[llvm] AMDGPU: Remove flat/global fmin/fmax intrinsics (PR #105642)

[Matt Arsenault via llvm-commits]

arsenm wrote:

> @arsenm I am not ok with the removal of these intrinsics that allow direct call to some of these atomics. I think you should keep them while you still do the changes that you want with the other atomicrmw. 

The changes are already completed and we are at the final point of removal. 

> Your change is not backwards compatible. It is not yet clear to me if calling create atomicrmw is equivalent to calling the old intrisincs.

The intrinsics are strictly less expressive than atomicrmw. There is no plus to using the intrinsics, and it is trivial to migrate to an atomicrmw that will produce the ISA instruction. The old intrinsics had broken scope and ordering arguments which did not work. You only need to decide whether the use case is OK with breaking remote memory, fine grained memory, or both.

https://github.com/llvm/llvm-project/pull/105642