Solar Speckle Image Reconstruction Based On Multi-core CPU

The energy of the earth mainly comes from the sun.Observing the solar active region with the large aperture telescope is the most effective means to understand the sun and reduce the influence of the solar activity on the human beings.The ground-based astronomical telescope could not reach the theoretical resolution because of the influence of atmospheric turbulence.It is necessary to use the adaptive optical system or image post-processing to obtain the diffraction limited image.At present,the combination of Speckle Interferometry and Speckle Masking to reconstruct frequency domain is the most commonly used reconstruction algorithm in the astronomical observations.The speckle image reconstruction needs parallel computing to accelerate the program due to the large amount of computation and time-consuming.Speckle image reconstruction with inefficient programming languages such as:IDL,Matlab could not be used in large-scale reconstructed.It takes more than ten hours to reconstruct a 1280 x 1280 pixel image in Matlab.It is necessary to use the efficient programming language and the parallel platform to reduce computing time.Common parallel standards and parallel platforms are compared in this paper.According to different steps of the algorithm,different parallel schedules and optimized methods are applied.This paper introduces the parallel standard OpenMP and MPI,Many Integrated Core based on X86 architecture and parallel standard CUDA based on GPU.The Speckle Interference and the Speckle Mask on solar image reconstruction are also described in this paperSolar speckle image reconstruction research is carried out under parallel computing and image reconstruction theories with a server equipped with Xeon E5-2670 v3 with Intel CPU and Intel Xeon Phi 7120.Efficient programming language could not be separated from highly optimized code.Under the guidance of the program optimization theory,the elapsed time of the serial version program isreduced from 12.5 minutes to 289 seconds.In order to reduce the elapsed time further,the multi-core CPU is chose as the parallel platform and the OpenMP is chose as the parallel standard.The optimized parallel program completes the reconstruction of a1280 x 1280 pixel image within 30 seconds using 12 cores.Compared with single core operation,the speedup is 9.66 and the parallel efficiency is about 80%.The OpenMP could use one more computational core than MPI,it could also use shared memory to get a better speedup.The program based on Multi-core CPU meets the needs of observed data processing on photosphere and chromosphere.The program contributes to the subsequent study of solar physics.