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

[Florian Hahn via llvm-commits]

================
@@ -2518,6 +2567,10 @@ class VPDerivedIVRecipe : public VPSingleDefRecipe {
   /// for floating point inductions.
   const FPMathOperator *FPBinOp;
 
+  VPCanonicalIVPHIRecipe *getCanonicalIV() const {
----------------
fhahn wrote:

After this patch, it/s not guaranteed that the second operand is the canonical IV. Better to completely remove the helper. Also, the documentation for the recipe needs updating above.

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