[llvm] X86: Emit adox instead of adc for overflow add (PR #216609)

[Craig Topper via llvm-commits]

================
@@ -59769,6 +59786,47 @@ static SDValue combineADC(SDNode *N, SelectionDAG &DAG,
   return SDValue();
 }
 
+// Optimize RES, EFLAGS = X86ISD::ADOX LHS, RHS, EFLAGS
+static SDValue combineADOX(SDNode *N, SelectionDAG &DAG) {
+  SDValue LHS = N->getOperand(0);
+  SDValue RHS = N->getOperand(1);
+  SDValue CarryIn = N->getOperand(2);
+  auto *LHSC = dyn_cast<ConstantSDNode>(LHS);
+  auto *RHSC = dyn_cast<ConstantSDNode>(RHS);
+
+  // Canonicalize constant to RHS.
+  if (LHSC && !RHSC)
+    return DAG.getNode(X86ISD::ADOX, SDLoc(N), N->getVTList(), RHS, LHS,
+                       CarryIn);
+
+  // Fold ADOX(ADD(X,Y),0,Carry) -> ADOX(X,Y,Carry)
+  // ISD::ADD produces no EFLAGS, so we only need to ensure the ADOX's own
+  // flag output is not needed before replacing it with a 2-operand ADOX.
+  SDValue AddOp = LHS;
+
+  // Unwrap extensions. This changes what the wide intermediate represents,
+  // but it is safe because X86ISD::ADOX currently only has one producer
----------------
topperc wrote:

We shouldn't make assumptions like this. This is the kind of thing that's easy to break in the future.

https://github.com/llvm/llvm-project/pull/216609