[llvm] [AMDGPU] Balance VM_CNT histories across branches (PR #221115)

[Dan Zimmerman via llvm-commits]

danzimm wrote:

Thanks for the link @arsenm -- I didn't appropriately mark this PR, apologies, I didn't know the annotation was needed. I'll update the summary too, but codex helped write this. I audited every line, of course, and made edits of my own, too. I've only worked on closed source forks, before, so if there's code smell it's (fortunately or unfortunately depending on your point of view) due to myself more than AI.

Thanks for taking a look at the PR more closely this time. Does it sound reasonable to sort out the conversation with Pierre first? Or should I respond/address your comments in parallel? On that topic, do you have any thoughts to the comment I left above responding to Pierre?

https://github.com/llvm/llvm-project/pull/221115