Fork自 PaddlePaddle / Paddle
Add the use_nnpack parameter in ExpandConvLayer, so that the convolution calculation can be switched to the NNPACK function.