How to understand "Fprop, Dgrad, Wgrad" in CUTLASS

hello, nv experts,
I’m reading the source code of CUTLASS about convolution.
I found there are 3 type convolutions in CUTLASS: Fprop, Dgrad, Wgrad。
I don’t know their fullname, and the differences of their implementation.
Is there anyone would like to teach me?

Quick Google search says (from the CUTLASS GitHub repository):

Operators: forward (Fprop), backward data gradient (Dgrad), and backward weight gradient (Wgrad) convolution

You are already on the right track: “Use the source, Luke!”


thank you