Extension with OpenCL of the two-level MPI+OpenMP parallelization for CFD simulations on heterogeneous systems

The paper is devoted to the extended parallelization for numerical experiments in fluid dynamics and aeroacoustics on heterogeneous systems that combines computing units of different architectures, namely CPU and GPU. A hybrid two-level MPI+OpenMP parallel model is extended with OpenCL in order to engage GPUs. In doing so the third level of parallelism appears. A model of an algorithm for unstructured meshes is presented.