https://hgpu.org/?p=26930
CoDL: Efficient CPU-GPU Co-execution for Deep Learning Inference on Mobile Devices