[InstCombine] Fold Xor with or disjoint

Implement a missing optimization to fold (A | B) ^ C to (A ^ C) ^ B

Author: AmrDeveloper

llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp

@@ -4693,7 +4693,20 @@ Instruction *InstCombinerImpl::visitXor(BinaryOperator &I) {
   // calls in there are unnecessary as SimplifyDemandedInstructionBits should
   // have already taken care of those cases.
   Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);
-  Value *M;
+  Value *X, *Y, *M;
+
+  // (X | Y) ^ M -> (X ^ M) ^ Y
+  // (X | Y) ^ M -> (Y ^ M) ^ X
+  if (match(&I, m_c_Xor(m_OneUse(m_DisjointOr(m_Value(X), m_Value(Y))),
+                        m_Value(M)))) {
+    if (Value *XorAC =
+            simplifyBinOp(Instruction::Xor, X, M, SQ.getWithInstruction(&I)))
+      return BinaryOperator::CreateXor(XorAC, Y);
+
+    if (Value *XorBC = simplifyBinOp(Instruction::Xor, Y, M, SQ))
+      return BinaryOperator::CreateXor(XorBC, X);
+  }
+
   if (match(&I, m_c_Xor(m_c_And(m_Not(m_Value(M)), m_Value()),
                         m_c_And(m_Deferred(M), m_Value())))) {
     if (isGuaranteedNotToBeUndef(M))
@@ -4705,7 +4718,6 @@ Instruction *InstCombinerImpl::visitXor(BinaryOperator &I) {
   if (Instruction *Xor = visitMaskedMerge(I, Builder))
     return Xor;
 
-  Value *X, *Y;
   Constant *C1;
   if (match(Op1, m_Constant(C1))) {
     Constant *C2;

llvm/test/Transforms/InstCombine/xor.ll

@@ -1489,48 +1489,51 @@ define i4 @PR96857_xor_without_noundef(i4  %val0, i4  %val1, i4 %val2) {
 define i32 @or_disjoint_with_xor(i32 %a, i32 %b) {
 ; CHECK-LABEL: @or_disjoint_with_xor(
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[TMP0:%.*]] = xor i32 [[A:%.*]], -1
-; CHECK-NEXT:    [[XOR:%.*]] = and i32 [[B:%.*]], [[TMP0]]
-; CHECK-NEXT:    ret i32 [[XOR]]
+; CHECK-NEXT:    ret i32 [[B:%.*]]
 ;
 entry:
   %or = or disjoint i32 %a, %b
   %xor = xor i32 %or, %a
   ret i32 %xor
 }
 
-define i32 @xor_with_or_disjoint(i32 %a, i32 %b, i32 %c) {
-; CHECK-LABEL: @xor_with_or_disjoint(
+define i32 @xor_with_or_disjoint_ab(i32 %a, i32 %b) {
+; CHECK-LABEL: @xor_with_or_disjoint_ab(
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[TMP0:%.*]] = xor i32 [[A:%.*]], -1
-; CHECK-NEXT:    [[XOR:%.*]] = and i32 [[B:%.*]], [[TMP0]]
-; CHECK-NEXT:    ret i32 [[XOR]]
+; CHECK-NEXT:    ret i32 [[B:%.*]]
 ;
 entry:
   %or = or disjoint i32 %a, %b
   %xor = xor i32 %a, %or
   ret i32 %xor
 }
 
-define <2 x i32> @or_disjoint_with_xor_vec(<2 x i32> %a, < 2 x i32> %b, <2 x i32> %c) {
+define i32 @xor_with_or_disjoint_ba(i32 %a, i32 %b) {
+; CHECK-LABEL: @xor_with_or_disjoint_ba(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    ret i32 [[B:%.*]]
+;
+entry:
+  %or = or disjoint i32 %b, %a
+  %xor = xor i32 %b, %or
+  ret i32 %xor
+}
+
+define <2 x i32> @or_disjoint_with_xor_vec(<2 x i32> %a, < 2 x i32> %b) {
 ; CHECK-LABEL: @or_disjoint_with_xor_vec(
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[TMP0:%.*]] = xor <2 x i32> [[A:%.*]], <i32 -1, i32 -1>
-; CHECK-NEXT:    [[XOR:%.*]] = and <2 x i32> [[B:%.*]], [[TMP0]]
-; CHECK-NEXT:    ret <2 x i32> [[XOR]]
+; CHECK-NEXT:    ret <2 x i32> [[B:%.*]]
 ;
 entry:
   %or = or disjoint <2 x i32> %a, %b
   %xor = xor <2 x i32> %or, %a
   ret <2 x i32> %xor
 }
 
-define <2 x i32> @xor_with_or_disjoint_vec(<2 x i32> %a, < 2 x i32> %b, <2 x i32> %c) {
+define <2 x i32> @xor_with_or_disjoint_vec(<2 x i32> %a, < 2 x i32> %b) {
 ; CHECK-LABEL: @xor_with_or_disjoint_vec(
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[TMP0:%.*]] = xor <2 x i32> [[A:%.*]], <i32 -1, i32 -1>
-; CHECK-NEXT:    [[XOR:%.*]] = and <2 x i32> [[B:%.*]], [[TMP0]]
-; CHECK-NEXT:    ret <2 x i32> [[XOR]]
+; CHECK-NEXT:    ret <2 x i32> [[B:%.*]]
 ;
 entry:
   %or = or disjoint <2 x i32> %a, %b