaluation,1990:81-86.
[8] SCHUBERT F. Numerical time-domain modeling of linear and nonlinear ultrasonic wave propagation using finite integration techniques—Theory and applications[J]. Ultrasonics,2004,42(1-9):221-229.
[9]丁辉.计算超声学———声场分析及应用[M].北京:科学出版社,2010:33-36.DING H. Computational ultrasonics—Analysis and application of ultrasonic fiels[M]. Beijing:Science Press,2010:33-36(in Chinese).
[10]余涛.超声波在混凝土中传播的数值模拟[D].长沙:中南大学,2013:4-9.YU T. Numerical simulation of ultrasonic wave propagation in concrete[D]. Changsha:Central South University,2013:4-9(in Chinese).
[11] BERENGER J. A perfectly matched layer for the absorption of electromagnetic waves[J]. Journal of Computational Physics,1994,114(2):185-200.
[12]廉西猛,单联瑜,隋志强,等.地震正演数值模拟完全匹配层吸收边界条件研究综述[J].地球物理学进展,2015,30(4):1725-1733.LIAN X M,SHAN L Y,SUI Z Q,et al. An overview of research on perfectly matched layers absorbing boundary condition of seismic forward numerical simulation[J]. Progress in Geophysics,2015,30(4):1725-1733(in Chinese).
[13]刘洋.波动方程时空域有限差分数值解及吸收边界条件研究进展[J].石油地球物理勘探,2014,49(1):35-46.LIU Y. The review of finite difference numerical solution for wave equation in time domain and obsorption boundary conditions[J]. Oil Geophysical Prospecting,2014,49(1):35-46(in Chinese).
[14]秦臻,任培罡,姚姚,等.弹性波正演模拟中PML吸收边界条件的改进[J].地球科学———中国地质大学学报,2009,34(4):658-664.QIN Z,REN P G,YAO Y,et al. Improvement of PML absorbing boundary conditions in elastic wave forward modeling[J]. Earth Science—Journal of China University of Geosciences,2009,34(4):658-664(in Chinese).
[15]卢风顺,宋君强,银福康,等. CPU/GPU协同并行计算研究综述[J].计算机科学,2011,38(3):5-9.LU F S,SONG J Q,YIN F K,et al. Survey of CPU/GPU synergetic parallel computing[J]. Computer Science,2011,38(3):5-9(in Chinese).
[16] SANDERS J,KANDROT E. GPU高性能编程CUDA实战[M].聂雪军,等,译.北京:机械工业出版社,2011:84-100.SANDERS J,KANDROT E. CUDA by example———An introduction to general-purpose GPU programming[M]. NIE X J,et al.,translated. Beijing:China Machine Press,2011:84-100(in Chinese).
[17]方民权,张卫民,方建滨,等. GPU编程与优化:大众高性能计算[M].北京:清华大学出版社,2016:273-276.FANG M Q,ZHANG W M,FANG J B,et al. GPU programming and code optimization:High performance computing for the masses[M]. Beijing:Tsinghua University Press,2016:273-276(in Chinese).
[2] FELLINGER P,MARKLEIN R,LANGENBERG K J,et al. Numerical modeling of elastic wave propagation and scattering with EFIT—Elastodynamic finite integration technique[J]. Wave Motion,1995,21:47-66.
[3]张霞,何兴无. CUDA平台下的超声弹性成像并行处理算法[J].计算机与数字工程,2012,40(9):113-116.ZHANG X,HE X W. A Parallel algorithm of ultrasound strainimaging based on CUDA[J]. Computer&Digital Engineering,2012,40(9):113-116(in Chinese).
[4]贾春刚,郭立新,刘伟.基于GPU的并行FDTD方法在二维粗糙面散射中的应用[J].电波科学学报,2016,31(4):683-687.JIA C G,GUO L X,LIU W. GPU-based FDTD method for analysis of electromagnetic scattering from a 2D rough surface[J].Chinese Journal of Radio Science,2016,31(4):683-687(in Chinese).
[5]付小波,马中高,余嘉顺,等.基于多图形处理单元加速的各向异性弹性波正演模拟[J].科学技术与工程,2018,18(11):16-22.FU X B,MA Z G,YU J S,et al. Anisotropic elastic wave forward modeling based on multiple graphics processing unit[J].Science Technology and Engineering,2018,18(11):16-22(in Chinese).
[6]杨尚琴.地震正演数值模拟仿真计算的并行优化设计方法[J].地球物理学进展,2017,32(3):1290-1296.YANG S Q. Parallel optimization design method for seismic forward modeling numerical simulation calculation[J]. Progress in Geophysics,2017,32(3):1290-1296(in Chinese).
[7] FELLINGER F,LANGENBERG K J. Numerical techniques for elastic wave propagation and scattering[C]∥Proceedings of the IUTAM Symposium on Elastic Wave Propagation and Ultrasonic Evaluation,1990:81-86.
[8] SCHUBERT F. Numerical time-domain modeling of linear and nonlinear ultrasonic wave propagation using finite integration techniques—Theory and applications[J]. Ultrasonics,2004,42(1-9):221-229.
[9]丁辉.计算超声学———声场分析及应用[M].北京:科学出版社,2010:33-36.DING H. Computational ultrasonics—Analysis and application of ultrasonic fiels[M]. Beijing:Science Press,2010:33-36(in Chinese).
[10]余涛.超声波在混凝土中传播的数值模拟[D].长沙:中南大学,2013:4-9.YU T. Numerical simulation of ultrasonic wave propagation in concrete[D]. Changsha:Central South University,2013:4-9(in Chinese).
[11] BERENGER J. A perfectly matched layer for the absorption of electromagnetic waves[J]. Journal of Computational Physics,1994,114(2):185-200.
[12]廉西猛,单联瑜,隋志强,等.地震正演数值模拟完全匹配层吸收边界条件研究综述[J].地球物理学进展,2015,30(4):1725-1733.LIAN X M,SHAN L Y,SUI Z Q,et al. An overview of research on perfectly matched layers absorbing boundary condition of seismic forward numerical simulation[J]. Progress in Geophysics,2015,30(4):1725-1733(in Chinese).
[13]刘洋.波动方程时空域有限差分数值解及吸收边界条件研究进展[J].石油地球物理勘探,2014,49(1):35-46.LIU Y. The review of finite difference numerical solution for wave equation in time domain and obsorption boundary conditions[J]. Oil Geophysical Prospecting,2014,49(1):35-46(in Chinese).
[14]秦臻,任培罡,姚姚,等.弹性波正演模拟中PML吸收边界条件的改进[J].地球科学———中国地质大学学报,2009,34(4):658-664.QIN Z,REN P G,YAO Y,et al. Improvement of PML absorbing boundary conditions in elastic wave forward modeling[J]. Earth Science—Journal of China University of Geosciences,2009,34(4):658-664(in Chinese).
[15]卢风顺,宋君强,银福康,等. CPU/GPU协同并行计算研究综述[J].计算机科学,2011,38(3):5-9.LU F S,SONG J Q,YIN F K,et al. Survey of CPU/GPU synergetic parallel computing[J]. Computer Science,2011,38(3):5-9(in Chinese).
[16] SANDERS J,KANDROT E. GPU高性能编程CUDA实战[M].聂雪军,等,译.北京:机械工业出版社,2011:84-100.SANDERS J,KANDROT E. CUDA by example———An introduction to general-purpose GPU programming[M]. NIE X J,et al.,translated. Beijing:China Machine Press,2011:84-100(in Chinese).
[17]方民权,张卫民,方建滨,等. GPU编程与优化:大众高性能计算[M].北京:清华大学出版社,2016:273-276.FANG M Q,ZHANG W M,FANG J B,et al. GPU programming and code optimization:High performance computing for the masses[M]. Beijing:Tsinghua University Press,2016:273-276(in Chinese).