(X | C1) + C2 --> (X | C1) ^ C1 iff (C1 == -C2)
I verified the correctness using Alive:
https://rise4fun.com/Alive/YNV
This transform enables the following transform that already exists in
instcombine:
(X | Y) ^ Y --> X & ~Y
As a result, the full expected transform is:
(X | C1) + C2 --> X & ~C1 iff (C1 == -C2)
There already exists the transform in the sub case:
(X | Y) - Y --> X & ~Y
However this does not trigger in the case where Y is constant due to an earlier
transform:
X - (-C) --> X + C
With this new add fold, both the add and sub constant cases are handled.
Patch by Chris Dawson.
Differential Revision: https://reviews.llvm.org/D61517
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@360185
91177308-0d34-0410-b5e6-
96231b3b80d8
if (!match(Op1, m_APInt(C)))
return nullptr;
+ // (X | C2) + C --> (X | C2) ^ C2 iff (C2 == -C)
+ const APInt *C2;
+ if (match(Op0, m_Or(m_Value(), m_APInt(C2))) && *C2 == -*C)
+ return BinaryOperator::CreateXor(Op0, ConstantInt::get(Add.getType(), *C2));
+
if (C->isSignMask()) {
// If wrapping is not allowed, then the addition must set the sign bit:
// X + (signmask) --> X | signmask
// Is this add the last step in a convoluted sext?
// add(zext(xor i16 X, -32768), -32768) --> sext X
Type *Ty = Add.getType();
- const APInt *C2;
if (match(Op0, m_ZExt(m_Xor(m_Value(X), m_APInt(C2)))) &&
C2->isMinSignedValue() && C2->sext(Ty->getScalarSizeInBits()) == *C)
return CastInst::Create(Instruction::SExt, X, Ty);
; (X | C1) + C2 --> (X | C1) ^ C1 iff (C1 == -C2)
define i32 @test44(i32 %A) {
; CHECK-LABEL: @test44(
-; CHECK-NEXT: [[B:%.*]] = or i32 [[A:%.*]], 123
-; CHECK-NEXT: [[C:%.*]] = add nsw i32 [[B]], -123
-; CHECK-NEXT: ret i32 [[C]]
+; CHECK-NEXT: [[TMP1:%.*]] = and i32 [[A:%.*]], -124
+; CHECK-NEXT: ret i32 [[TMP1]]
;
%B = or i32 %A, 123
%C = add i32 %B, -123
define i32 @test44_extra_use(i32 %A) {
; CHECK-LABEL: @test44_extra_use(
; CHECK-NEXT: [[B:%.*]] = or i32 [[A:%.*]], 123
-; CHECK-NEXT: [[C:%.*]] = add nsw i32 [[B]], -123
-; CHECK-NEXT: [[D:%.*]] = mul i32 [[B]], [[C]]
+; CHECK-NEXT: [[TMP1:%.*]] = and i32 [[A]], -124
+; CHECK-NEXT: [[D:%.*]] = mul i32 [[B]], [[TMP1]]
; CHECK-NEXT: ret i32 [[D]]
;
%B = or i32 %A, 123
define <2 x i32> @test44_vec(<2 x i32> %A) {
; CHECK-LABEL: @test44_vec(
-; CHECK-NEXT: [[B:%.*]] = or <2 x i32> [[A:%.*]], <i32 123, i32 123>
-; CHECK-NEXT: [[C:%.*]] = add nsw <2 x i32> [[B]], <i32 -123, i32 -123>
-; CHECK-NEXT: ret <2 x i32> [[C]]
+; CHECK-NEXT: [[TMP1:%.*]] = and <2 x i32> [[A:%.*]], <i32 -124, i32 -124>
+; CHECK-NEXT: ret <2 x i32> [[TMP1]]
;
%B = or <2 x i32> %A, <i32 123, i32 123>
%C = add <2 x i32> %B, <i32 -123, i32 -123>
; Check (X | Y) - Y --> X & ~Y when Y is a constant
define i32 @test70(i32 %A) {
; CHECK-LABEL: @test70(
-; CHECK-NEXT: [[B:%.*]] = or i32 [[A:%.*]], 123
-; CHECK-NEXT: [[C:%.*]] = add nsw i32 [[B]], -123
-; CHECK-NEXT: ret i32 [[C]]
+; CHECK-NEXT: [[TMP1:%.*]] = and i32 [[A:%.*]], -124
+; CHECK-NEXT: ret i32 [[TMP1]]
;
%B = or i32 %A, 123
%C = sub i32 %B, 123
define { i8, i1 } @uadd_always_overflow(i8 %x) nounwind {
; CHECK-LABEL: @uadd_always_overflow(
-; CHECK-NEXT: [[Y:%.*]] = or i8 [[X:%.*]], -64
-; CHECK-NEXT: [[A:%.*]] = add nsw i8 [[Y]], 64
-; CHECK-NEXT: [[TMP1:%.*]] = insertvalue { i8, i1 } { i8 undef, i1 true }, i8 [[A]], 0
-; CHECK-NEXT: ret { i8, i1 } [[TMP1]]
+; CHECK-NEXT: [[TMP1:%.*]] = and i8 [[X:%.*]], 63
+; CHECK-NEXT: [[TMP2:%.*]] = insertvalue { i8, i1 } { i8 undef, i1 true }, i8 [[TMP1]], 0
+; CHECK-NEXT: ret { i8, i1 } [[TMP2]]
;
%y = or i8 %x, 192
%a = call { i8, i1 } @llvm.uadd.with.overflow.i8(i8 %y, i8 64)