Fast 3-Point Correlation Function Approximation on GPU

摘要

The problem of 3-Point Correlation Function (3PCF) in astrophysics processes megabytes data with complex calculations, which is an important tool for calculating properties of heterogeneous systems, but its algorithmic complex is a notorious problem. The fast 3PCF Approximation algorithm can improve the efficiency by reduce the precision of result. In this paper, we are going to introduce a design of this algorithm on GPU, which is 13× speedup over a single CPU. Moreover, we will optimize it in the calculation details: converting the 3D arrays to 1D, padding Os to arrays and shrinking the kernel array. Finally, this algorithm can achieve 27× speedup additional, and 347× speedup over a single CPU.