[llvm] [msan] Generalize handlePairwiseShadowOrIntrinsic, and handle x86 pairwise add/sub (PR #127567)

[Thurston Dang via llvm-commits]

================
@@ -2608,8 +2608,15 @@ struct MemorySanitizerVisitor : public InstVisitor<MemorySanitizerVisitor> {
   /// e.g., <2 x i32> @llvm.aarch64.neon.saddlp.v2i32.v4i16(<4 x i16>)
   ///       <16 x i8> @llvm.aarch64.neon.addp.v16i8(<16 x i8>, <16 x i8>)
   ///
-  /// TODO: adapt this function to handle horizontal add/sub?
-  void handlePairwiseShadowOrIntrinsic(IntrinsicInst &I) {
+  /// Optionally, reinterpret the parameters to have elements of a specified
+  /// width. For example:
+  ///     @llvm.x86.ssse3.phadd.w(<1 x i64> [[VAR1]], <1 x i64> [[VAR2]])
+  /// conceptually operates on
+  ///     (<4 x i16> [[VAR1]], <4 x i16> [[VAR2]])
+  /// and can be handled with ReinterpretElemWidth == 16.
+  void
+  handlePairwiseShadowOrIntrinsic(IntrinsicInst &I,
----------------
thurstond wrote:

Done

https://github.com/llvm/llvm-project/pull/127567