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

[Matt Arsenault via llvm-commits]

arsenm wrote:

> That is an indirect way doing that.
> 

Yes, that's my point, it's redundant. 

> What this patch is doing can allow us not to set `.hidden_queue_ptr` unnecessarily. Currently we set it when the function attribute `amdgpu-no-queue-ptr` is absent even for GFX9+ with COV5+. 


> Right. We should only use the implicitarg field for queue_ptr for cov5. I do remember I have to update many LIT tests due to less sgpr usage.

But you still can use the queue pointer, it's not gone. The logic should be able to differentiate should not use queue pointer and queue pointer does not exist 



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