How Can I Know Which Convolution Algo Was Used When I Set "torch.backends.cudnn.benchmark = True"?

It is validated that convolution process can be accelerated in Pytorch by setting “torch.backends.cudnn.benchmark = True”.I want to know which convolution algorithm(FFT, Winograd and GEMM)was selected. Moreover, if it is possible to set the convolution algo manually?
I wonder if there is any way to know which convolution algo was used, and how to set the specific algo manually.