Compose Softmax Mainloop with Collective

[osayamenja]

Hello team!

I want to perform a fused gemm + softmax. I noticed CUTLASS provides an implementation here. So, how would I compose DefaultMmaSoftmaxMainloopFusion with a collective?

My understanding from reading the code is that the fusion is a collective itself that I can construct with its type arguments, is this correct?

I checked thoroughly and did not find any examples but please, refer me if there are any!