[QST] [CuTeDSL] Redudant wait in Hopper example

[simveit]

In the Hopper example we perform (in L859)

cute.nvgpu.warpgroup.wait_group(k_pipe_mmas)

As I understand it this is not necessary because we wait for all the commited wgmma instructions before we perform the epilog (in L918):

cute.nvgpu.warpgroup.wait_group(0)

[tridao]

No it's necessary, you need to make sure that wgmma has finished before releasing the smem

mainloop_pipeline.consumer_release(mainloop_consumer_release_state)

[simveit]

thanks @tridao