[llvm] [LV] Vectorize Epilogues for loops with small VF but high IC (PR #108190)

[Sjoerd Meijer via llvm-commits]

sjoerdmeijer wrote:

I first did a quick performance run for 527.cam4_r from SPEC FP. I was hoping it would avoid the regression that is result of my extra interleaving patch (#100385), but unfortunately this patch didn't make a difference. I.e., with the extra interleaving for the V2, this patch didn't manage to bring back the lost perf. 

What I will plan to do next:
- I haven't looked why this is not helping. I guess because epilogue vectorization is not kicking in, but will debug why that isn't the case,
- will kick off some more perf runs. 

https://github.com/llvm/llvm-project/pull/108190