[llvm] [AMDGPU] Add writelane and readlane pseudos for SGPR spilling (PR #69923)

[Carl Ritson via llvm-commits]

https://github.com/perlfu commented:

Change looks mostly fine to me, splitting these out to dedicated pseudos certainly makes sense; however, I have some doubts (in line) about whether we want to treat the pseudos exactly like v_readlane/v_writelane.

Also curious about the contents of the follow up patch dependent on this?
Having spent a bit of time investigating bugs in shaders with heavy SGPR in the last few weeks, I have a few thoughts about optimizing away some of these v_readlane/v_writelane operations, but perhaps you are already working on that?

https://github.com/llvm/llvm-project/pull/69923