Deep Learning Inference on Heterogeneous Mobile Processors: Potentials and Pitfalls