其他知识库

天文软件开发中迫切需要在单机环境下进行高性能数据处理工作,但由于机器配置不同,采用传统的多线程、CUDA(Compute Unified Device Architecture)+GPU(Graphic Processing Unit)等方式都存在明显的局限,不利于天文软件的快速移植和无缝运行.对明安图频谱射电日像仪(Mingant U Sp Ectral Radioheliograph,MUSER)数据处理系统开发中所采用的Open CL(Open Computing Language)技术进行介绍,并基于Open CL实现H¨ogbom CLEAN算法.整体工作通过Python语言和Py Open CL扩展包实现并行洁化处理.实验结果表明:基于Open CL实现的CLEAN算法与基于CUDA实现的CLEAN算法具有大致相当的运行效率,同时也可以无需修改代码直接实现纯CPU(Central Processing Unit)环境下的高性能数据处理,解决了对CUDA+GPU环境依赖的问题,在保证MUSER数据处理系统洁化过程性能的基础上,提高了系统对硬件平台的适应性.该工作验证了Open CL在科学数据处理中的可用性,可以预见:由于Open CL所具有的异构环境下高性能计算特性,Open CL将是未来天文高性能软件开发的首选技术.

It’s urgent to carry out high-performance data processing with a single machine in the development of astronomical software. However, due to the different configuration of the machine, traditional programming techniques such as multi-threading, and CUDA (Compute Unified Device Architecture)+GPU (Graphic Processing Unit) have obvious limitations in portability and seamlessness between different operation systems. The OpenCL (Open Computing Language) used in the development of MUSER (MingantU SpEctral Radioheliograph) data processing system is introduced. And the H¨ogbom CLEAN algorithm is re-implemented into parallel CLEAN algorithm by the Python language and PyOpenCL extended package. The experimental results show that the CLEAN algorithm based on OpenCL has approximately equally operating efficiency compared with the former CLEAN algorithm based on CUDA. More important, the data processing in merely CPU (Central Processing Unit) environment of this system can also achieve high performance, which has solved the problem of environmental dependence of CUDA+GPU. Overall, the research improves the adaptability of the system with emphasis on performance of MUSER image clean computing. In the meanwhile, the realization of OpenCL in MUSER proves its availability in scientific data processing. In view of the high-performance computing features of OpenCL in heterogeneous environment, it will probably become the preferred technology in the future high-performance astronomical software development.