[llvm] [CGP]: Optimize mul.overflow. (PR #148343)

[David Green via llvm-commits]

================
@@ -322,8 +322,11 @@ class AArch64TargetLowering : public TargetLowering {
   }
 
   // Return true if the target wants to optimize the mul overflow intrinsic
-  // by detecting if there is no overflow.
-  bool shouldOptimizeMulOverflowIntrinsic() const override { return true; }
+  // for the given \p VT.
+  bool shouldOptimizeMulOverflowIntrinsic(LLVMContext &Context,
+                                          EVT VT) const override {
+    return getTypeAction(Context, VT) == TypeExpandInteger;
----------------
davemgreen wrote:

Does this just mean VT == MVT::i64? Does it apply to vector types? It might be simpler to be explicit.

https://github.com/llvm/llvm-project/pull/148343