New instruction for inter-CTA barrier in future GPUs?

[wzhcz8902]

cutlass/include/cutlass/semaphore.h

Line 52 in 5c447dd

/// CTA-wide semaphore for inter-CTA synchronization.

I saw similar synchronization mechanism using semaphores between CTAs in cub, do you think it might be good to have an instruction that does this, or will this be a hardware feature in the future generation of GPUs?

[thakkarV]

Hopper adds a bunch of stuff for accelerated intern cluster sync