[llvm] [msan] Implement support for avx512fp16.mask.{add/sub/mul/div/max/min}.sh.round (PR #137441)

[Florian Mayer via llvm-commits]

================
@@ -4312,6 +4312,65 @@ struct MemorySanitizerVisitor : public InstVisitor<MemorySanitizerVisitor> {
     setOriginForNaryOp(I);
   }
 
+  // For sh.* compiler intrinsics:
+  //   llvm.x86.avx512fp16.mask.{add/sub/mul/div/max/min}.sh.round
+  //     (<8 x half>, <8 x half>, <8 x half>, i8,  i32)
+  //      A           B           WriteThru   Mask RoundingMode
+  //
+  // DstShadow[0] = Mask[0] ? (AShadow[0] | BShadow[0]) : WriteThruShadow[0]
+  // DstShadow[1..7] = AShadow[1..7]
+  void visitGenericScalarHalfwordInst(IntrinsicInst &I) {
+    IRBuilder<> IRB(&I);
+
+    assert(I.arg_size() == 5);
+    Value *A = I.getOperand(0);
+    Value *B = I.getOperand(1);
+    Value *WriteThrough = I.getOperand(2);
+    Value *Mask = I.getOperand(3);
+    Value *RoundingMode = I.getOperand(4);
+
+    // Technically, we could probably just check whether the LSB is initialized
+    insertShadowCheck(Mask, &I);
----------------
fmayer wrote:

... but?

https://github.com/llvm/llvm-project/pull/137441