Parallel and high-performance computing

A heterogeneous system combines general-purpose CPU cores with a GPU accelerator. Which workload division is most appropriate when the computation contains both irregular control flow and large regular data-parallel regions?