[llvm] [MachineCSE][AArch64] Coalesce subregister copies behind an opt-in target hook (PR #224593)

[Madhur Amilkanthwar via llvm-commits]

madhur13490 wrote:

@sjoerdmeijer Regressions? That is surprising. I might have missed something but if you are looking at test changes then those are purely because RA chose different registers now. Can you please point me to the particular case? 

By enlarge, if I read @arsenm's reply then I don't think the logic is in queation. The plan of productization is subject to debate due to PPC being a blocker. 

https://github.com/llvm/llvm-project/pull/224593