Why Slow than origin?

[StonepageVan]

I use your code but slow than the origin function do you know why? my code is self.cheap_operation = nn.Conv2d(init_channels, new_channels, dw_size, 1,dw_size // 2, groups=init_channels, bias=False) and I replace it with your function like this

self.cheap_operation = DepthwiseConv2d(init_channels, new_channels, dw_size, 1,dw_size // 2, groups=init_channels, bias=False)

do you know how to fix the problem?

[rosinality]

This is just a try to implement efficient kernels, and I cannot ensure this is faster than pytorch native kernels.