foldedNhwcToNhwcKernel<__half, __half, float, bool=1> shows up on profiler timeline

cudnnConvolutionBackwardData() setup to run in NHWC for some reason triggers these kernels after each ALGO1, any suggestions what these are?

The name itself implies conversion from nhwc to nhwc, which doesn’t seem to make much sense.

Could you please let us know if you are still facing this issue?