[llvm] [LoopPeel] Peel to make Phis loop inductions (PR #121104)

[Ryotaro Kasuga via llvm-commits]

https://github.com/kasuga-fj commented:

> > Note that in some cases there was unnecessary peeling when tried with llvm-test-suite. The causes include peeling for a remainder loop of vectorization and the limitations of analysis by SCEV. However, as far as I've tried, these unnecessary peels do not affect performance.
>
> Do you have an example handy? Generally having some unnecessary peeling is expected, but peeling remainder loops is not great.

I found such cases when I first submitted this patch (I used SCEV to detect IVs at the time), but I just tried to run llvm-test-suite and it didn't reproduce. I'm now checking the details of the difference with and without this patch, but there does not appear to be a significant difference. The following loops were newly peeled.

- `MultiSource/Applications/ClamAV/libclamav_matcher-ac.c:888:6`
- `MultiSource/Applications/JM/ldecod/vlc.c:886:3`
- `MultiSource/Applications/JM/ldecod/vlc.c:961:3`
- `MultiSource/Applications/JM/lencod/cabac.c:1492:3`
- `MultiSource/Applications/oggenc/oggenc.c:47108:7`
- `MultiSource/Benchmarks/Prolangs-C/TimberWolfMC/finalpin.c:434:2`
- `MultiSource/Benchmarks/TSVC/CrossingThresholds-dbl/../tsc.inc:3310:3`
- `MultiSource/Benchmarks/TSVC/CrossingThresholds-dbl/../tsc.inc:3342:3`
- `MultiSource/Benchmarks/TSVC/CrossingThresholds-flt/../tsc.inc:3310:3`
- `MultiSource/Benchmarks/TSVC/CrossingThresholds-flt/../tsc.inc:3342:3`

https://github.com/llvm/llvm-project/pull/121104