[InstCombine] Optimize 'xor-and/or-select' sequence to 'or' for bool

[Qi-Hu]

Optimize xor-and/or-select sequence to or : https://alive2.llvm.org/ce/z/52ZT62, https://alive2.llvm.org/ce/z/CHCJQd.

[llvmbot]

@llvm/pr-subscribers-llvm-transforms

Changes

Optimize `xor`-`and`-`select` sequence to `or` : https://alive2.llvm.org/ce/z/52ZT62 -- Full diff: https://github.com//pull/66394.diff

2 Files Affected:

• (modified) llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp (+8)
• (added) llvm/test/Transforms/InstCombine/fold-xor-and-select-i1.ll (+14)

diff --git a/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp b/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
index e2965e7a5703976..8a0ba6f766b4ad1 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
@@ -3202,6 +3202,14 @@ Instruction *InstCombinerImpl::foldSelectOfBools(SelectInst &SI) {
     }
   }
 
+  // select (~b & a), a, b -> or a, b
+  // only for scalar types
+  if (match(CondVal, m_And(m_Not(m_Specific(FalseVal)), m_Specific(TrueVal))) &&
+      TrueVal->getType()->isIntegerTy(1) &&
+      FalseVal->getType()->isIntegerTy(1)) {
+    return BinaryOperator::CreateOr(TrueVal, FalseVal);
+  }
+
   return nullptr;
 }
 
diff --git a/llvm/test/Transforms/InstCombine/fold-xor-and-select-i1.ll b/llvm/test/Transforms/InstCombine/fold-xor-and-select-i1.ll
new file mode 100644
index 000000000000000..a5870a3ce9d973f
--- /dev/null
+++ b/llvm/test/Transforms/InstCombine/fold-xor-and-select-i1.ll
@@ -0,0 +1,14 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -passes=instcombine -S | FileCheck %s
+
+define i1 @max_if(i1 %a, i1 %b) {
+; CHECK-LABEL: define i1 @max_if
+; CHECK-SAME: (i1 [[A:%.*]], i1 [[B:%.*]]) {
+; CHECK-NEXT:    [[TMP1:%.*]] = or i1 [[A]], [[B]]
+; CHECK-NEXT:    ret i1 [[TMP1]]
+;
+  %1 = xor i1 %b, true
+  %cmp = and i1 %1, %a
+  %2 = select i1 %cmp, i1 %a, i1 %b
+  ret i1 %2
+}

[bryanpkc]

This pattern is similar to, but more specific than, the one on line 3109. You should probably move this transform above that one.

[bryanpkc]

Correction: Both your patterns are more specific than the ones between lines 3089 and 3109. Move your code above those.

[bryanpkc]

These scalar type checks are not necessary. When given vector operands, the select instruction will perform the selection element by element.

[bryanpkc]

Add your tests into llvm/test/Transforms/InstCombine/select-and-or.ll, instead of creating new files.

[nikic]

Hasn't this already been implemented in https://reviews.llvm.org/D158983?

Please split test additions into a separate commit so the actual impact of the change can be seen. If the only benefit of the additional transform is replacing a logical or with a bitwise or, then I don't think this is needed.

[Qi-Hu]

Hasn't this already been implemented in https://reviews.llvm.org/D158983?

Please split test additions into a separate commit so the actual impact of the change can be seen. If the only benefit of the additional transform is replacing a logical or with a bitwise or, then I don't think this is needed.

Indeed, this is already implemented in D158983. Thanks for pointing it out. I'll close this PR.