基于OpenCL的射电干涉阵成像网格化算法实现
|
摘要
天文软件开发与应用中迫切需要在单机环境下进行高性能的科学数据处理,由于机器配置不同,采用传统的CUDA+GPU技术存在明显的局限,不利于天文软件的快速移植和无缝运行。针对明安图射电频谱日像仪数据处理中的网格化(Gridding)算法,采用并行计算OpenCL技术进行多线程编程实现。实验结果表明,基于OpenCL实现的网格化算法不仅能够在图形处理器上运行,而且能够在纯中央处理器上运行。当选择在图形处理器上执行时,算法的执行效率与基于CUDA实现的网格化算法执行效率大致相当,但算法不局限于NVIDIA GPU,解决了算法对CUDA+GPU的依赖;同时算法也能在纯中央处理器上较快速地执行,适用于单机模式下进行天文软件的开发和测试,也便于天文软件的应用与推广。
It's urgent to carry out high-performance scientific data processing with a single machine in the development and application of astronomical software.However,due to the different configurations of machines,the traditional CUDA + GPU technology has obvious limitations in portability and seamlessness.According to gridding algorithm in MingantU SpEctral Radioheliograph(MUSER) data processing,the OpenCL technology is used in parallel to implement multi-thread programming.The experimental results show that the gridding algorithm based on OpenCL can not only run on GPU,but also merely on CPU.While choosing execution on GPU,the execution efficiency of gridding algorithm is approximately equal to that on CUDA.At the same time,the algorithm is not limited to the NVIDIA' s GPU,which has solved the problem of environmental dependence of CUDA + GPU.And the algorithm also has an acceptable execution efficiency in implementation with the merely CPU,which is suitable for development and testing astronomy software with a single machine and will facilitate the application and promotion of astronomical software.
It's urgent to carry out high-performance scientific data processing with a single machine in the development and application of astronomical software.However,due to the different configurations of machines,the traditional CUDA + GPU technology has obvious limitations in portability and seamlessness.According to gridding algorithm in MingantU SpEctral Radioheliograph(MUSER) data processing,the OpenCL technology is used in parallel to implement multi-thread programming.The experimental results show that the gridding algorithm based on OpenCL can not only run on GPU,but also merely on CPU.While choosing execution on GPU,the execution efficiency of gridding algorithm is approximately equal to that on CUDA.At the same time,the algorithm is not limited to the NVIDIA' s GPU,which has solved the problem of environmental dependence of CUDA + GPU.And the algorithm also has an acceptable execution efficiency in implementation with the merely CPU,which is suitable for development and testing astronomy software with a single machine and will facilitate the application and promotion of astronomical software.
引文
[1]YAN Y,ZHANG J,WANG W,et al.The Chinese Spectral Radioheliograph-CSRH[J].Earth Moon&Planets,2009,104(1-4):97-100.
[2]赖铖,梅盈,邓辉,等.MUSER可见度数据积分方法与实现[J].天文研究与技术,2018,15(1):78-86.
[3]周鑫磊,王威,王锋,等.基于QT的MUSER观测数据多屏图形化实时显示的设计与实现[J].天文研究与技术,2015,12(4):503-509.
[4]SHI C,WANG F,DENG H,et al.High performance negative database for massive data management system of the Mingantu Spectral Radioheliograph[J].Publications of the Astronomical Society of the Pacific,2017,129(978):084501(10pp).
[5]DAI H M,MEI Y,WANG W,et al.An auto-flag method of radio visibility data based on support vector machine[J].Chinese Astronomy&Astrophysics,2017,41(1):125-135.
[6]WANG F,DENG H,WANG W.High performance distributed data processing pipeline for Chinese Spectral RadioHeliograph[C]//Radio Science Conference.2015.
[7]WEI S L,WANG F,DENG H,et al.OpenCluster:a flexible distributed computing framework for astronomical data processing[J].Publications of the Astronomical Society of the Pacific,2016,129(972):024001(14pp).
[8]WANG F,MEI Y,DENG H,et al.Distributed data-processing pipeline for Mingantu Ultrawide Spectral Radioheliograph[J].Publications of the Astronomical Society of the Pacific,2015,127(950):383-396.
[9]陈泰燃,王威,王锋,等.基于MPI的高性能UVFITS数据合成研究与应用[J].天文研究与技术,2016,13(2):184-189.
[10]MEI Y,WANG F,WANG W,et al.GPU-based high-performance imaging for Mingantu Spectral Radio Heliograph[J].Publications of the Astronomical Society of the Pacific,2018,130(983):014503(11pp).
[11]冯勇,陈坤,邓辉,等.基于OpenCL的MUSER CLEAN算法研究与实现[J].天文学报,2017,58(2):55-64.
[12]HGBOM J A.Aperture synthesis with a non-regular distribution of interferometer baselines[J].Astronomy&Astrophysics Supplement,1974,15(15):417-426.
[13]THOMPSON A R,BRACEWELL R N.Interpolation and fourier transformation of fringe visibilities[J].Astronomical Journal,1973,79(1):11-24.
[14]SEDARAT H,NISHIMURA D G.On the optimality of the gridding reconstruction algorithm[J].IEEE Transactions on Medical Imaging,2000,19(4):306-317.
[15]O'SULLIVAN J D.A fast sinc function gridding algorithm for fourier inversion in computer tomography[J].IEEE Transactions on Medical Imaging,1985,4(4):200-207.
[16]JACKSON J I,MEYER C H,NISHIMURA D G,et al.Selection of a convolution function for fourier inversion using gridding[J].IEEE Transactions on Medical Imaging,1991,10(3):473-478.
[17]BOONE F.Weighting interferometric data for direct imaging[J].Experimental Astronomy,2013,36(1-2):77-104.
[2]赖铖,梅盈,邓辉,等.MUSER可见度数据积分方法与实现[J].天文研究与技术,2018,15(1):78-86.
[3]周鑫磊,王威,王锋,等.基于QT的MUSER观测数据多屏图形化实时显示的设计与实现[J].天文研究与技术,2015,12(4):503-509.
[4]SHI C,WANG F,DENG H,et al.High performance negative database for massive data management system of the Mingantu Spectral Radioheliograph[J].Publications of the Astronomical Society of the Pacific,2017,129(978):084501(10pp).
[5]DAI H M,MEI Y,WANG W,et al.An auto-flag method of radio visibility data based on support vector machine[J].Chinese Astronomy&Astrophysics,2017,41(1):125-135.
[6]WANG F,DENG H,WANG W.High performance distributed data processing pipeline for Chinese Spectral RadioHeliograph[C]//Radio Science Conference.2015.
[7]WEI S L,WANG F,DENG H,et al.OpenCluster:a flexible distributed computing framework for astronomical data processing[J].Publications of the Astronomical Society of the Pacific,2016,129(972):024001(14pp).
[8]WANG F,MEI Y,DENG H,et al.Distributed data-processing pipeline for Mingantu Ultrawide Spectral Radioheliograph[J].Publications of the Astronomical Society of the Pacific,2015,127(950):383-396.
[9]陈泰燃,王威,王锋,等.基于MPI的高性能UVFITS数据合成研究与应用[J].天文研究与技术,2016,13(2):184-189.
[10]MEI Y,WANG F,WANG W,et al.GPU-based high-performance imaging for Mingantu Spectral Radio Heliograph[J].Publications of the Astronomical Society of the Pacific,2018,130(983):014503(11pp).
[11]冯勇,陈坤,邓辉,等.基于OpenCL的MUSER CLEAN算法研究与实现[J].天文学报,2017,58(2):55-64.
[12]HGBOM J A.Aperture synthesis with a non-regular distribution of interferometer baselines[J].Astronomy&Astrophysics Supplement,1974,15(15):417-426.
[13]THOMPSON A R,BRACEWELL R N.Interpolation and fourier transformation of fringe visibilities[J].Astronomical Journal,1973,79(1):11-24.
[14]SEDARAT H,NISHIMURA D G.On the optimality of the gridding reconstruction algorithm[J].IEEE Transactions on Medical Imaging,2000,19(4):306-317.
[15]O'SULLIVAN J D.A fast sinc function gridding algorithm for fourier inversion in computer tomography[J].IEEE Transactions on Medical Imaging,1985,4(4):200-207.
[16]JACKSON J I,MEYER C H,NISHIMURA D G,et al.Selection of a convolution function for fourier inversion using gridding[J].IEEE Transactions on Medical Imaging,1991,10(3):473-478.
[17]BOONE F.Weighting interferometric data for direct imaging[J].Experimental Astronomy,2013,36(1-2):77-104.