Parallel Likelihood Function Evaluation on Heterogeneous Many-core Systems

This paper describes a parallel implementation that allows the evalua- tions of the likelihood function for data analysis methods t o run cooperatively on heterogeneous computational devices (i.e. CPU and GPU) belonging to a single computational node. The implementation is able to split and balance the work- load needed for the evaluation of the likelihood function in corresponding sub- workloads to be executed in parallel on each computational device. The CPU paral- lelization is implemented using OpenMP, while the GPU implementation is based on OpenCL. The comparison of the performance of these implementations for dif- ferent configurations and different hardware systems are re ported. Tests are based on a real data analysis carried out in the high energy physics community.