Comparison of hardware accelerators for some ALICE online computing applications

ALICE (A Large Ion Collider Experiment) is an experiment which studies about interacting matter and the quark-gluon plasma at the European Organization for Nuclear Research (CERN) Large Hadron Collider (LHC). Due to the detector upgrade in 2018, 1TB/s data are expected to flow from the detector and the computing system needs to process these data online. To speed up the online processing, different hardware accelerators must be tested thoroughly to see whether they are suitable for the tasks. In this paper, we propose a benchmark method for the ALICE O^2 project on various accelerators. There are three targeted computing platform for benchmark, namely, Graphics Processing Unit (GPU), Many-Integrated Core (MIC), and Accelerated Processing Unit (APU). For completeness, CUDA, OpenMP and OpenCL were used to implement the selected algorithms, which correspond to some of the real processing tasks to be implemented for ALICE. The initial results together with the discussion on algorithm optimizations and hardware limitations are also presented.