[llvm] [InstCombine] Don't require GEP in indexed compare fold (PR #81614)

[Nikita Popov via llvm-commits]

nikic wrote:

> > Huh, this seems to have a larger impact than I expected: http://llvm-compile-time-tracker.com/compare.php?from=f0db35b93f31ea5d6ff9bd4791fb6755b5a5bb9b&to=c77f948ef66114ea99f0daf5a72d3646bb5845ff&stat=instructions%3Au
> > clang stage2 becomes 0.6% larger and 0.5% faster as a result of this change.
> 
> stage2 is using the bootstrapped clang with this commit?

Yes.

> Is the `clang build:instruction:u` the change in instructions for stage1 to build stage2? If so seems like there is also a .3% regression there?

Right. This is likely correlated with the code size increase.

https://github.com/llvm/llvm-project/pull/81614