A taxonomy of approaches for workload placement on heterogeneous CPU-GPU systems is available at link