[llvm] [AMDGPU] Skip float and double atomicrmw fadd expansion on gfx12 (PR #100217)

[Stanislav Mekhanoshin via llvm-commits]

rampitec wrote:

> > Please just close this and do not touch any atomic code. I have a large stack starting at #96759 that replaces all of this
> 
> OK, but I do not see there changes to the gfx12 tests similar to what I have here.

I see now, it is in another patch: https://github.com/llvm/llvm-project/pull/96760

https://github.com/llvm/llvm-project/pull/100217