[llvm] [NFC][LV]: Differentiate between Pred and Unpred masked operations. (PR #169509)

[Hassnaa Hamdi via llvm-commits]

================
@@ -714,9 +717,16 @@ class LoopVectorizationLegality {
   AssumptionCache *AC;
 
   /// While vectorizing these instructions we have to generate a
-  /// call to the appropriate masked intrinsic or drop them in case of
-  /// conditional assumes.
-  SmallPtrSet<const Instruction *, 8> MaskedOp;
+  /// call to the appropriate masked intrinsic or drop them.
+  /// In order to differentiate between control flow introduced at the source
+  /// level and that introduced by the loop vectoriser during tail-folding, we
+  /// keep two lists:
+  /// 1) UnpredMaskedOp - instructions that need masking if we are
+  ///    in conditionally executed block.
+  /// 2) PredMaskedOp - instructions that need masking if we are in
+  ///    a predicated loop.
+  SmallPtrSet<const Instruction *, 8> UnpredMaskedOps;
+  SmallPtrSet<const Instruction *, 8> PredMaskedOps;
----------------
hassnaaHamdi wrote:

These are the names I used in the draft patch:
`SmallPtrSet<const Instruction *, 8> MaskedOp;`
 ` SmallPtrSet<const Instruction *, 8> TailFoldedMaskedOp;`
I think these are better ?

https://github.com/llvm/llvm-project/pull/169509