[llvm] [LoopVectorizer] Prune VFs based on plan register pressure (PR #132190)

[Sam Tebbs via llvm-commits]

SamTebbs33 wrote:

> This is causing serious (~40-50%) regressions in the LCALS benchmarks from the llvm-test-suite (https://github.com/llvm/llvm-test-suite/tree/main/MicroBenchmarks/LCALS) on x86. `BM_VOL3D_CALC_LAMBDA` and `BM_VOL3D_CALC_RAW` are affected.
> 
> Is this the same problem as the one being addressed in #141736?

It looks like it could be the same problem since X86 doesn't enable maximizing vector bandwidth.

https://github.com/llvm/llvm-project/pull/132190