Can you control conv algorithm using CUDNN API?

I would like to know if you can choose the algorithm viz. FFT, Winograd etc. used for convolution in C++ API for Pytorch using CUDNN API as shown here.


This is not possible at the moment. There are a couple of issues open for this: and

Thanks I came across those and thought may be its possible to have fine grained control from C++ API instead of python. But it seems that C++ API just exposes the C++ backend of the same python frontend.