[llvm] [LV, VP]VP intrinsics support for the Loop Vectorizer + adding new tail-folding mode using EVL. (PR #76172)

[via llvm-commits]

================
@@ -5342,6 +5394,10 @@ LoopVectorizationCostModel::selectInterleaveCount(ElementCount VF,
   if (!isScalarEpilogueAllowed())
     return 1;
 
+  // Do not interleave if EVL is preferred and no User IC is specified.
----------------
ayalz wrote:

The decision to avoiding unrolling due to EVL may be worth a DEBUG message.

https://github.com/llvm/llvm-project/pull/76172