Parallel Optimization Research And Implementation Of Radio Interference Array Imaging Algorithm

With the continuous construction of the new generation of radio telescopes,the scale of astronomical observation data has increased dramatically,and research on high-performance data processing methods is crucial.Because of different machine configurations,the traditional methods,such as OpenMP and GPU+CUDA,have obvious shortcomings on a single-machine environment,which are not conducive to the rapid operation or transplantation of the system.Taking the data processing of MingantU SpEctral Radioheliograph(MUSER)as an example,this paper analyses the data processing of radio interferometric array imaging,and further uses Open Computing Language(OpenCL)to study and implement the parallel optimization of radio interferometric array imaging algorithm.The specific research contents are as follows:1.The principle of key algorithms(gridding and clean)involved in the imaging process of radio interferometric array are analyzed,and the parallel optimization of the algorithms are studied based on the programming principle of OpenCL architecture;2.Based on OpenCL,the gridding and clean algorithm are re-implemented by multi-threaded programming,and successfully deployed in MUSER data processing system;3.To test the performance of the imaging algorithm based on OpenCL under different system configurations(CPU and GPU).And the performance of the imaging algorithm is compared and analyzed with the previous serial implementation and GPU + CUDA implementation.The experimental results show that the imaging algorithm based on OpenCL can run not only in GPU environment,but also in CPU environment,which solves the dependence of the algorithm on GPU environment and improves the adaptability of the algorithm to hardware platform.At the same time,in the CPU environment,the efficiency of imaging algorithm based on OpenCL is much higher than that of serial implementation.In the GPU environment,the efficiency of imaging algorithm based on OpenCL is approximately equal to that of GPU+CUDA.With the need of popularization and application of MUSER data processing system,this paper improves the MUSER data processing system based on the cross-Hardware platform characteristics of the imaging algorithm implemented by OpenCL,and provides convenience for researchers to use the MUSER data processing system.OpenCL extends the heterogeneous system from CPU+NVIDIA GPU mode to CPU+multi-core computing device mode.Due to the change of this heterogeneous system mode,it can be predicted that OpenCL will probably become the preferred technology for developing high-performance astronomical data software.