diff --git a/llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp b/llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp --- a/llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp +++ b/llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp @@ -616,6 +616,19 @@ if (match(I->getOperand(1), m_APInt(SA))) { uint64_t ShiftAmt = SA->getLimitedValue(BitWidth-1); + // If we are just demanding the shifted sign bit and below, then this can + // be treated as an ASHR in disguise. + if (ShiftAmt == DemandedMask.countLeadingZeros()) { + // If we only want bits that already match the signbit then we don't + // need to shift. + unsigned NumHiDemandedBits = + BitWidth - DemandedMask.countTrailingZeros(); + unsigned SignBits = + ComputeNumSignBits(I->getOperand(0), Depth + 1, CxtI); + if (SignBits >= NumHiDemandedBits) + return I->getOperand(0); + } + // Unsigned shift right. APInt DemandedMaskIn(DemandedMask.shl(ShiftAmt)); diff --git a/llvm/test/Transforms/InstCombine/modulo.ll b/llvm/test/Transforms/InstCombine/modulo.ll --- a/llvm/test/Transforms/InstCombine/modulo.ll +++ b/llvm/test/Transforms/InstCombine/modulo.ll @@ -4,9 +4,8 @@ define i32 @PR21929(i32 %x) { ; CHECK-LABEL: @PR21929( ; CHECK-NEXT: [[REM_I:%.*]] = srem i32 [[X:%.*]], 2 -; CHECK-NEXT: [[TMP1:%.*]] = lshr i32 [[REM_I]], 30 -; CHECK-NEXT: [[TMP2:%.*]] = and i32 [[TMP1]], 2 -; CHECK-NEXT: [[RET_I:%.*]] = add nsw i32 [[TMP2]], [[REM_I]] +; CHECK-NEXT: [[TMP1:%.*]] = and i32 [[REM_I]], 2 +; CHECK-NEXT: [[RET_I:%.*]] = add nsw i32 [[TMP1]], [[REM_I]] ; CHECK-NEXT: ret i32 [[RET_I]] ; %rem.i = srem i32 %x, 2 @@ -19,9 +18,8 @@ define <2 x i32> @PR21929_vec(<2 x i32> %x) { ; CHECK-LABEL: @PR21929_vec( ; CHECK-NEXT: [[REM_I:%.*]] = srem <2 x i32> [[X:%.*]], -; CHECK-NEXT: [[TMP1:%.*]] = lshr <2 x i32> [[REM_I]], -; CHECK-NEXT: [[TMP2:%.*]] = and <2 x i32> [[TMP1]], -; CHECK-NEXT: [[RET_I:%.*]] = add nsw <2 x i32> [[TMP2]], [[REM_I]] +; CHECK-NEXT: [[TMP1:%.*]] = and <2 x i32> [[REM_I]], +; CHECK-NEXT: [[RET_I:%.*]] = add nsw <2 x i32> [[TMP1]], [[REM_I]] ; CHECK-NEXT: ret <2 x i32> [[RET_I]] ; %rem.i = srem <2 x i32> %x,