Hey everyone
Does anybody know the exact difference between the 3 options CuDNN offers from conv layer forward please?
The 3 options are: Implicit_gemm, Implicit_precomp_gemm, gemm
I know that gemm is the caffe's default way of doing im2col then matrix multiplication
What about the others please?