[llvm] [X86] Optimized ADD + ADC to ADC (PR #173543)

[Sebastian Buchwald via llvm-commits]

================
@@ -18,15 +18,18 @@ define i32 @test_i32_add_add_idx(i32 %x, i32 %y, i32 %z) nounwind {
 ; X86-LABEL: test_i32_add_add_idx:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    addl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    btl $30, {{[0-9]+}}(%esp)
-; X86-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    adcl $0, %eax
----------------
UniQP wrote:

This should not change.

It looks like `needCarryOrOverflowFlag` looks at users of the `SDNode` rather than at users of the passed `SDValue`. The following change should fix this:
```diff
 static bool needCarryOrOverflowFlag(SDValue Flags) {
   assert(Flags.getValueType() == MVT::i32 && "Unexpected VT!");
 
-  for (const SDNode *User : Flags->users()) {
+  for (const SDUse &Use : Flags->uses()) {
+    if (Use != Flags)
+      continue;
+    const SDNode* User = Use.getUser();
     X86::CondCode CC;
     switch (User->getOpcode()) {
     default:
```

https://github.com/llvm/llvm-project/pull/173543