[llvm] [InstCombine] Fold xor(zext(or disjoint X, C1), C2) (PR #214970)

[Andreas Jonson via llvm-commits]

=?utf-8?b?7KGw7JiB7KeEKEpveW91bmdqaW4p?Message-ID:
In-Reply-To: <llvm.org/llvm/llvm-project/pull/214970 at github.com>


================
@@ -1775,6 +1775,27 @@ static Instruction *foldLogicCastConstant(BinaryOperator &Logic, CastInst *Cast,
   Type *DestTy = Logic.getType();
   Type *SrcTy = Cast->getSrcTy();
 
+  // xor (zext (or disjoint X, NarrowC)), WideC
+  //   -> xor (zext X), WideC ^ zext(NarrowC)
+  if (LogicOpc == Instruction::Xor && SrcTy->isIntegerTy() &&
+      DestTy->isIntegerTy()) {
+    Value *X;
+    ConstantInt *NarrowC, *WideC;
+
+    if (match(C, m_ConstantInt(WideC)) &&
+        match(Cast,
+              m_OneUse(m_ZExt(m_OneUse(
+                  m_c_DisjointOr(m_Value(X), m_ConstantInt(NarrowC))))))) {
----------------
andjo403 wrote:

no need to use the m_c_DisjointOr as constants is swaped to be on the right before his fold
and if m_APInt is used instead this also works for splat vectors so then there is no need to check the types as zext must be integer or vector integers
```suggestion
                  m_DisjointOr(m_Value(X), m_APInt(NarrowC))))))) {
```

https://github.com/llvm/llvm-project/pull/214970