Torch.istft with real input


I note this line in Pytorch documentation for torch.istft about the input of this function:

Deprecated since version 1.8.0: Real input is deprecated, use complex inputs as returned by stft(..., return_complex=True) instead.

I have some codes which do use real input with torch.istft. Will it produce different results from using this function with complex input after version 1.8.0? Will the backward gradient be different in training?