Created by: grygielski
Add support for asymetric padding (along with "VALID" and "SAME" padding algorithms) in MKL-DNN operators.
Current state:
-
Pooling
-
Convolution
-
Convolution Transpose
MKL-DNN operators with asymetric padding support: pool2d, conv2d, conv3d, conv_transpose Possible improvements: Add more UTs
@jczaja @luotao1 @bingyanghuang please review