what problem will i meet when implementing an binconv2d op?

[zhuyinheng]

Thanks for all these work! It's the only xnor gpu kernel I found so far.
but I notice that:

# This is not a binary op right now...
                h_conv2 = tf.nn.relu(self.conv2d(h_pool1_bin, Wb_conv2))
                #h_conv2 = tf.nn.relu(self.conv2d(h_pool1, W_conv2))

and I plan to implement an binconv2 base on this repo(GPU kernel). I am wondering if there are some difficulty that is hard to solve? is that because of the limitation of kernel:

Limitations
XNOR GEMM op currently only works for square matrices that are powers of 2, with smallest N being 512.