Parallelization of the Friends-of-Friends Algorithm for a Hybrid Computing with OpenACC

The use of accelerators to process code portions that deal with large amount of data is seen as a promissory field in hybrid computing. The Cosmology area is an example for recording giant and heterogeneous databases. Classification procedures are an important tool for data analysis. A very known algorithm to classify astronomical data is the Friends-of-Friends (FoF) algorithm. One framework for cosmological studies is to apply N-body simulations, and the analysis for such data can be done with the FoF. In this work, we present parallel versions of the FoF with O(N²) time complexity using the OpenACC standard in an environment with CPU and GPU. The results show that the algorithm can be well explored in this context, with an expressive performance improvement when rewriting the original program code to the hybrid environment and with a significant improvement when parallelized with minimal modifications using only compiling directives.