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

[David Green via llvm-commits]

davemgreen wrote:

> I tried both options (max epilogue trip count and average, i.e, / 2 ) on some micro benchmarks, https://github.com/llvm/llvm-test-suite/pull/165, and it looks like avg is strictly worse

Thanks for checking. I don't know how much I would trust micro-benchmarks, but that probably makes sense if it was using the average factor. I imagine what we would really want would not be the "performance of the average trip count", but the "average performance of all the trip counts". The maximum might well be the best at approximating of that.

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