Hi, I’m trying to create custom variant of strided convolutions (fwd+bwd). I’m trying to find the code showing how Pytorch does it but find no relevant code in functional.py. I’ve been able to use e.g. torch._C._cudnn_convolution_backward_data but can’t find the files with the C code either (though I’m a novice at C).
My plan was to use this code as basis and then modify it as needed.