[llvm] [AMDGPU] Lower `llvm.amdgcn.queue.ptr` instrinsic to using implicit kernel argument if feasible (PR #103490)

[Matt Arsenault via llvm-commits]

https://github.com/arsenm commented:

I think the intrinsic should be interpreted as the user SGPR value. There is already a representation for load from implicit argument pointer, and we don't need 2 ways of doing that.

https://github.com/llvm/llvm-project/pull/103490