[APInt] Use lshrInPlace to replace lshr where possible

[craig.topper]

Summary

This patch uses lshrInPlace to replace code where the object that lshr is called on is being overwritten with the result.

This adds an lshrInPlace(const APInt &) version as well.

This also makes lshrInPlace return a reference to *this so it can be used in some other special cases.

For example:

Elt = Elt.lshr(Index).zextOrTrunc(Length);

to

Elt = Elt.lshrInPlace(Index).zextOrTrunc(Length);

[RKSimon]

include/llvm/ADT/APInt.h

892 ↗

(On Diff #95531)

Other than operators we don't tend to use this pattern to return *this - is it a good idea to introduce it?

[hans]

Besides the return *this issue, this looks great!

include/llvm/ADT/APInt.h

892 ↗

(On Diff #95531)

I'm also not sure about this one. While I can see that it would be convenient, it doesn't seem like it would apply in that many places, and it could potentially be a source of subtle bugs, i.e. the reader might not notice the the variable gets modified in-place when this is used as part of a larger expression.

[craig.topper]

I'll remove the return reference.

[craig.topper]

Dont' return *this by reference from lshrInPlace

[hans]

lgtm

llvm/trunk/include/llvm/ADT/APInt.h

public:
   /// \brief Arithmetic right-shift function.
   ///
   /// Arithmetic right-shift this APInt by shiftAmt.
   APInt ashr(const APInt &shiftAmt) const;
 
   /// \brief Logical right-shift function.
   ///
   /// Logical right-shift this APInt by shiftAmt.
-  APInt lshr(const APInt &shiftAmt) const;
+  APInt lshr(const APInt &ShiftAmt) const {
+    APInt R(*this);
+    R.lshrInPlace(ShiftAmt);
+    return R;
+  }
+
+  /// Logical right-shift this APInt by ShiftAmt in place.
+  void lshrInPlace(const APInt &ShiftAmt);
 
   /// \brief Left-shift function.
   ///
   /// Left-shift this APInt by shiftAmt.
   APInt shl(const APInt &shiftAmt) const;
 
   /// \brief Rotate left by rotateAmt.
   APInt rotl(const APInt &rotateAmt) const;

llvm/trunk/lib/Analysis/ScalarEvolution.cpp

static const SCEV *BinomialCoefficient(const SCEV *It, unsigned K,
   // Other overflow doesn't matter because we only care about the bottom
   // W bits of the result.
   APInt OddFactorial(W, 1);
   unsigned T = 1;
   for (unsigned i = 3; i <= K; ++i) {
     APInt Mult(W, i);
     unsigned TwoFactors = Mult.countTrailingZeros();
     T += TwoFactors;
-    Mult = Mult.lshr(TwoFactors);
+    Mult.lshrInPlace(TwoFactors);
     OddFactorial *= Mult;
   }
 
   // We need at least W + T bits for the multiplication step
   unsigned CalculationBits = W + T;
 
   // Calculate 2^T, at width T+W.
   APInt DivFactor = APInt::getOneBitSet(CalculationBits, T);

llvm/trunk/lib/Analysis/ValueTracking.cpp

for (auto &AssumeVH : Q.AC->assumptionsFor(V)) {
     } else if (match(Arg, m_c_ICmp(Pred, m_Shl(m_V, m_ConstantInt(C)),
                                    m_Value(A))) &&
                Pred == ICmpInst::ICMP_EQ &&
                isValidAssumeForContext(I, Q.CxtI, Q.DT)) {
       APInt RHSKnownZero(BitWidth, 0), RHSKnownOne(BitWidth, 0);
       computeKnownBits(A, RHSKnownZero, RHSKnownOne, Depth+1, Query(Q, I));
       // For those bits in RHS that are known, we can propagate them to known
       // bits in V shifted to the right by C.
-      KnownZero |= RHSKnownZero.lshr(C->getZExtValue());
-      KnownOne  |= RHSKnownOne.lshr(C->getZExtValue());
+      RHSKnownZero.lshrInPlace(C->getZExtValue());
+      KnownZero |= RHSKnownZero;
+      RHSKnownOne.lshrInPlace(C->getZExtValue());
+      KnownOne  |= RHSKnownOne;
     // assume(~(v << c) = a)
     } else if (match(Arg, m_c_ICmp(Pred, m_Not(m_Shl(m_V, m_ConstantInt(C))),
                                    m_Value(A))) &&
                Pred == ICmpInst::ICMP_EQ &&
                isValidAssumeForContext(I, Q.CxtI, Q.DT)) {
       APInt RHSKnownZero(BitWidth, 0), RHSKnownOne(BitWidth, 0);
       computeKnownBits(A, RHSKnownZero, RHSKnownOne, Depth+1, Query(Q, I));
       // For those bits in RHS that are known, we can propagate them inverted
       // to known bits in V shifted to the right by C.
-      KnownZero |= RHSKnownOne.lshr(C->getZExtValue());
-      KnownOne  |= RHSKnownZero.lshr(C->getZExtValue());
+      RHSKnownOne.lshrInPlace(C->getZExtValue());
+      KnownZero |= RHSKnownOne;
+      RHSKnownZero.lshrInPlace(C->getZExtValue());
+      KnownOne  |= RHSKnownZero;
     // assume(v >> c = a)
     } else if (match(Arg,
                      m_c_ICmp(Pred, m_CombineOr(m_LShr(m_V, m_ConstantInt(C)),
                                                 m_AShr(m_V, m_ConstantInt(C))),
                               m_Value(A))) &&
                Pred == ICmpInst::ICMP_EQ &&
                isValidAssumeForContext(I, Q.CxtI, Q.DT)) {
       APInt RHSKnownZero(BitWidth, 0), RHSKnownOne(BitWidth, 0);
case Instruction::Shl: {
 
     computeKnownBitsFromShiftOperator(I, KnownZero, KnownOne,
                                       KnownZero2, KnownOne2, Depth, Q, KZF,
                                       KOF);
     break;
   }
   case Instruction::LShr: {
     // (ushr X, C1) & C2 == 0   iff  (-1 >> C1) & C2 == 0
-    auto KZF = [BitWidth](const APInt &KnownZero, unsigned ShiftAmt) {
-      return KnownZero.lshr(ShiftAmt) |
-             // High bits known zero.
-             APInt::getHighBitsSet(BitWidth, ShiftAmt);
+    auto KZF = [](const APInt &KnownZero, unsigned ShiftAmt) {
+      APInt KZResult = KnownZero.lshr(ShiftAmt);
+      KZResult.setHighBits(ShiftAmt);
+      return KZResult;
     };
 
     auto KOF = [](const APInt &KnownOne, unsigned ShiftAmt) {
       return KnownOne.lshr(ShiftAmt);
     };
 
     computeKnownBitsFromShiftOperator(I, KnownZero, KnownOne,
                                       KnownZero2, KnownOne2, Depth, Q, KZF,

llvm/trunk/lib/CodeGen/AsmPrinter/AsmPrinter.cpp

if (DL.isBigEndian()) {
       // [chunk1][chunk2] ... [chunkN].
       // The most significant chunk is chunkN and it should be emitted first.
       // However, due to the alignment issue chunkN contains useless bits.
       // Realign the chunks so that they contain only useless information:
       // ExtraBits     0       1       (BitWidth / 64) - 1
       //       chu[nk1 chu][nk2 chu] ... [nkN-1 chunkN]
       ExtraBits = Realigned.getRawData()[0] &
         (((uint64_t)-1) >> (64 - ExtraBitsSize));
-      Realigned = Realigned.lshr(ExtraBitsSize);
+      Realigned.lshrInPlace(ExtraBitsSize);
     } else
       ExtraBits = Realigned.getRawData()[BitWidth / 64];
   }
 
   // We don't expect assemblers to support integer data directives
   // for more than 64 bits, so we emit the data in at most 64-bit
   // quantities at a time.
   const uint64_t *RawData = Realigned.getRawData();

llvm/trunk/lib/CodeGen/CodeGenPrepare.cpp

case llvm::Instruction::And: {
       break;
     }
 
     case llvm::Instruction::Shl: {
       auto *ShlC = dyn_cast<ConstantInt>(I->getOperand(1));
       if (!ShlC)
         return false;
       uint64_t ShiftAmt = ShlC->getLimitedValue(BitWidth - 1);
-      auto ShlDemandBits = APInt::getAllOnesValue(BitWidth).lshr(ShiftAmt);
-      DemandBits |= ShlDemandBits;
+      DemandBits.setLowBits(BitWidth - ShiftAmt);
       break;
     }
 
     case llvm::Instruction::Trunc: {
       EVT TruncVT = TLI->getValueType(*DL, I->getType());
       unsigned TruncBitWidth = TruncVT.getSizeInBits();
-      auto TruncBits = APInt::getAllOnesValue(TruncBitWidth).zext(BitWidth);
-      DemandBits |= TruncBits;
+      DemandBits.setLowBits(TruncBitWidth);
       break;
     }
 
     default:
       return false;
     }
   }
 

llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

if (ConstantSDNode *N0C1 = isConstOrConstSplat(N0.getOperand(1))) {
         APInt Mask = APInt::getHighBitsSet(OpSizeInBits, OpSizeInBits - c1);
         SDValue Shift;
         if (c2 > c1) {
           Mask = Mask.shl(c2 - c1);
           SDLoc DL(N);
           Shift = DAG.getNode(ISD::SHL, DL, VT, N0.getOperand(0),
                               DAG.getConstant(c2 - c1, DL, N1.getValueType()));
         } else {
-          Mask = Mask.lshr(c1 - c2);
+          Mask.lshrInPlace(c1 - c2);
           SDLoc DL(N);
           Shift = DAG.getNode(ISD::SRL, DL, VT, N0.getOperand(0),
                               DAG.getConstant(c1 - c2, DL, N1.getValueType()));
         }
         SDLoc DL(N0);
         return DAG.getNode(ISD::AND, DL, VT, Shift,
                            DAG.getConstant(Mask, DL, VT));
       }
if (N1C && N0.getOpcode() == ISD::ANY_EXTEND) {
     if (!LegalTypes || TLI.isTypeDesirableForOp(ISD::SRL, SmallVT)) {
       uint64_t ShiftAmt = N1C->getZExtValue();
       SDLoc DL0(N0);
       SDValue SmallShift = DAG.getNode(ISD::SRL, DL0, SmallVT,
                                        N0.getOperand(0),
                           DAG.getConstant(ShiftAmt, DL0,
                                           getShiftAmountTy(SmallVT)));
       AddToWorklist(SmallShift.getNode());
-      APInt Mask = APInt::getAllOnesValue(OpSizeInBits).lshr(ShiftAmt);
+      APInt Mask = APInt::getLowBitsSet(OpSizeInBits, OpSizeInBits - ShiftAmt);
       SDLoc DL(N);
       return DAG.getNode(ISD::AND, DL, VT,
                          DAG.getNode(ISD::ANY_EXTEND, DL, VT, SmallShift),
                          DAG.getConstant(Mask, DL, VT));
     }
   }
 
   // fold (srl (sra X, Y), 31) -> (srl X, 31).  This srl only looks at the sign
for (const SDValue &Op : BV->op_values()) {
     }
 
     APInt OpVal = cast<ConstantSDNode>(Op)->
                   getAPIntValue().zextOrTrunc(SrcBitSize);
 
     for (unsigned j = 0; j != NumOutputsPerInput; ++j) {
       APInt ThisVal = OpVal.trunc(DstBitSize);
       Ops.push_back(DAG.getConstant(ThisVal, DL, DstEltVT));
-      OpVal = OpVal.lshr(DstBitSize);
+      OpVal.lshrInPlace(DstBitSize);
     }
 
     // For big endian targets, swap the order of the pieces of each element.
     if (DAG.getDataLayout().isBigEndian())
       std::reverse(Ops.end()-NumOutputsPerInput, Ops.end());
   }
 
   return DAG.getBuildVector(VT, DL, Ops);
for (int i = 0; i != NumSubElts; ++i) {
         Bits = cast<ConstantSDNode>(Elt)->getAPIntValue();
       else if (isa<ConstantFPSDNode>(Elt))
         Bits = cast<ConstantFPSDNode>(Elt)->getValueAPF().bitcastToAPInt();
       else
         return SDValue();
 
       // Extract the sub element from the constant bit mask.
       if (DAG.getDataLayout().isBigEndian()) {
-        Bits = Bits.lshr((Split - SubIdx - 1) * NumSubBits);
+        Bits.lshrInPlace((Split - SubIdx - 1) * NumSubBits);
       } else {
-        Bits = Bits.lshr(SubIdx * NumSubBits);
+        Bits.lshrInPlace(SubIdx * NumSubBits);
       }
 
       if (Split > 1)
         Bits = Bits.trunc(NumSubBits);
 
       if (Bits.isAllOnesValue())
         Indices.push_back(i);
       else if (Bits == 0)

llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAG.cpp

if (const APInt *ShAmt = getValidShiftAmountConstant(Op)) {
       // Low bits are known zero.
       KnownZero.setLowBits(ShAmt->getZExtValue());
     }
     break;
   case ISD::SRL:
     if (const APInt *ShAmt = getValidShiftAmountConstant(Op)) {
       computeKnownBits(Op.getOperand(0), KnownZero, KnownOne, DemandedElts,
                        Depth + 1);
-      KnownZero = KnownZero.lshr(*ShAmt);
-      KnownOne  = KnownOne.lshr(*ShAmt);
+      KnownZero.lshrInPlace(*ShAmt);
+      KnownOne.lshrInPlace(*ShAmt);
       // High bits are known zero.
       KnownZero.setHighBits(ShAmt->getZExtValue());
     }
     break;
   case ISD::SRA:
     if (const APInt *ShAmt = getValidShiftAmountConstant(Op)) {
       computeKnownBits(Op.getOperand(0), KnownZero, KnownOne, DemandedElts,
                        Depth + 1);
-      KnownZero = KnownZero.lshr(*ShAmt);
-      KnownOne  = KnownOne.lshr(*ShAmt);
+      KnownZero.lshrInPlace(*ShAmt);
+      KnownOne.lshrInPlace(*ShAmt);
       // If we know the value of the sign bit, then we know it is copied across
       // the high bits by the shift amount.
       APInt SignBit = APInt::getSignBit(BitWidth);
-      SignBit = SignBit.lshr(*ShAmt);  // Adjust to where it is now in the mask.
+      SignBit.lshrInPlace(*ShAmt);  // Adjust to where it is now in the mask.
       if (KnownZero.intersects(SignBit)) {
         KnownZero.setHighBits(ShAmt->getZExtValue());// New bits are known zero.
       } else if (KnownOne.intersects(SignBit)) {
         KnownOne.setHighBits(ShAmt->getZExtValue()); // New bits are known one.
       }
     }
     break;
   case ISD::SIGN_EXTEND_INREG: {

llvm/trunk/lib/CodeGen/SelectionDAG/TargetLowering.cpp

if (ConstantSDNode *SA = dyn_cast<ConstantSDNode>(Op.getOperand(1))) {
         }
       }
 
       // Compute the new bits that are at the top now.
       if (SimplifyDemandedBits(InOp, InDemandedMask,
                                KnownZero, KnownOne, TLO, Depth+1))
         return true;
       assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
-      KnownZero = KnownZero.lshr(ShAmt);
-      KnownOne  = KnownOne.lshr(ShAmt);
+      KnownZero.lshrInPlace(ShAmt);
+      KnownOne.lshrInPlace(ShAmt);
 
       KnownZero.setHighBits(ShAmt);  // High bits known zero.
     }
     break;
   case ISD::SRA:
     // If this is an arithmetic shift right and only the low-bit is set, we can
     // always convert this into a logical shr, even if the shift amount is
     // variable.  The low bit of the shift cannot be an input sign bit unless
if (ConstantSDNode *SA = dyn_cast<ConstantSDNode>(Op.getOperand(1))) {
       APInt HighBits = APInt::getHighBitsSet(BitWidth, ShAmt);
       if (HighBits.intersects(NewMask))
         InDemandedMask |= APInt::getSignBit(VT.getScalarSizeInBits());
 
       if (SimplifyDemandedBits(Op.getOperand(0), InDemandedMask,
                                KnownZero, KnownOne, TLO, Depth+1))
         return true;
       assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
-      KnownZero = KnownZero.lshr(ShAmt);
-      KnownOne  = KnownOne.lshr(ShAmt);
+      KnownZero.lshrInPlace(ShAmt);
+      KnownOne.lshrInPlace(ShAmt);
 
       // Handle the sign bit, adjusted to where it is now in the mask.
       APInt SignBit = APInt::getSignBit(BitWidth).lshr(ShAmt);
 
       // If the input sign bit is known to be zero, or if none of the top bits
       // are demanded, turn this into an unsigned shift right.
       if (KnownZero.intersects(SignBit) || (HighBits & ~NewMask) == HighBits) {
         SDNodeFlags Flags;
if (Op.getOperand(0).getNode()->hasOneUse()) {
         if (TLO.LegalTypes()) {
           uint64_t ShVal = ShAmt->getZExtValue();
           Shift = TLO.DAG.getConstant(ShVal, dl,
                                       getShiftAmountTy(Op.getValueType(), DL));
         }
 
         APInt HighBits = APInt::getHighBitsSet(OperandBitWidth,
                                                OperandBitWidth - BitWidth);
-        HighBits = HighBits.lshr(ShAmt->getZExtValue()).trunc(BitWidth);
+        HighBits.lshrInPlace(ShAmt->getZExtValue());
+        HighBits = HighBits.trunc(BitWidth);
 
         if (ShAmt->getZExtValue() < BitWidth && !(HighBits & NewMask)) {
           // None of the shifted in bits are needed.  Add a truncate of the
           // shift input, then shift it.
           SDValue NewTrunc = TLO.DAG.getNode(ISD::TRUNCATE, dl,
                                              Op.getValueType(),
                                              In.getOperand(0));
           return TLO.CombineTo(Op, TLO.DAG.getNode(ISD::SRL, dl,
if (C1.getMinSignedBits() <= 64 &&
         ISD::CondCode NewCond = Cond;
         if (AdjOne) {
           ShiftBits = C1.countTrailingOnes();
           NewC = NewC + 1;
           NewCond = (Cond == ISD::SETULE) ? ISD::SETULT : ISD::SETUGE;
         } else {
           ShiftBits = C1.countTrailingZeros();
         }
-        NewC = NewC.lshr(ShiftBits);
+        NewC.lshrInPlace(ShiftBits);
         if (ShiftBits && NewC.getMinSignedBits() <= 64 &&
           isLegalICmpImmediate(NewC.getSExtValue())) {
           auto &DL = DAG.getDataLayout();
           EVT ShiftTy = DCI.isBeforeLegalize()
                             ? getPointerTy(DL)
                             : getShiftAmountTy(N0.getValueType(), DL);
           EVT CmpTy = N0.getValueType();
           SDValue Shift = DAG.getNode(ISD::SRL, dl, CmpTy, N0,

llvm/trunk/lib/ExecutionEngine/Interpreter/Execution.cpp

if (DstNum < SrcNum) {
       // Example: bitcast <2 x i64> <i64 0, i64 1> to <4 x i32>
       unsigned Ratio = DstNum / SrcNum;
       for (unsigned i = 0; i < SrcNum; i++) {
         unsigned ShiftAmt = isLittleEndian ? 0 : DstBitSize * (Ratio - 1);
         for (unsigned j = 0; j < Ratio; j++) {
           GenericValue Elt;
           Elt.IntVal = Elt.IntVal.zext(SrcBitSize);
           Elt.IntVal = TempSrc.AggregateVal[i].IntVal;
-          Elt.IntVal = Elt.IntVal.lshr(ShiftAmt);
+          Elt.IntVal.lshrInPlace(ShiftAmt);
           // it could be DstBitSize == SrcBitSize, so check it
           if (DstBitSize < SrcBitSize)
             Elt.IntVal = Elt.IntVal.trunc(DstBitSize);
           ShiftAmt += isLittleEndian ? DstBitSize : -DstBitSize;
           TempDst.AggregateVal.push_back(Elt);
         }
       }
     }

llvm/trunk/lib/IR/ConstantFold.cpp

static Constant *ExtractConstantBytes(Constant *C, unsigned ByteStart,
   assert(ByteSize && "Must be accessing some piece");
   assert(ByteStart+ByteSize <= CSize && "Extracting invalid piece from input");
   assert(ByteSize != CSize && "Should not extract everything");
 
   // Constant Integers are simple.
   if (ConstantInt *CI = dyn_cast<ConstantInt>(C)) {
     APInt V = CI->getValue();
     if (ByteStart)
-      V = V.lshr(ByteStart*8);
+      V.lshrInPlace(ByteStart*8);
     V = V.trunc(ByteSize*8);
     return ConstantInt::get(CI->getContext(), V);
   }
 
   // In the input is a constant expr, we might be able to recursively simplify.
   // If not, we definitely can't do anything.
   ConstantExpr *CE = dyn_cast<ConstantExpr>(C);
   if (!CE) return nullptr;

llvm/trunk/lib/Support/APFloat.cpp

if (!FormatPrecision) {
 
     // FormatPrecision = 2 + floor(significandBits / lg_2(10))
     FormatPrecision = 2 + semantics->precision * 59 / 196;
   }
 
   // Ignore trailing binary zeros.
   int trailingZeros = significand.countTrailingZeros();
   exp += trailingZeros;
-  significand = significand.lshr(trailingZeros);
+  significand.lshrInPlace(trailingZeros);
 
   // Change the exponent from 2^e to 10^e.
   if (exp == 0) {
     // Nothing to do.
   } else if (exp > 0) {
     // Just shift left.
     significand = significand.zext(semantics->precision + exp);
     significand <<= exp;

llvm/trunk/lib/Support/APInt.cpp

for (unsigned i = breakWord+1; i < getNumWords(); ++i)
     val[i] = fillValue;
   APInt Result(val, BitWidth);
   Result.clearUnusedBits();
   return Result;
 }
 
 /// Logical right-shift this APInt by shiftAmt.
 /// @brief Logical right-shift function.
-APInt APInt::lshr(const APInt &shiftAmt) const {
-  return lshr((unsigned)shiftAmt.getLimitedValue(BitWidth));
+void APInt::lshrInPlace(const APInt &shiftAmt) {
+  lshrInPlace((unsigned)shiftAmt.getLimitedValue(BitWidth));
 }
 
 /// Logical right-shift this APInt by shiftAmt.
 /// @brief Logical right-shift function.
 void APInt::lshrInPlace(unsigned ShiftAmt) {
   if (isSingleWord()) {
     if (ShiftAmt >= BitWidth)
       VAL = 0;
     else
       VAL >>= ShiftAmt;
     return;
   }
 
-  return tcShiftRight(pVal, getNumWords(), ShiftAmt);
+  tcShiftRight(pVal, getNumWords(), ShiftAmt);
 }
 
 /// Left-shift this APInt by shiftAmt.
 /// @brief Left-shift function.
 APInt APInt::shl(const APInt &shiftAmt) const {
   // It's undefined behavior in C to shift by BitWidth or greater.
   return shl((unsigned)shiftAmt.getLimitedValue(BitWidth));
 }
void APInt::toString(SmallVectorImpl<char> &Str, unsigned Radix,
   if (Radix == 2 || Radix == 8 || Radix == 16) {
     // Just shift tmp right for each digit width until it becomes zero
     unsigned ShiftAmt = (Radix == 16 ? 4 : (Radix == 8 ? 3 : 1));
     unsigned MaskAmt = Radix - 1;
 
     while (Tmp != 0) {
       unsigned Digit = unsigned(Tmp.getRawData()[0]) & MaskAmt;
       Str.push_back(Digits[Digit]);
-      Tmp = Tmp.lshr(ShiftAmt);
+      Tmp.lshrInPlace(ShiftAmt);
     }
   } else {
     APInt divisor(Radix == 10? 4 : 8, Radix);
     while (Tmp != 0) {
       APInt APdigit(1, 0);
       APInt tmp2(Tmp.getBitWidth(), 0);
       divide(Tmp, Tmp.getNumWords(), divisor, divisor.getNumWords(), &tmp2,
              &APdigit);

llvm/trunk/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp

if (MSB >= Imm) {
     OpUsefulBits = OpUsefulBits.shl(Imm);
   } else {
     OpUsefulBits = OpUsefulBits.shl(MSB + 1);
     --OpUsefulBits;
     // The interesting part will be shifted in the result
     OpUsefulBits = OpUsefulBits.shl(OpUsefulBits.getBitWidth() - Imm);
     getUsefulBits(Op, OpUsefulBits, Depth + 1);
     // The interesting part was at zero in the argument
-    OpUsefulBits = OpUsefulBits.lshr(OpUsefulBits.getBitWidth() - Imm);
+    OpUsefulBits.lshrInPlace(OpUsefulBits.getBitWidth() - Imm);
   }
 
   UsefulBits &= OpUsefulBits;
 }
 
 static void getUsefulBitsFromUBFM(SDValue Op, APInt &UsefulBits,
                                   unsigned Depth) {
   uint64_t Imm =
static void getUsefulBitsFromOrWithShiftedReg(SDValue Op, APInt &UsefulBits,
   Mask.clearAllBits();
   Mask.flipAllBits();
 
   if (AArch64_AM::getShiftType(ShiftTypeAndValue) == AArch64_AM::LSL) {
     // Shift Left
     uint64_t ShiftAmt = AArch64_AM::getShiftValue(ShiftTypeAndValue);
     Mask = Mask.shl(ShiftAmt);
     getUsefulBits(Op, Mask, Depth + 1);
-    Mask = Mask.lshr(ShiftAmt);
+    Mask.lshrInPlace(ShiftAmt);
   } else if (AArch64_AM::getShiftType(ShiftTypeAndValue) == AArch64_AM::LSR) {
     // Shift Right
     // We do not handle AArch64_AM::ASR, because the sign will change the
     // number of useful bits
     uint64_t ShiftAmt = AArch64_AM::getShiftValue(ShiftTypeAndValue);
-    Mask = Mask.lshr(ShiftAmt);
+    Mask.lshrInPlace(ShiftAmt);
     getUsefulBits(Op, Mask, Depth + 1);
     Mask = Mask.shl(ShiftAmt);
   } else
     return;
 
   UsefulBits &= Mask;
 }
 
if (MSB >= Imm) {
 
     OpUsefulBits = OpUsefulBits.shl(Width);
     --OpUsefulBits;
     OpUsefulBits = OpUsefulBits.shl(LSB);
 
     if (Op.getOperand(1) == Orig) {
       // Copy the bits from the result to the zero bits.
       Mask = ResultUsefulBits & OpUsefulBits;
-      Mask = Mask.lshr(LSB);
+      Mask.lshrInPlace(LSB);
     }
 
     if (Op.getOperand(0) == Orig)
       Mask |= (ResultUsefulBits & ~OpUsefulBits);
   }
 
   UsefulBits &= Mask;
 }

llvm/trunk/lib/Target/NVPTX/NVPTXAsmPrinter.cpp

void NVPTXAsmPrinter::bufferAggregateConstant(const Constant *CPV,
   int Bytes;
 
   // Integers of arbitrary width
   if (const ConstantInt *CI = dyn_cast<ConstantInt>(CPV)) {
     APInt Val = CI->getValue();
     for (unsigned I = 0, E = DL.getTypeAllocSize(CPV->getType()); I < E; ++I) {
       uint8_t Byte = Val.getLoBits(8).getZExtValue();
       aggBuffer->addBytes(&Byte, 1, 1);
-      Val = Val.lshr(8);
+      Val.lshrInPlace(8);
     }
     return;
   }
 
   // Old constants
   if (isa<ConstantArray>(CPV) || isa<ConstantVector>(CPV)) {
     if (CPV->getNumOperands())
       for (unsigned i = 0, e = CPV->getNumOperands(); i != e; ++i)

llvm/trunk/lib/Target/X86/X86ISelLowering.cpp

for (int i = 0, Size = Mask.size(); i < Size; ++i) {
     // the (larger) source element must be UNDEF/ZERO.
     if ((Size % V.getNumOperands()) == 0) {
       int Scale = Size / V->getNumOperands();
       SDValue Op = V.getOperand(M / Scale);
       if (Op.isUndef() || X86::isZeroNode(Op))
         Zeroable.setBit(i);
       else if (ConstantSDNode *Cst = dyn_cast<ConstantSDNode>(Op)) {
         APInt Val = Cst->getAPIntValue();
-        Val = Val.lshr((M % Scale) * ScalarSizeInBits);
+        Val.lshrInPlace((M % Scale) * ScalarSizeInBits);
         Val = Val.getLoBits(ScalarSizeInBits);
         if (Val == 0)
           Zeroable.setBit(i);
       } else if (ConstantFPSDNode *Cst = dyn_cast<ConstantFPSDNode>(Op)) {
         APInt Val = Cst->getValueAPF().bitcastToAPInt();
-        Val = Val.lshr((M % Scale) * ScalarSizeInBits);
+        Val.lshrInPlace((M % Scale) * ScalarSizeInBits);
         Val = Val.getLoBits(ScalarSizeInBits);
         if (Val == 0)
           Zeroable.setBit(i);
       }
       continue;
     }
 
     // If the BUILD_VECTOR has more elements then all the (smaller) source
if (auto *ShiftImm = dyn_cast<ConstantSDNode>(Op.getOperand(1))) {
       DAG.computeKnownBits(Op.getOperand(0), KnownZero, KnownOne, Depth + 1);
       unsigned ShAmt = ShiftImm->getZExtValue();
       if (Opc == X86ISD::VSHLI) {
         KnownZero = KnownZero << ShAmt;
         KnownOne = KnownOne << ShAmt;
         // Low bits are known zero.
         KnownZero.setLowBits(ShAmt);
       } else {
-        KnownZero = KnownZero.lshr(ShAmt);
-        KnownOne = KnownOne.lshr(ShAmt);
+        KnownZero.lshrInPlace(ShAmt);
+        KnownOne.lshrInPlace(ShAmt);
         // High bits are known zero.
         KnownZero.setHighBits(ShAmt);
       }
     }
     break;
   }
   case X86ISD::VZEXT: {
     SDValue N0 = Op.getOperand(0);
assert(EltBits.size() == VT.getVectorNumElements() &&
            "Unexpected shift value type");
     unsigned ShiftImm = ShiftVal.getZExtValue();
     for (APInt &Elt : EltBits) {
       if (X86ISD::VSHLI == Opcode)
         Elt = Elt.shl(ShiftImm);
       else if (X86ISD::VSRAI == Opcode)
         Elt = Elt.ashr(ShiftImm);
       else
-        Elt = Elt.lshr(ShiftImm);
+        Elt.lshrInPlace(ShiftImm);
     }
     return getConstVector(EltBits, UndefElts, VT.getSimpleVT(), DAG, SDLoc(N));
   }
 
   return SDValue();
 }
 
 static SDValue combineVectorInsert(SDNode *N, SelectionDAG &DAG,

llvm/trunk/lib/Transforms/InstCombine/InstCombineCalls.cpp

if ((Length % 8) == 0 && (Index % 8) == 0) {
           ConstantAggregateZero::get(ShufTy), ConstantVector::get(ShuffleMask));
       return Builder.CreateBitCast(SV, II.getType());
     }
 
     // Constant Fold - shift Index'th bit to lowest position and mask off
     // Length bits.
     if (CI0) {
       APInt Elt = CI0->getValue();
-      Elt = Elt.lshr(Index).zextOrTrunc(Length);
+      Elt.lshrInPlace(Index);
+      Elt = Elt.zextOrTrunc(Length);
       return LowConstantHighUndef(Elt.getZExtValue());
     }
 
     // If we were an EXTRQ call, we'll save registers if we convert to EXTRQI.
     if (II.getIntrinsicID() == Intrinsic::x86_sse4a_extrq) {
       Value *Args[] = {Op0, CILength, CIIndex};
       Module *M = II.getModule();
       Value *F = Intrinsic::getDeclaration(M, Intrinsic::x86_sse4a_extrqi);
for (unsigned I = 0; I < NumElts; ++I) {
     }
 
     APInt Index = cast<ConstantInt>(COp)->getValue();
     Index = Index.zextOrTrunc(32).getLoBits(2);
 
     // The PD variants uses bit 1 to select per-lane element index, so
     // shift down to convert to generic shuffle mask index.
     if (IsPD)
-      Index = Index.lshr(1);
+      Index.lshrInPlace(1);
 
     // The _256 variants are a bit trickier since the mask bits always index
     // into the corresponding 128 half. In order to convert to a generic
     // shuffle, we have to make that explicit.
     Index += APInt(32, (I / NumLaneElts) * NumLaneElts);
 
     Indexes[I] = ConstantInt::get(MaskEltTy, Index);
   }

llvm/trunk/lib/Transforms/InstCombine/InstCombineShifts.cpp

if (TrOp && I.isLogicalShift() && TrOp->isShift() &&
       // The mask we constructed says what the trunc would do if occurring
       // between the shifts.  We want to know the effect *after* the second
       // shift.  We know that it is a logical shift by a constant, so adjust the
       // mask as appropriate.
       if (I.getOpcode() == Instruction::Shl)
         MaskV <<= Op1C->getZExtValue();
       else {
         assert(I.getOpcode() == Instruction::LShr && "Unknown logical shift");
-        MaskV = MaskV.lshr(Op1C->getZExtValue());
+        MaskV.lshrInPlace(Op1C->getZExtValue());
       }
 
       // shift1 & 0x00FF
       Value *And = Builder->CreateAnd(NSh,
                                       ConstantInt::get(I.getContext(), MaskV),
                                       TI->getName());
 
       // Return the value truncated to the interesting size.

llvm/trunk/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp

if (ConstantInt *SA = dyn_cast<ConstantInt>(I->getOperand(1))) {
       // they are zero).
       if (cast<LShrOperator>(I)->isExact())
         DemandedMaskIn.setLowBits(ShiftAmt);
 
       if (SimplifyDemandedBits(I, 0, DemandedMaskIn, KnownZero, KnownOne,
                                Depth + 1))
         return I;
       assert(!(KnownZero & KnownOne) && "Bits known to be one AND zero?");
-      KnownZero = KnownZero.lshr(ShiftAmt);
-      KnownOne  = KnownOne.lshr(ShiftAmt);
+      KnownZero.lshrInPlace(ShiftAmt);
+      KnownOne.lshrInPlace(ShiftAmt);
       if (ShiftAmt)
         KnownZero.setHighBits(ShiftAmt);  // high bits known zero.
     }
     break;
   case Instruction::AShr:
     // If this is an arithmetic shift right and only the low-bit is set, we can
     // always convert this into a logical shr, even if the shift amount is
     // variable.  The low bit of the shift cannot be an input sign bit unless
if (ConstantInt *SA = dyn_cast<ConstantInt>(I->getOperand(1))) {
         DemandedMaskIn.setLowBits(ShiftAmt);
 
       if (SimplifyDemandedBits(I, 0, DemandedMaskIn, KnownZero, KnownOne,
                                Depth + 1))
         return I;
       assert(!(KnownZero & KnownOne) && "Bits known to be one AND zero?");
       // Compute the new bits that are at the top now.
       APInt HighBits(APInt::getHighBitsSet(BitWidth, ShiftAmt));
-      KnownZero = KnownZero.lshr(ShiftAmt);
-      KnownOne  = KnownOne.lshr(ShiftAmt);
+      KnownZero.lshrInPlace(ShiftAmt);
+      KnownOne.lshrInPlace(ShiftAmt);
 
       // Handle the sign bits.
       APInt SignBit(APInt::getSignBit(BitWidth));
       // Adjust to where it is now in the mask.
-      SignBit = SignBit.lshr(ShiftAmt);
+      SignBit.lshrInPlace(ShiftAmt);
 
       // If the input sign bit is known to be zero, or if none of the top bits
       // are demanded, turn this into an unsigned shift right.
       if (BitWidth <= ShiftAmt || KnownZero[BitWidth-ShiftAmt-1] ||
           (HighBits & ~DemandedMask) == HighBits) {
         // Perform the logical shift right.
         BinaryOperator *NewVal = BinaryOperator::CreateLShr(I->getOperand(0),
                                                             SA, I->getName());