Three-Point Correlation Function Parallel Algorithm Based on MPI
Lingyan Yin, Ce Yu, Jizhou Sun, Xu Liu, Jian Kun Xiao, Chao Sun · 2013
The computation of the three-point correlation function (3PCF) is a critical challenge in astrophysics. An algorithm, named as RCSF (recursive convolution for scalar fields), has been proposed to solve 3PCF by using a filter matrix to reduce the computation load. In this paper, we accelerate the 3PCF by parallel implementation of RCSF. The proposed parallel algorithm, denoted as p-RCSF, splits the 3PCF problem into sub-tasks and the tasks are assigned to each process evenly. The computation load of each task is further reduced by identifying filter matrices consisting of large number of zero elements. The proposed algorithm is capable of significantly accelerating the RCSF with higher accuracy in comparison with previous work. Experimental results show that p-RCSF is able to accelerate the RCSF by 71 times using 72 processes without loss in accuracy. In addition, the p-RCSF improves the accuracy to 99.9% which is higher than before.