The problem about fusing depthwise convolution layer and convolution layer

[yangjing69]

1. We analyzed the code and find that ChaiDNN v2 fuse depthwise convolution layer with convolution layer, and the quantization mode of depthwise convolution is Dynamic only. When we use Dynamic quantization method, we find that the calculation result of depthwise convolution layer and convolution layer fusion becomes all zero, why?
2. After depthwise convolution layer with convolution layer fusion, does the quantization method of depthwise convolution layer support the Xilinx quantization method?