[InstCombine] Fold (A & ~B) | (A ^ B) -> A ^ B

[wentywenty]

This patch implements the fold:
(A & ~B) | (A ^ B) -> A ^ B

The term (A & ~B) implies (A=1, B=0), which is already covered by
the XOR operation (A=1, B=0) | (A=0, B=1).
Therefore, the OR and the AND are redundant, leaving just A ^ B.

Related issue: #113711

[github-actions]

Thank you for submitting a Pull Request (PR) to the LLVM Project!

This PR will be automatically labeled and the relevant teams will be notified.

If you wish to, you can add reviewers by using the "Reviewers" section on this page.

If this is not working for you, it is probably because you do not have write permissions for the repository. In which case you can instead tag reviewers by name in a comment by using @ followed by their GitHub username.

If you have received no comments on your PR for a week, you can request a review by "ping"ing the PR by adding a comment “Ping”. The common courtesy "ping" rate is once a week. Please remember that you are asking for valuable time from other developers.

If you have further questions, they may be answered by the LLVM GitHub User Guide.

You can also ask questions in a comment on this PR, on the LLVM Discord or on the forums.

[llvmbot]

@llvm/pr-subscribers-llvm-transforms

Author: wentywenty (wentywenty)

Changes

[InstCombine] Drop one-use check for (A & ~B) | (A ^ B) -> A ^ B

This patch removes the m_OneUse constraint for the fold:
(A & ~B) | (A ^ B) -> A ^ B

Previously, this optimization was limited to cases where the operands had
a single use. However, replacing the 'or' with the existing 'xor' operand
is always profitable because it strictly removes one instruction (the 'or'),
regardless of whether the intermediate 'and' or 'xor' are used elsewhere.

Tests have been updated to reflect this multi-use optimization.

Related issue: #113711

---

Full diff: https://github.com/llvm/llvm-project/pull/171047.diff

2 Files Affected:

• (modified) llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp (+5)
• (added) llvm/test/Transforms/InstCombine/or-xor-fold.ll (+58)

diff --git a/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp b/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
index ba5568b00441b..821086879339c 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
@@ -1982,6 +1982,11 @@ static Instruction *foldOrToXor(BinaryOperator &I,
       match(Op1, m_c_And(m_Not(m_Specific(A)), m_Specific(B))))
     return BinaryOperator::CreateXor(A, B);
 
+  // (A & ~B) | (A ^ B) -> A ^ B
+  if (match(Op0, m_c_And(m_Value(A), m_Not(m_Value(B)))) &&
+      match(Op1, m_c_Xor(m_Specific(A), m_Specific(B))))
+    return replaceInstUsesWith(I, Op1);
+
   return nullptr;
 }
 
diff --git a/llvm/test/Transforms/InstCombine/or-xor-fold.ll b/llvm/test/Transforms/InstCombine/or-xor-fold.ll
new file mode 100644
index 0000000000000..57797a33335b8
--- /dev/null
+++ b/llvm/test/Transforms/InstCombine/or-xor-fold.ll
@@ -0,0 +1,58 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -passes=instcombine -S | FileCheck %s
+
+declare void @use(i32)
+
+; (A & ~B) | (A ^ B) -> A ^ B
+
+define i32 @test_basic(i32 %a, i32 %b) {
+; CHECK-LABEL: @test_basic(
+; CHECK-NEXT:    [[RES:%.*]] = xor i32 [[A:%.*]], [[B:%.*]]
+; CHECK-NEXT:    ret i32 [[RES]]
+;
+  %not_b = xor i32 %b, -1
+  %and = and i32 %a, %not_b
+  %xor = xor i32 %a, %b
+  %res = or i32 %and, %xor
+  ret i32 %res
+}
+
+define <2 x i32> @test_vector(<2 x i32> %a, <2 x i32> %b) {
+; CHECK-LABEL: @test_vector(
+; CHECK-NEXT:    [[RES:%.*]] = xor <2 x i32> [[A:%.*]], [[B:%.*]]
+; CHECK-NEXT:    ret <2 x i32> [[RES]]
+;
+  %not_b = xor <2 x i32> %b, <i32 -1, i32 -1>
+  %and = and <2 x i32> %a, %not_b
+  %xor = xor <2 x i32> %a, %b
+  %res = or <2 x i32> %and, %xor
+  ret <2 x i32> %res
+}
+
+define i32 @test_commute(i32 %a, i32 %b) {
+; CHECK-LABEL: @test_commute(
+; CHECK-NEXT:    [[RES:%.*]] = xor i32 [[A:%.*]], [[B:%.*]]
+; CHECK-NEXT:    ret i32 [[RES]]
+;
+  %not_b = xor i32 %b, -1
+  %and = and i32 %a, %not_b
+  %xor = xor i32 %a, %b
+  %res = or i32 %xor, %and  ;
+  ret i32 %res
+}
+
+define i32 @test_multiuse(i32 %a, i32 %b) {
+; CHECK-LABEL: @test_multiuse(
+; CHECK-NEXT:    [[NOT_B:%.*]] = xor i32 [[B:%.*]], -1
+; CHECK-NEXT:    [[AND:%.*]] = and i32 [[A:%.*]], [[NOT_B]]
+; CHECK-NEXT:    call void @use(i32 [[AND]])
+; CHECK-NEXT:    [[XOR:%.*]] = xor i32 [[A]], [[B]]
+; CHECK-NEXT:    ret i32 [[XOR]]
+;
+  %not_b = xor i32 %b, -1
+  %and = and i32 %a, %not_b
+  call void @use(i32 %and)
+  %xor = xor i32 %a, %b
+  %res = or i32 %and, %xor
+  ret i32 %res
+}

[nikic]

Please see https://llvm.org/docs/InstCombineContributorGuide.html#precommit-tests.

All of your tests already fold without your patch: https://llvm.godbolt.org/z/ffs3MYGGT