Offloading computational tasks of hybrid MPI+OpenMP/OmpSs-2 applications to GPUs