[InstCombine] add (ashr (shl i32 X, 31), 31), 1 --> and (not X), 1

This is another step towards favoring 'not' ops over random 'xor' in IR:
https://bugs.llvm.org/show_bug.cgi?id=32706

This transformation may have occurred in longer IR sequences using computeKnownBits,
but that could be much more expensive to calculate.

As the scalar result shows, we do not currently favor 'not' in all cases. The 'not'
created by the transform is transformed again (unnecessarily). Vectors don't have
this problem because vectors are (wrongly) excluded from several other combines.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@302659 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/InstCombine/InstCombineAddSub.cpp b/lib/Transforms/InstCombine/InstCombineAddSub.cpp
index 659b6c3..3709360 100644 (file)
--- a/lib/Transforms/InstCombine/InstCombineAddSub.cpp
+++ b/lib/Transforms/InstCombine/InstCombineAddSub.cpp
@@ -1079,6 +1079,16 @@ static Instruction *foldAddWithConstant(BinaryOperator &Add,
     return new ZExtInst(Builder.CreateNUWAdd(X, NewC), Ty);
   }
 
+  // Shifts and add used to flip and mask off the low bit:
+  // add (ashr (shl i32 X, 31), 31), 1 --> and (not X), 1
+  const APInt *C3;
+  if (*C == 1 && match(Op0, m_OneUse(m_AShr(m_Shl(m_Value(X), m_APInt(C2)),
+                                            m_APInt(C3)))) &&
+      C2 == C3 && *C2 == Ty->getScalarSizeInBits() - 1) {
+    Value *NotX = Builder.CreateNot(X);
+    return BinaryOperator::CreateAnd(NotX, ConstantInt::get(Ty, 1));
+  }
+
   return nullptr;
 }
 

diff --git a/test/Transforms/InstCombine/add.ll b/test/Transforms/InstCombine/add.ll
index 955ffaa..5f7101e 100644 (file)
--- a/test/Transforms/InstCombine/add.ll
+++ b/test/Transforms/InstCombine/add.ll
@@ -31,9 +31,8 @@ define <2 x i32> @select_0_or_1_from_bool_vec(<2 x i1> %x) {
 
 define i32 @flip_and_mask(i32 %x) {
 ; CHECK-LABEL: @flip_and_mask(
-; CHECK-NEXT:    [[SHL:%.*]] = shl i32 %x, 31
-; CHECK-NEXT:    [[SHR:%.*]] = ashr exact i32 [[SHL]], 31
-; CHECK-NEXT:    [[INC:%.*]] = add nsw i32 [[SHR]], 1
+; CHECK-NEXT:    [[TMP1:%.*]] = and i32 %x, 1
+; CHECK-NEXT:    [[INC:%.*]] = xor i32 [[TMP1]], 1
 ; CHECK-NEXT:    ret i32 [[INC]]
 ;
   %shl = shl i32 %x, 31
@@ -44,9 +43,8 @@ define i32 @flip_and_mask(i32 %x) {
 
 define <2 x i8> @flip_and_mask_splat(<2 x i8> %x) {
 ; CHECK-LABEL: @flip_and_mask_splat(
-; CHECK-NEXT:    [[SHL:%.*]] = shl <2 x i8> %x, <i8 7, i8 7>
-; CHECK-NEXT:    [[SHR:%.*]] = ashr exact <2 x i8> [[SHL]], <i8 7, i8 7>
-; CHECK-NEXT:    [[INC:%.*]] = add nsw <2 x i8> [[SHR]], <i8 1, i8 1>
+; CHECK-NEXT:    [[TMP1:%.*]] = xor <2 x i8> %x, <i8 1, i8 1>
+; CHECK-NEXT:    [[INC:%.*]] = and <2 x i8> [[TMP1]], <i8 1, i8 1>
 ; CHECK-NEXT:    ret <2 x i8> [[INC]]
 ;
   %shl = shl <2 x i8> %x, <i8 7, i8 7>