[llvm] Reapply "[LICM] Fold associative binary ops to promote code hoisting (#81608)" (PR #100377)

[Yingwei Zheng via llvm-commits]

dtcxzyw wrote:

Hi @rj-jesus @nikic

My CI reported that this patch caused significant performance regressions on `SingleSource/Benchmarks/Stanford/Queens` (+50% instructions) and `SingleSource/Benchmarks/Misc/himenobmtxpa` (+20% instructions). Can you have a look?
See also https://github.com/dtcxzyw/llvm-ci/issues/1176.



https://github.com/llvm/llvm-project/pull/100377