Reducing overheads of dynamic scheduling on heterogeneous chips