[llvm] [LoopVectorize] Enhance Vectorization decisions for predicate tail-folded loops with low trip counts (PR #69588)

[Igor Kirillov via llvm-commits]

igogo-x86 wrote:

> I might be missing something, but IIUC the patch may take different decisions in the presence of PGO data now, so there's some new behavior that should be testable?

Added a test that now shows the difference in decisions depending on tail-folding type

https://github.com/llvm/llvm-project/pull/69588