Home | History | Annotate | Download | only in InstCombine
      1 //===- InstCombineMulDivRem.cpp -------------------------------------------===//
      2 //
      3 //                     The LLVM Compiler Infrastructure
      4 //
      5 // This file is distributed under the University of Illinois Open Source
      6 // License. See LICENSE.TXT for details.
      7 //
      8 //===----------------------------------------------------------------------===//
      9 //
     10 // This file implements the visit functions for mul, fmul, sdiv, udiv, fdiv,
     11 // srem, urem, frem.
     12 //
     13 //===----------------------------------------------------------------------===//
     14 
     15 #include "InstCombineInternal.h"
     16 #include "llvm/Analysis/InstructionSimplify.h"
     17 #include "llvm/IR/IntrinsicInst.h"
     18 #include "llvm/IR/PatternMatch.h"
     19 using namespace llvm;
     20 using namespace PatternMatch;
     21 
     22 #define DEBUG_TYPE "instcombine"
     23 
     24 
     25 /// The specific integer value is used in a context where it is known to be
     26 /// non-zero.  If this allows us to simplify the computation, do so and return
     27 /// the new operand, otherwise return null.
     28 static Value *simplifyValueKnownNonZero(Value *V, InstCombiner &IC,
     29                                         Instruction &CxtI) {
     30   // If V has multiple uses, then we would have to do more analysis to determine
     31   // if this is safe.  For example, the use could be in dynamically unreached
     32   // code.
     33   if (!V->hasOneUse()) return nullptr;
     34 
     35   bool MadeChange = false;
     36 
     37   // ((1 << A) >>u B) --> (1 << (A-B))
     38   // Because V cannot be zero, we know that B is less than A.
     39   Value *A = nullptr, *B = nullptr, *One = nullptr;
     40   if (match(V, m_LShr(m_OneUse(m_Shl(m_Value(One), m_Value(A))), m_Value(B))) &&
     41       match(One, m_One())) {
     42     A = IC.Builder->CreateSub(A, B);
     43     return IC.Builder->CreateShl(One, A);
     44   }
     45 
     46   // (PowerOfTwo >>u B) --> isExact since shifting out the result would make it
     47   // inexact.  Similarly for <<.
     48   BinaryOperator *I = dyn_cast<BinaryOperator>(V);
     49   if (I && I->isLogicalShift() &&
     50       isKnownToBeAPowerOfTwo(I->getOperand(0), IC.getDataLayout(), false, 0,
     51                              IC.getAssumptionCache(), &CxtI,
     52                              IC.getDominatorTree())) {
     53     // We know that this is an exact/nuw shift and that the input is a
     54     // non-zero context as well.
     55     if (Value *V2 = simplifyValueKnownNonZero(I->getOperand(0), IC, CxtI)) {
     56       I->setOperand(0, V2);
     57       MadeChange = true;
     58     }
     59 
     60     if (I->getOpcode() == Instruction::LShr && !I->isExact()) {
     61       I->setIsExact();
     62       MadeChange = true;
     63     }
     64 
     65     if (I->getOpcode() == Instruction::Shl && !I->hasNoUnsignedWrap()) {
     66       I->setHasNoUnsignedWrap();
     67       MadeChange = true;
     68     }
     69   }
     70 
     71   // TODO: Lots more we could do here:
     72   //    If V is a phi node, we can call this on each of its operands.
     73   //    "select cond, X, 0" can simplify to "X".
     74 
     75   return MadeChange ? V : nullptr;
     76 }
     77 
     78 
     79 /// True if the multiply can not be expressed in an int this size.
     80 static bool MultiplyOverflows(const APInt &C1, const APInt &C2, APInt &Product,
     81                               bool IsSigned) {
     82   bool Overflow;
     83   if (IsSigned)
     84     Product = C1.smul_ov(C2, Overflow);
     85   else
     86     Product = C1.umul_ov(C2, Overflow);
     87 
     88   return Overflow;
     89 }
     90 
     91 /// \brief True if C2 is a multiple of C1. Quotient contains C2/C1.
     92 static bool IsMultiple(const APInt &C1, const APInt &C2, APInt &Quotient,
     93                        bool IsSigned) {
     94   assert(C1.getBitWidth() == C2.getBitWidth() &&
     95          "Inconsistent width of constants!");
     96 
     97   // Bail if we will divide by zero.
     98   if (C2.isMinValue())
     99     return false;
    100 
    101   // Bail if we would divide INT_MIN by -1.
    102   if (IsSigned && C1.isMinSignedValue() && C2.isAllOnesValue())
    103     return false;
    104 
    105   APInt Remainder(C1.getBitWidth(), /*Val=*/0ULL, IsSigned);
    106   if (IsSigned)
    107     APInt::sdivrem(C1, C2, Quotient, Remainder);
    108   else
    109     APInt::udivrem(C1, C2, Quotient, Remainder);
    110 
    111   return Remainder.isMinValue();
    112 }
    113 
    114 /// \brief A helper routine of InstCombiner::visitMul().
    115 ///
    116 /// If C is a vector of known powers of 2, then this function returns
    117 /// a new vector obtained from C replacing each element with its logBase2.
    118 /// Return a null pointer otherwise.
    119 static Constant *getLogBase2Vector(ConstantDataVector *CV) {
    120   const APInt *IVal;
    121   SmallVector<Constant *, 4> Elts;
    122 
    123   for (unsigned I = 0, E = CV->getNumElements(); I != E; ++I) {
    124     Constant *Elt = CV->getElementAsConstant(I);
    125     if (!match(Elt, m_APInt(IVal)) || !IVal->isPowerOf2())
    126       return nullptr;
    127     Elts.push_back(ConstantInt::get(Elt->getType(), IVal->logBase2()));
    128   }
    129 
    130   return ConstantVector::get(Elts);
    131 }
    132 
    133 /// \brief Return true if we can prove that:
    134 ///    (mul LHS, RHS)  === (mul nsw LHS, RHS)
    135 bool InstCombiner::WillNotOverflowSignedMul(Value *LHS, Value *RHS,
    136                                             Instruction &CxtI) {
    137   // Multiplying n * m significant bits yields a result of n + m significant
    138   // bits. If the total number of significant bits does not exceed the
    139   // result bit width (minus 1), there is no overflow.
    140   // This means if we have enough leading sign bits in the operands
    141   // we can guarantee that the result does not overflow.
    142   // Ref: "Hacker's Delight" by Henry Warren
    143   unsigned BitWidth = LHS->getType()->getScalarSizeInBits();
    144 
    145   // Note that underestimating the number of sign bits gives a more
    146   // conservative answer.
    147   unsigned SignBits =
    148       ComputeNumSignBits(LHS, 0, &CxtI) + ComputeNumSignBits(RHS, 0, &CxtI);
    149 
    150   // First handle the easy case: if we have enough sign bits there's
    151   // definitely no overflow.
    152   if (SignBits > BitWidth + 1)
    153     return true;
    154 
    155   // There are two ambiguous cases where there can be no overflow:
    156   //   SignBits == BitWidth + 1    and
    157   //   SignBits == BitWidth
    158   // The second case is difficult to check, therefore we only handle the
    159   // first case.
    160   if (SignBits == BitWidth + 1) {
    161     // It overflows only when both arguments are negative and the true
    162     // product is exactly the minimum negative number.
    163     // E.g. mul i16 with 17 sign bits: 0xff00 * 0xff80 = 0x8000
    164     // For simplicity we just check if at least one side is not negative.
    165     bool LHSNonNegative, LHSNegative;
    166     bool RHSNonNegative, RHSNegative;
    167     ComputeSignBit(LHS, LHSNonNegative, LHSNegative, /*Depth=*/0, &CxtI);
    168     ComputeSignBit(RHS, RHSNonNegative, RHSNegative, /*Depth=*/0, &CxtI);
    169     if (LHSNonNegative || RHSNonNegative)
    170       return true;
    171   }
    172   return false;
    173 }
    174 
    175 Instruction *InstCombiner::visitMul(BinaryOperator &I) {
    176   bool Changed = SimplifyAssociativeOrCommutative(I);
    177   Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);
    178 
    179   if (Value *V = SimplifyVectorOp(I))
    180     return replaceInstUsesWith(I, V);
    181 
    182   if (Value *V = SimplifyMulInst(Op0, Op1, DL, TLI, DT, AC))
    183     return replaceInstUsesWith(I, V);
    184 
    185   if (Value *V = SimplifyUsingDistributiveLaws(I))
    186     return replaceInstUsesWith(I, V);
    187 
    188   // X * -1 == 0 - X
    189   if (match(Op1, m_AllOnes())) {
    190     BinaryOperator *BO = BinaryOperator::CreateNeg(Op0, I.getName());
    191     if (I.hasNoSignedWrap())
    192       BO->setHasNoSignedWrap();
    193     return BO;
    194   }
    195 
    196   // Also allow combining multiply instructions on vectors.
    197   {
    198     Value *NewOp;
    199     Constant *C1, *C2;
    200     const APInt *IVal;
    201     if (match(&I, m_Mul(m_Shl(m_Value(NewOp), m_Constant(C2)),
    202                         m_Constant(C1))) &&
    203         match(C1, m_APInt(IVal))) {
    204       // ((X << C2)*C1) == (X * (C1 << C2))
    205       Constant *Shl = ConstantExpr::getShl(C1, C2);
    206       BinaryOperator *Mul = cast<BinaryOperator>(I.getOperand(0));
    207       BinaryOperator *BO = BinaryOperator::CreateMul(NewOp, Shl);
    208       if (I.hasNoUnsignedWrap() && Mul->hasNoUnsignedWrap())
    209         BO->setHasNoUnsignedWrap();
    210       if (I.hasNoSignedWrap() && Mul->hasNoSignedWrap() &&
    211           Shl->isNotMinSignedValue())
    212         BO->setHasNoSignedWrap();
    213       return BO;
    214     }
    215 
    216     if (match(&I, m_Mul(m_Value(NewOp), m_Constant(C1)))) {
    217       Constant *NewCst = nullptr;
    218       if (match(C1, m_APInt(IVal)) && IVal->isPowerOf2())
    219         // Replace X*(2^C) with X << C, where C is either a scalar or a splat.
    220         NewCst = ConstantInt::get(NewOp->getType(), IVal->logBase2());
    221       else if (ConstantDataVector *CV = dyn_cast<ConstantDataVector>(C1))
    222         // Replace X*(2^C) with X << C, where C is a vector of known
    223         // constant powers of 2.
    224         NewCst = getLogBase2Vector(CV);
    225 
    226       if (NewCst) {
    227         unsigned Width = NewCst->getType()->getPrimitiveSizeInBits();
    228         BinaryOperator *Shl = BinaryOperator::CreateShl(NewOp, NewCst);
    229 
    230         if (I.hasNoUnsignedWrap())
    231           Shl->setHasNoUnsignedWrap();
    232         if (I.hasNoSignedWrap()) {
    233           uint64_t V;
    234           if (match(NewCst, m_ConstantInt(V)) && V != Width - 1)
    235             Shl->setHasNoSignedWrap();
    236         }
    237 
    238         return Shl;
    239       }
    240     }
    241   }
    242 
    243   if (ConstantInt *CI = dyn_cast<ConstantInt>(Op1)) {
    244     // (Y - X) * (-(2**n)) -> (X - Y) * (2**n), for positive nonzero n
    245     // (Y + const) * (-(2**n)) -> (-constY) * (2**n), for positive nonzero n
    246     // The "* (2**n)" thus becomes a potential shifting opportunity.
    247     {
    248       const APInt &   Val = CI->getValue();
    249       const APInt &PosVal = Val.abs();
    250       if (Val.isNegative() && PosVal.isPowerOf2()) {
    251         Value *X = nullptr, *Y = nullptr;
    252         if (Op0->hasOneUse()) {
    253           ConstantInt *C1;
    254           Value *Sub = nullptr;
    255           if (match(Op0, m_Sub(m_Value(Y), m_Value(X))))
    256             Sub = Builder->CreateSub(X, Y, "suba");
    257           else if (match(Op0, m_Add(m_Value(Y), m_ConstantInt(C1))))
    258             Sub = Builder->CreateSub(Builder->CreateNeg(C1), Y, "subc");
    259           if (Sub)
    260             return
    261               BinaryOperator::CreateMul(Sub,
    262                                         ConstantInt::get(Y->getType(), PosVal));
    263         }
    264       }
    265     }
    266   }
    267 
    268   // Simplify mul instructions with a constant RHS.
    269   if (isa<Constant>(Op1)) {
    270     // Try to fold constant mul into select arguments.
    271     if (SelectInst *SI = dyn_cast<SelectInst>(Op0))
    272       if (Instruction *R = FoldOpIntoSelect(I, SI))
    273         return R;
    274 
    275     if (isa<PHINode>(Op0))
    276       if (Instruction *NV = FoldOpIntoPhi(I))
    277         return NV;
    278 
    279     // Canonicalize (X+C1)*CI -> X*CI+C1*CI.
    280     {
    281       Value *X;
    282       Constant *C1;
    283       if (match(Op0, m_OneUse(m_Add(m_Value(X), m_Constant(C1))))) {
    284         Value *Mul = Builder->CreateMul(C1, Op1);
    285         // Only go forward with the transform if C1*CI simplifies to a tidier
    286         // constant.
    287         if (!match(Mul, m_Mul(m_Value(), m_Value())))
    288           return BinaryOperator::CreateAdd(Builder->CreateMul(X, Op1), Mul);
    289       }
    290     }
    291   }
    292 
    293   if (Value *Op0v = dyn_castNegVal(Op0)) {   // -X * -Y = X*Y
    294     if (Value *Op1v = dyn_castNegVal(Op1)) {
    295       BinaryOperator *BO = BinaryOperator::CreateMul(Op0v, Op1v);
    296       if (I.hasNoSignedWrap() &&
    297           match(Op0, m_NSWSub(m_Value(), m_Value())) &&
    298           match(Op1, m_NSWSub(m_Value(), m_Value())))
    299         BO->setHasNoSignedWrap();
    300       return BO;
    301     }
    302   }
    303 
    304   // (X / Y) *  Y = X - (X % Y)
    305   // (X / Y) * -Y = (X % Y) - X
    306   {
    307     Value *Op1C = Op1;
    308     BinaryOperator *BO = dyn_cast<BinaryOperator>(Op0);
    309     if (!BO ||
    310         (BO->getOpcode() != Instruction::UDiv &&
    311          BO->getOpcode() != Instruction::SDiv)) {
    312       Op1C = Op0;
    313       BO = dyn_cast<BinaryOperator>(Op1);
    314     }
    315     Value *Neg = dyn_castNegVal(Op1C);
    316     if (BO && BO->hasOneUse() &&
    317         (BO->getOperand(1) == Op1C || BO->getOperand(1) == Neg) &&
    318         (BO->getOpcode() == Instruction::UDiv ||
    319          BO->getOpcode() == Instruction::SDiv)) {
    320       Value *Op0BO = BO->getOperand(0), *Op1BO = BO->getOperand(1);
    321 
    322       // If the division is exact, X % Y is zero, so we end up with X or -X.
    323       if (PossiblyExactOperator *SDiv = dyn_cast<PossiblyExactOperator>(BO))
    324         if (SDiv->isExact()) {
    325           if (Op1BO == Op1C)
    326             return replaceInstUsesWith(I, Op0BO);
    327           return BinaryOperator::CreateNeg(Op0BO);
    328         }
    329 
    330       Value *Rem;
    331       if (BO->getOpcode() == Instruction::UDiv)
    332         Rem = Builder->CreateURem(Op0BO, Op1BO);
    333       else
    334         Rem = Builder->CreateSRem(Op0BO, Op1BO);
    335       Rem->takeName(BO);
    336 
    337       if (Op1BO == Op1C)
    338         return BinaryOperator::CreateSub(Op0BO, Rem);
    339       return BinaryOperator::CreateSub(Rem, Op0BO);
    340     }
    341   }
    342 
    343   /// i1 mul -> i1 and.
    344   if (I.getType()->getScalarType()->isIntegerTy(1))
    345     return BinaryOperator::CreateAnd(Op0, Op1);
    346 
    347   // X*(1 << Y) --> X << Y
    348   // (1 << Y)*X --> X << Y
    349   {
    350     Value *Y;
    351     BinaryOperator *BO = nullptr;
    352     bool ShlNSW = false;
    353     if (match(Op0, m_Shl(m_One(), m_Value(Y)))) {
    354       BO = BinaryOperator::CreateShl(Op1, Y);
    355       ShlNSW = cast<ShlOperator>(Op0)->hasNoSignedWrap();
    356     } else if (match(Op1, m_Shl(m_One(), m_Value(Y)))) {
    357       BO = BinaryOperator::CreateShl(Op0, Y);
    358       ShlNSW = cast<ShlOperator>(Op1)->hasNoSignedWrap();
    359     }
    360     if (BO) {
    361       if (I.hasNoUnsignedWrap())
    362         BO->setHasNoUnsignedWrap();
    363       if (I.hasNoSignedWrap() && ShlNSW)
    364         BO->setHasNoSignedWrap();
    365       return BO;
    366     }
    367   }
    368 
    369   // If one of the operands of the multiply is a cast from a boolean value, then
    370   // we know the bool is either zero or one, so this is a 'masking' multiply.
    371   //   X * Y (where Y is 0 or 1) -> X & (0-Y)
    372   if (!I.getType()->isVectorTy()) {
    373     // -2 is "-1 << 1" so it is all bits set except the low one.
    374     APInt Negative2(I.getType()->getPrimitiveSizeInBits(), (uint64_t)-2, true);
    375 
    376     Value *BoolCast = nullptr, *OtherOp = nullptr;
    377     if (MaskedValueIsZero(Op0, Negative2, 0, &I)) {
    378       BoolCast = Op0;
    379       OtherOp = Op1;
    380     } else if (MaskedValueIsZero(Op1, Negative2, 0, &I)) {
    381       BoolCast = Op1;
    382       OtherOp = Op0;
    383     }
    384 
    385     if (BoolCast) {
    386       Value *V = Builder->CreateSub(Constant::getNullValue(I.getType()),
    387                                     BoolCast);
    388       return BinaryOperator::CreateAnd(V, OtherOp);
    389     }
    390   }
    391 
    392   if (!I.hasNoSignedWrap() && WillNotOverflowSignedMul(Op0, Op1, I)) {
    393     Changed = true;
    394     I.setHasNoSignedWrap(true);
    395   }
    396 
    397   if (!I.hasNoUnsignedWrap() &&
    398       computeOverflowForUnsignedMul(Op0, Op1, &I) ==
    399           OverflowResult::NeverOverflows) {
    400     Changed = true;
    401     I.setHasNoUnsignedWrap(true);
    402   }
    403 
    404   return Changed ? &I : nullptr;
    405 }
    406 
    407 /// Detect pattern log2(Y * 0.5) with corresponding fast math flags.
    408 static void detectLog2OfHalf(Value *&Op, Value *&Y, IntrinsicInst *&Log2) {
    409   if (!Op->hasOneUse())
    410     return;
    411 
    412   IntrinsicInst *II = dyn_cast<IntrinsicInst>(Op);
    413   if (!II)
    414     return;
    415   if (II->getIntrinsicID() != Intrinsic::log2 || !II->hasUnsafeAlgebra())
    416     return;
    417   Log2 = II;
    418 
    419   Value *OpLog2Of = II->getArgOperand(0);
    420   if (!OpLog2Of->hasOneUse())
    421     return;
    422 
    423   Instruction *I = dyn_cast<Instruction>(OpLog2Of);
    424   if (!I)
    425     return;
    426   if (I->getOpcode() != Instruction::FMul || !I->hasUnsafeAlgebra())
    427     return;
    428 
    429   if (match(I->getOperand(0), m_SpecificFP(0.5)))
    430     Y = I->getOperand(1);
    431   else if (match(I->getOperand(1), m_SpecificFP(0.5)))
    432     Y = I->getOperand(0);
    433 }
    434 
    435 static bool isFiniteNonZeroFp(Constant *C) {
    436   if (C->getType()->isVectorTy()) {
    437     for (unsigned I = 0, E = C->getType()->getVectorNumElements(); I != E;
    438          ++I) {
    439       ConstantFP *CFP = dyn_cast_or_null<ConstantFP>(C->getAggregateElement(I));
    440       if (!CFP || !CFP->getValueAPF().isFiniteNonZero())
    441         return false;
    442     }
    443     return true;
    444   }
    445 
    446   return isa<ConstantFP>(C) &&
    447          cast<ConstantFP>(C)->getValueAPF().isFiniteNonZero();
    448 }
    449 
    450 static bool isNormalFp(Constant *C) {
    451   if (C->getType()->isVectorTy()) {
    452     for (unsigned I = 0, E = C->getType()->getVectorNumElements(); I != E;
    453          ++I) {
    454       ConstantFP *CFP = dyn_cast_or_null<ConstantFP>(C->getAggregateElement(I));
    455       if (!CFP || !CFP->getValueAPF().isNormal())
    456         return false;
    457     }
    458     return true;
    459   }
    460 
    461   return isa<ConstantFP>(C) && cast<ConstantFP>(C)->getValueAPF().isNormal();
    462 }
    463 
    464 /// Helper function of InstCombiner::visitFMul(BinaryOperator(). It returns
    465 /// true iff the given value is FMul or FDiv with one and only one operand
    466 /// being a normal constant (i.e. not Zero/NaN/Infinity).
    467 static bool isFMulOrFDivWithConstant(Value *V) {
    468   Instruction *I = dyn_cast<Instruction>(V);
    469   if (!I || (I->getOpcode() != Instruction::FMul &&
    470              I->getOpcode() != Instruction::FDiv))
    471     return false;
    472 
    473   Constant *C0 = dyn_cast<Constant>(I->getOperand(0));
    474   Constant *C1 = dyn_cast<Constant>(I->getOperand(1));
    475 
    476   if (C0 && C1)
    477     return false;
    478 
    479   return (C0 && isFiniteNonZeroFp(C0)) || (C1 && isFiniteNonZeroFp(C1));
    480 }
    481 
    482 /// foldFMulConst() is a helper routine of InstCombiner::visitFMul().
    483 /// The input \p FMulOrDiv is a FMul/FDiv with one and only one operand
    484 /// being a constant (i.e. isFMulOrFDivWithConstant(FMulOrDiv) == true).
    485 /// This function is to simplify "FMulOrDiv * C" and returns the
    486 /// resulting expression. Note that this function could return NULL in
    487 /// case the constants cannot be folded into a normal floating-point.
    488 ///
    489 Value *InstCombiner::foldFMulConst(Instruction *FMulOrDiv, Constant *C,
    490                                    Instruction *InsertBefore) {
    491   assert(isFMulOrFDivWithConstant(FMulOrDiv) && "V is invalid");
    492 
    493   Value *Opnd0 = FMulOrDiv->getOperand(0);
    494   Value *Opnd1 = FMulOrDiv->getOperand(1);
    495 
    496   Constant *C0 = dyn_cast<Constant>(Opnd0);
    497   Constant *C1 = dyn_cast<Constant>(Opnd1);
    498 
    499   BinaryOperator *R = nullptr;
    500 
    501   // (X * C0) * C => X * (C0*C)
    502   if (FMulOrDiv->getOpcode() == Instruction::FMul) {
    503     Constant *F = ConstantExpr::getFMul(C1 ? C1 : C0, C);
    504     if (isNormalFp(F))
    505       R = BinaryOperator::CreateFMul(C1 ? Opnd0 : Opnd1, F);
    506   } else {
    507     if (C0) {
    508       // (C0 / X) * C => (C0 * C) / X
    509       if (FMulOrDiv->hasOneUse()) {
    510         // It would otherwise introduce another div.
    511         Constant *F = ConstantExpr::getFMul(C0, C);
    512         if (isNormalFp(F))
    513           R = BinaryOperator::CreateFDiv(F, Opnd1);
    514       }
    515     } else {
    516       // (X / C1) * C => X * (C/C1) if C/C1 is not a denormal
    517       Constant *F = ConstantExpr::getFDiv(C, C1);
    518       if (isNormalFp(F)) {
    519         R = BinaryOperator::CreateFMul(Opnd0, F);
    520       } else {
    521         // (X / C1) * C => X / (C1/C)
    522         Constant *F = ConstantExpr::getFDiv(C1, C);
    523         if (isNormalFp(F))
    524           R = BinaryOperator::CreateFDiv(Opnd0, F);
    525       }
    526     }
    527   }
    528 
    529   if (R) {
    530     R->setHasUnsafeAlgebra(true);
    531     InsertNewInstWith(R, *InsertBefore);
    532   }
    533 
    534   return R;
    535 }
    536 
    537 Instruction *InstCombiner::visitFMul(BinaryOperator &I) {
    538   bool Changed = SimplifyAssociativeOrCommutative(I);
    539   Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);
    540 
    541   if (Value *V = SimplifyVectorOp(I))
    542     return replaceInstUsesWith(I, V);
    543 
    544   if (isa<Constant>(Op0))
    545     std::swap(Op0, Op1);
    546 
    547   if (Value *V =
    548           SimplifyFMulInst(Op0, Op1, I.getFastMathFlags(), DL, TLI, DT, AC))
    549     return replaceInstUsesWith(I, V);
    550 
    551   bool AllowReassociate = I.hasUnsafeAlgebra();
    552 
    553   // Simplify mul instructions with a constant RHS.
    554   if (isa<Constant>(Op1)) {
    555     // Try to fold constant mul into select arguments.
    556     if (SelectInst *SI = dyn_cast<SelectInst>(Op0))
    557       if (Instruction *R = FoldOpIntoSelect(I, SI))
    558         return R;
    559 
    560     if (isa<PHINode>(Op0))
    561       if (Instruction *NV = FoldOpIntoPhi(I))
    562         return NV;
    563 
    564     // (fmul X, -1.0) --> (fsub -0.0, X)
    565     if (match(Op1, m_SpecificFP(-1.0))) {
    566       Constant *NegZero = ConstantFP::getNegativeZero(Op1->getType());
    567       Instruction *RI = BinaryOperator::CreateFSub(NegZero, Op0);
    568       RI->copyFastMathFlags(&I);
    569       return RI;
    570     }
    571 
    572     Constant *C = cast<Constant>(Op1);
    573     if (AllowReassociate && isFiniteNonZeroFp(C)) {
    574       // Let MDC denote an expression in one of these forms:
    575       // X * C, C/X, X/C, where C is a constant.
    576       //
    577       // Try to simplify "MDC * Constant"
    578       if (isFMulOrFDivWithConstant(Op0))
    579         if (Value *V = foldFMulConst(cast<Instruction>(Op0), C, &I))
    580           return replaceInstUsesWith(I, V);
    581 
    582       // (MDC +/- C1) * C => (MDC * C) +/- (C1 * C)
    583       Instruction *FAddSub = dyn_cast<Instruction>(Op0);
    584       if (FAddSub &&
    585           (FAddSub->getOpcode() == Instruction::FAdd ||
    586            FAddSub->getOpcode() == Instruction::FSub)) {
    587         Value *Opnd0 = FAddSub->getOperand(0);
    588         Value *Opnd1 = FAddSub->getOperand(1);
    589         Constant *C0 = dyn_cast<Constant>(Opnd0);
    590         Constant *C1 = dyn_cast<Constant>(Opnd1);
    591         bool Swap = false;
    592         if (C0) {
    593           std::swap(C0, C1);
    594           std::swap(Opnd0, Opnd1);
    595           Swap = true;
    596         }
    597 
    598         if (C1 && isFiniteNonZeroFp(C1) && isFMulOrFDivWithConstant(Opnd0)) {
    599           Value *M1 = ConstantExpr::getFMul(C1, C);
    600           Value *M0 = isNormalFp(cast<Constant>(M1)) ?
    601                       foldFMulConst(cast<Instruction>(Opnd0), C, &I) :
    602                       nullptr;
    603           if (M0 && M1) {
    604             if (Swap && FAddSub->getOpcode() == Instruction::FSub)
    605               std::swap(M0, M1);
    606 
    607             Instruction *RI = (FAddSub->getOpcode() == Instruction::FAdd)
    608                                   ? BinaryOperator::CreateFAdd(M0, M1)
    609                                   : BinaryOperator::CreateFSub(M0, M1);
    610             RI->copyFastMathFlags(&I);
    611             return RI;
    612           }
    613         }
    614       }
    615     }
    616   }
    617 
    618   if (Op0 == Op1) {
    619     if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(Op0)) {
    620       // sqrt(X) * sqrt(X) -> X
    621       if (AllowReassociate && II->getIntrinsicID() == Intrinsic::sqrt)
    622         return replaceInstUsesWith(I, II->getOperand(0));
    623 
    624       // fabs(X) * fabs(X) -> X * X
    625       if (II->getIntrinsicID() == Intrinsic::fabs) {
    626         Instruction *FMulVal = BinaryOperator::CreateFMul(II->getOperand(0),
    627                                                           II->getOperand(0),
    628                                                           I.getName());
    629         FMulVal->copyFastMathFlags(&I);
    630         return FMulVal;
    631       }
    632     }
    633   }
    634 
    635   // Under unsafe algebra do:
    636   // X * log2(0.5*Y) = X*log2(Y) - X
    637   if (AllowReassociate) {
    638     Value *OpX = nullptr;
    639     Value *OpY = nullptr;
    640     IntrinsicInst *Log2;
    641     detectLog2OfHalf(Op0, OpY, Log2);
    642     if (OpY) {
    643       OpX = Op1;
    644     } else {
    645       detectLog2OfHalf(Op1, OpY, Log2);
    646       if (OpY) {
    647         OpX = Op0;
    648       }
    649     }
    650     // if pattern detected emit alternate sequence
    651     if (OpX && OpY) {
    652       BuilderTy::FastMathFlagGuard Guard(*Builder);
    653       Builder->setFastMathFlags(Log2->getFastMathFlags());
    654       Log2->setArgOperand(0, OpY);
    655       Value *FMulVal = Builder->CreateFMul(OpX, Log2);
    656       Value *FSub = Builder->CreateFSub(FMulVal, OpX);
    657       FSub->takeName(&I);
    658       return replaceInstUsesWith(I, FSub);
    659     }
    660   }
    661 
    662   // Handle symmetric situation in a 2-iteration loop
    663   Value *Opnd0 = Op0;
    664   Value *Opnd1 = Op1;
    665   for (int i = 0; i < 2; i++) {
    666     bool IgnoreZeroSign = I.hasNoSignedZeros();
    667     if (BinaryOperator::isFNeg(Opnd0, IgnoreZeroSign)) {
    668       BuilderTy::FastMathFlagGuard Guard(*Builder);
    669       Builder->setFastMathFlags(I.getFastMathFlags());
    670 
    671       Value *N0 = dyn_castFNegVal(Opnd0, IgnoreZeroSign);
    672       Value *N1 = dyn_castFNegVal(Opnd1, IgnoreZeroSign);
    673 
    674       // -X * -Y => X*Y
    675       if (N1) {
    676         Value *FMul = Builder->CreateFMul(N0, N1);
    677         FMul->takeName(&I);
    678         return replaceInstUsesWith(I, FMul);
    679       }
    680 
    681       if (Opnd0->hasOneUse()) {
    682         // -X * Y => -(X*Y) (Promote negation as high as possible)
    683         Value *T = Builder->CreateFMul(N0, Opnd1);
    684         Value *Neg = Builder->CreateFNeg(T);
    685         Neg->takeName(&I);
    686         return replaceInstUsesWith(I, Neg);
    687       }
    688     }
    689 
    690     // (X*Y) * X => (X*X) * Y where Y != X
    691     //  The purpose is two-fold:
    692     //   1) to form a power expression (of X).
    693     //   2) potentially shorten the critical path: After transformation, the
    694     //  latency of the instruction Y is amortized by the expression of X*X,
    695     //  and therefore Y is in a "less critical" position compared to what it
    696     //  was before the transformation.
    697     //
    698     if (AllowReassociate) {
    699       Value *Opnd0_0, *Opnd0_1;
    700       if (Opnd0->hasOneUse() &&
    701           match(Opnd0, m_FMul(m_Value(Opnd0_0), m_Value(Opnd0_1)))) {
    702         Value *Y = nullptr;
    703         if (Opnd0_0 == Opnd1 && Opnd0_1 != Opnd1)
    704           Y = Opnd0_1;
    705         else if (Opnd0_1 == Opnd1 && Opnd0_0 != Opnd1)
    706           Y = Opnd0_0;
    707 
    708         if (Y) {
    709           BuilderTy::FastMathFlagGuard Guard(*Builder);
    710           Builder->setFastMathFlags(I.getFastMathFlags());
    711           Value *T = Builder->CreateFMul(Opnd1, Opnd1);
    712 
    713           Value *R = Builder->CreateFMul(T, Y);
    714           R->takeName(&I);
    715           return replaceInstUsesWith(I, R);
    716         }
    717       }
    718     }
    719 
    720     if (!isa<Constant>(Op1))
    721       std::swap(Opnd0, Opnd1);
    722     else
    723       break;
    724   }
    725 
    726   return Changed ? &I : nullptr;
    727 }
    728 
    729 /// Try to fold a divide or remainder of a select instruction.
    730 bool InstCombiner::SimplifyDivRemOfSelect(BinaryOperator &I) {
    731   SelectInst *SI = cast<SelectInst>(I.getOperand(1));
    732 
    733   // div/rem X, (Cond ? 0 : Y) -> div/rem X, Y
    734   int NonNullOperand = -1;
    735   if (Constant *ST = dyn_cast<Constant>(SI->getOperand(1)))
    736     if (ST->isNullValue())
    737       NonNullOperand = 2;
    738   // div/rem X, (Cond ? Y : 0) -> div/rem X, Y
    739   if (Constant *ST = dyn_cast<Constant>(SI->getOperand(2)))
    740     if (ST->isNullValue())
    741       NonNullOperand = 1;
    742 
    743   if (NonNullOperand == -1)
    744     return false;
    745 
    746   Value *SelectCond = SI->getOperand(0);
    747 
    748   // Change the div/rem to use 'Y' instead of the select.
    749   I.setOperand(1, SI->getOperand(NonNullOperand));
    750 
    751   // Okay, we know we replace the operand of the div/rem with 'Y' with no
    752   // problem.  However, the select, or the condition of the select may have
    753   // multiple uses.  Based on our knowledge that the operand must be non-zero,
    754   // propagate the known value for the select into other uses of it, and
    755   // propagate a known value of the condition into its other users.
    756 
    757   // If the select and condition only have a single use, don't bother with this,
    758   // early exit.
    759   if (SI->use_empty() && SelectCond->hasOneUse())
    760     return true;
    761 
    762   // Scan the current block backward, looking for other uses of SI.
    763   BasicBlock::iterator BBI = I.getIterator(), BBFront = I.getParent()->begin();
    764 
    765   while (BBI != BBFront) {
    766     --BBI;
    767     // If we found a call to a function, we can't assume it will return, so
    768     // information from below it cannot be propagated above it.
    769     if (isa<CallInst>(BBI) && !isa<IntrinsicInst>(BBI))
    770       break;
    771 
    772     // Replace uses of the select or its condition with the known values.
    773     for (Instruction::op_iterator I = BBI->op_begin(), E = BBI->op_end();
    774          I != E; ++I) {
    775       if (*I == SI) {
    776         *I = SI->getOperand(NonNullOperand);
    777         Worklist.Add(&*BBI);
    778       } else if (*I == SelectCond) {
    779         *I = Builder->getInt1(NonNullOperand == 1);
    780         Worklist.Add(&*BBI);
    781       }
    782     }
    783 
    784     // If we past the instruction, quit looking for it.
    785     if (&*BBI == SI)
    786       SI = nullptr;
    787     if (&*BBI == SelectCond)
    788       SelectCond = nullptr;
    789 
    790     // If we ran out of things to eliminate, break out of the loop.
    791     if (!SelectCond && !SI)
    792       break;
    793 
    794   }
    795   return true;
    796 }
    797 
    798 
    799 /// This function implements the transforms common to both integer division
    800 /// instructions (udiv and sdiv). It is called by the visitors to those integer
    801 /// division instructions.
    802 /// @brief Common integer divide transforms
    803 Instruction *InstCombiner::commonIDivTransforms(BinaryOperator &I) {
    804   Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);
    805 
    806   // The RHS is known non-zero.
    807   if (Value *V = simplifyValueKnownNonZero(I.getOperand(1), *this, I)) {
    808     I.setOperand(1, V);
    809     return &I;
    810   }
    811 
    812   // Handle cases involving: [su]div X, (select Cond, Y, Z)
    813   // This does not apply for fdiv.
    814   if (isa<SelectInst>(Op1) && SimplifyDivRemOfSelect(I))
    815     return &I;
    816 
    817   if (Instruction *LHS = dyn_cast<Instruction>(Op0)) {
    818     const APInt *C2;
    819     if (match(Op1, m_APInt(C2))) {
    820       Value *X;
    821       const APInt *C1;
    822       bool IsSigned = I.getOpcode() == Instruction::SDiv;
    823 
    824       // (X / C1) / C2  -> X / (C1*C2)
    825       if ((IsSigned && match(LHS, m_SDiv(m_Value(X), m_APInt(C1)))) ||
    826           (!IsSigned && match(LHS, m_UDiv(m_Value(X), m_APInt(C1))))) {
    827         APInt Product(C1->getBitWidth(), /*Val=*/0ULL, IsSigned);
    828         if (!MultiplyOverflows(*C1, *C2, Product, IsSigned))
    829           return BinaryOperator::Create(I.getOpcode(), X,
    830                                         ConstantInt::get(I.getType(), Product));
    831       }
    832 
    833       if ((IsSigned && match(LHS, m_NSWMul(m_Value(X), m_APInt(C1)))) ||
    834           (!IsSigned && match(LHS, m_NUWMul(m_Value(X), m_APInt(C1))))) {
    835         APInt Quotient(C1->getBitWidth(), /*Val=*/0ULL, IsSigned);
    836 
    837         // (X * C1) / C2 -> X / (C2 / C1) if C2 is a multiple of C1.
    838         if (IsMultiple(*C2, *C1, Quotient, IsSigned)) {
    839           BinaryOperator *BO = BinaryOperator::Create(
    840               I.getOpcode(), X, ConstantInt::get(X->getType(), Quotient));
    841           BO->setIsExact(I.isExact());
    842           return BO;
    843         }
    844 
    845         // (X * C1) / C2 -> X * (C1 / C2) if C1 is a multiple of C2.
    846         if (IsMultiple(*C1, *C2, Quotient, IsSigned)) {
    847           BinaryOperator *BO = BinaryOperator::Create(
    848               Instruction::Mul, X, ConstantInt::get(X->getType(), Quotient));
    849           BO->setHasNoUnsignedWrap(
    850               !IsSigned &&
    851               cast<OverflowingBinaryOperator>(LHS)->hasNoUnsignedWrap());
    852           BO->setHasNoSignedWrap(
    853               cast<OverflowingBinaryOperator>(LHS)->hasNoSignedWrap());
    854           return BO;
    855         }
    856       }
    857 
    858       if ((IsSigned && match(LHS, m_NSWShl(m_Value(X), m_APInt(C1))) &&
    859            *C1 != C1->getBitWidth() - 1) ||
    860           (!IsSigned && match(LHS, m_NUWShl(m_Value(X), m_APInt(C1))))) {
    861         APInt Quotient(C1->getBitWidth(), /*Val=*/0ULL, IsSigned);
    862         APInt C1Shifted = APInt::getOneBitSet(
    863             C1->getBitWidth(), static_cast<unsigned>(C1->getLimitedValue()));
    864 
    865         // (X << C1) / C2 -> X / (C2 >> C1) if C2 is a multiple of C1.
    866         if (IsMultiple(*C2, C1Shifted, Quotient, IsSigned)) {
    867           BinaryOperator *BO = BinaryOperator::Create(
    868               I.getOpcode(), X, ConstantInt::get(X->getType(), Quotient));
    869           BO->setIsExact(I.isExact());
    870           return BO;
    871         }
    872 
    873         // (X << C1) / C2 -> X * (C2 >> C1) if C1 is a multiple of C2.
    874         if (IsMultiple(C1Shifted, *C2, Quotient, IsSigned)) {
    875           BinaryOperator *BO = BinaryOperator::Create(
    876               Instruction::Mul, X, ConstantInt::get(X->getType(), Quotient));
    877           BO->setHasNoUnsignedWrap(
    878               !IsSigned &&
    879               cast<OverflowingBinaryOperator>(LHS)->hasNoUnsignedWrap());
    880           BO->setHasNoSignedWrap(
    881               cast<OverflowingBinaryOperator>(LHS)->hasNoSignedWrap());
    882           return BO;
    883         }
    884       }
    885 
    886       if (*C2 != 0) { // avoid X udiv 0
    887         if (SelectInst *SI = dyn_cast<SelectInst>(Op0))
    888           if (Instruction *R = FoldOpIntoSelect(I, SI))
    889             return R;
    890         if (isa<PHINode>(Op0))
    891           if (Instruction *NV = FoldOpIntoPhi(I))
    892             return NV;
    893       }
    894     }
    895   }
    896 
    897   if (ConstantInt *One = dyn_cast<ConstantInt>(Op0)) {
    898     if (One->isOne() && !I.getType()->isIntegerTy(1)) {
    899       bool isSigned = I.getOpcode() == Instruction::SDiv;
    900       if (isSigned) {
    901         // If Op1 is 0 then it's undefined behaviour, if Op1 is 1 then the
    902         // result is one, if Op1 is -1 then the result is minus one, otherwise
    903         // it's zero.
    904         Value *Inc = Builder->CreateAdd(Op1, One);
    905         Value *Cmp = Builder->CreateICmpULT(
    906                          Inc, ConstantInt::get(I.getType(), 3));
    907         return SelectInst::Create(Cmp, Op1, ConstantInt::get(I.getType(), 0));
    908       } else {
    909         // If Op1 is 0 then it's undefined behaviour. If Op1 is 1 then the
    910         // result is one, otherwise it's zero.
    911         return new ZExtInst(Builder->CreateICmpEQ(Op1, One), I.getType());
    912       }
    913     }
    914   }
    915 
    916   // See if we can fold away this div instruction.
    917   if (SimplifyDemandedInstructionBits(I))
    918     return &I;
    919 
    920   // (X - (X rem Y)) / Y -> X / Y; usually originates as ((X / Y) * Y) / Y
    921   Value *X = nullptr, *Z = nullptr;
    922   if (match(Op0, m_Sub(m_Value(X), m_Value(Z)))) { // (X - Z) / Y; Y = Op1
    923     bool isSigned = I.getOpcode() == Instruction::SDiv;
    924     if ((isSigned && match(Z, m_SRem(m_Specific(X), m_Specific(Op1)))) ||
    925         (!isSigned && match(Z, m_URem(m_Specific(X), m_Specific(Op1)))))
    926       return BinaryOperator::Create(I.getOpcode(), X, Op1);
    927   }
    928 
    929   return nullptr;
    930 }
    931 
    932 /// dyn_castZExtVal - Checks if V is a zext or constant that can
    933 /// be truncated to Ty without losing bits.
    934 static Value *dyn_castZExtVal(Value *V, Type *Ty) {
    935   if (ZExtInst *Z = dyn_cast<ZExtInst>(V)) {
    936     if (Z->getSrcTy() == Ty)
    937       return Z->getOperand(0);
    938   } else if (ConstantInt *C = dyn_cast<ConstantInt>(V)) {
    939     if (C->getValue().getActiveBits() <= cast<IntegerType>(Ty)->getBitWidth())
    940       return ConstantExpr::getTrunc(C, Ty);
    941   }
    942   return nullptr;
    943 }
    944 
    945 namespace {
    946 const unsigned MaxDepth = 6;
    947 typedef Instruction *(*FoldUDivOperandCb)(Value *Op0, Value *Op1,
    948                                           const BinaryOperator &I,
    949                                           InstCombiner &IC);
    950 
    951 /// \brief Used to maintain state for visitUDivOperand().
    952 struct UDivFoldAction {
    953   FoldUDivOperandCb FoldAction; ///< Informs visitUDiv() how to fold this
    954                                 ///< operand.  This can be zero if this action
    955                                 ///< joins two actions together.
    956 
    957   Value *OperandToFold;         ///< Which operand to fold.
    958   union {
    959     Instruction *FoldResult;    ///< The instruction returned when FoldAction is
    960                                 ///< invoked.
    961 
    962     size_t SelectLHSIdx;        ///< Stores the LHS action index if this action
    963                                 ///< joins two actions together.
    964   };
    965 
    966   UDivFoldAction(FoldUDivOperandCb FA, Value *InputOperand)
    967       : FoldAction(FA), OperandToFold(InputOperand), FoldResult(nullptr) {}
    968   UDivFoldAction(FoldUDivOperandCb FA, Value *InputOperand, size_t SLHS)
    969       : FoldAction(FA), OperandToFold(InputOperand), SelectLHSIdx(SLHS) {}
    970 };
    971 }
    972 
    973 // X udiv 2^C -> X >> C
    974 static Instruction *foldUDivPow2Cst(Value *Op0, Value *Op1,
    975                                     const BinaryOperator &I, InstCombiner &IC) {
    976   const APInt &C = cast<Constant>(Op1)->getUniqueInteger();
    977   BinaryOperator *LShr = BinaryOperator::CreateLShr(
    978       Op0, ConstantInt::get(Op0->getType(), C.logBase2()));
    979   if (I.isExact())
    980     LShr->setIsExact();
    981   return LShr;
    982 }
    983 
    984 // X udiv C, where C >= signbit
    985 static Instruction *foldUDivNegCst(Value *Op0, Value *Op1,
    986                                    const BinaryOperator &I, InstCombiner &IC) {
    987   Value *ICI = IC.Builder->CreateICmpULT(Op0, cast<ConstantInt>(Op1));
    988 
    989   return SelectInst::Create(ICI, Constant::getNullValue(I.getType()),
    990                             ConstantInt::get(I.getType(), 1));
    991 }
    992 
    993 // X udiv (C1 << N), where C1 is "1<<C2"  -->  X >> (N+C2)
    994 static Instruction *foldUDivShl(Value *Op0, Value *Op1, const BinaryOperator &I,
    995                                 InstCombiner &IC) {
    996   Instruction *ShiftLeft = cast<Instruction>(Op1);
    997   if (isa<ZExtInst>(ShiftLeft))
    998     ShiftLeft = cast<Instruction>(ShiftLeft->getOperand(0));
    999 
   1000   const APInt &CI =
   1001       cast<Constant>(ShiftLeft->getOperand(0))->getUniqueInteger();
   1002   Value *N = ShiftLeft->getOperand(1);
   1003   if (CI != 1)
   1004     N = IC.Builder->CreateAdd(N, ConstantInt::get(N->getType(), CI.logBase2()));
   1005   if (ZExtInst *Z = dyn_cast<ZExtInst>(Op1))
   1006     N = IC.Builder->CreateZExt(N, Z->getDestTy());
   1007   BinaryOperator *LShr = BinaryOperator::CreateLShr(Op0, N);
   1008   if (I.isExact())
   1009     LShr->setIsExact();
   1010   return LShr;
   1011 }
   1012 
   1013 // \brief Recursively visits the possible right hand operands of a udiv
   1014 // instruction, seeing through select instructions, to determine if we can
   1015 // replace the udiv with something simpler.  If we find that an operand is not
   1016 // able to simplify the udiv, we abort the entire transformation.
   1017 static size_t visitUDivOperand(Value *Op0, Value *Op1, const BinaryOperator &I,
   1018                                SmallVectorImpl<UDivFoldAction> &Actions,
   1019                                unsigned Depth = 0) {
   1020   // Check to see if this is an unsigned division with an exact power of 2,
   1021   // if so, convert to a right shift.
   1022   if (match(Op1, m_Power2())) {
   1023     Actions.push_back(UDivFoldAction(foldUDivPow2Cst, Op1));
   1024     return Actions.size();
   1025   }
   1026 
   1027   if (ConstantInt *C = dyn_cast<ConstantInt>(Op1))
   1028     // X udiv C, where C >= signbit
   1029     if (C->getValue().isNegative()) {
   1030       Actions.push_back(UDivFoldAction(foldUDivNegCst, C));
   1031       return Actions.size();
   1032     }
   1033 
   1034   // X udiv (C1 << N), where C1 is "1<<C2"  -->  X >> (N+C2)
   1035   if (match(Op1, m_Shl(m_Power2(), m_Value())) ||
   1036       match(Op1, m_ZExt(m_Shl(m_Power2(), m_Value())))) {
   1037     Actions.push_back(UDivFoldAction(foldUDivShl, Op1));
   1038     return Actions.size();
   1039   }
   1040 
   1041   // The remaining tests are all recursive, so bail out if we hit the limit.
   1042   if (Depth++ == MaxDepth)
   1043     return 0;
   1044 
   1045   if (SelectInst *SI = dyn_cast<SelectInst>(Op1))
   1046     if (size_t LHSIdx =
   1047             visitUDivOperand(Op0, SI->getOperand(1), I, Actions, Depth))
   1048       if (visitUDivOperand(Op0, SI->getOperand(2), I, Actions, Depth)) {
   1049         Actions.push_back(UDivFoldAction(nullptr, Op1, LHSIdx - 1));
   1050         return Actions.size();
   1051       }
   1052 
   1053   return 0;
   1054 }
   1055 
   1056 Instruction *InstCombiner::visitUDiv(BinaryOperator &I) {
   1057   Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);
   1058 
   1059   if (Value *V = SimplifyVectorOp(I))
   1060     return replaceInstUsesWith(I, V);
   1061 
   1062   if (Value *V = SimplifyUDivInst(Op0, Op1, DL, TLI, DT, AC))
   1063     return replaceInstUsesWith(I, V);
   1064 
   1065   // Handle the integer div common cases
   1066   if (Instruction *Common = commonIDivTransforms(I))
   1067     return Common;
   1068 
   1069   // (x lshr C1) udiv C2 --> x udiv (C2 << C1)
   1070   {
   1071     Value *X;
   1072     const APInt *C1, *C2;
   1073     if (match(Op0, m_LShr(m_Value(X), m_APInt(C1))) &&
   1074         match(Op1, m_APInt(C2))) {
   1075       bool Overflow;
   1076       APInt C2ShlC1 = C2->ushl_ov(*C1, Overflow);
   1077       if (!Overflow) {
   1078         bool IsExact = I.isExact() && match(Op0, m_Exact(m_Value()));
   1079         BinaryOperator *BO = BinaryOperator::CreateUDiv(
   1080             X, ConstantInt::get(X->getType(), C2ShlC1));
   1081         if (IsExact)
   1082           BO->setIsExact();
   1083         return BO;
   1084       }
   1085     }
   1086   }
   1087 
   1088   // (zext A) udiv (zext B) --> zext (A udiv B)
   1089   if (ZExtInst *ZOp0 = dyn_cast<ZExtInst>(Op0))
   1090     if (Value *ZOp1 = dyn_castZExtVal(Op1, ZOp0->getSrcTy()))
   1091       return new ZExtInst(
   1092           Builder->CreateUDiv(ZOp0->getOperand(0), ZOp1, "div", I.isExact()),
   1093           I.getType());
   1094 
   1095   // (LHS udiv (select (select (...)))) -> (LHS >> (select (select (...))))
   1096   SmallVector<UDivFoldAction, 6> UDivActions;
   1097   if (visitUDivOperand(Op0, Op1, I, UDivActions))
   1098     for (unsigned i = 0, e = UDivActions.size(); i != e; ++i) {
   1099       FoldUDivOperandCb Action = UDivActions[i].FoldAction;
   1100       Value *ActionOp1 = UDivActions[i].OperandToFold;
   1101       Instruction *Inst;
   1102       if (Action)
   1103         Inst = Action(Op0, ActionOp1, I, *this);
   1104       else {
   1105         // This action joins two actions together.  The RHS of this action is
   1106         // simply the last action we processed, we saved the LHS action index in
   1107         // the joining action.
   1108         size_t SelectRHSIdx = i - 1;
   1109         Value *SelectRHS = UDivActions[SelectRHSIdx].FoldResult;
   1110         size_t SelectLHSIdx = UDivActions[i].SelectLHSIdx;
   1111         Value *SelectLHS = UDivActions[SelectLHSIdx].FoldResult;
   1112         Inst = SelectInst::Create(cast<SelectInst>(ActionOp1)->getCondition(),
   1113                                   SelectLHS, SelectRHS);
   1114       }
   1115 
   1116       // If this is the last action to process, return it to the InstCombiner.
   1117       // Otherwise, we insert it before the UDiv and record it so that we may
   1118       // use it as part of a joining action (i.e., a SelectInst).
   1119       if (e - i != 1) {
   1120         Inst->insertBefore(&I);
   1121         UDivActions[i].FoldResult = Inst;
   1122       } else
   1123         return Inst;
   1124     }
   1125 
   1126   return nullptr;
   1127 }
   1128 
   1129 Instruction *InstCombiner::visitSDiv(BinaryOperator &I) {
   1130   Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);
   1131 
   1132   if (Value *V = SimplifyVectorOp(I))
   1133     return replaceInstUsesWith(I, V);
   1134 
   1135   if (Value *V = SimplifySDivInst(Op0, Op1, DL, TLI, DT, AC))
   1136     return replaceInstUsesWith(I, V);
   1137 
   1138   // Handle the integer div common cases
   1139   if (Instruction *Common = commonIDivTransforms(I))
   1140     return Common;
   1141 
   1142   const APInt *Op1C;
   1143   if (match(Op1, m_APInt(Op1C))) {
   1144     // sdiv X, -1 == -X
   1145     if (Op1C->isAllOnesValue())
   1146       return BinaryOperator::CreateNeg(Op0);
   1147 
   1148     // sdiv exact X, C  -->  ashr exact X, log2(C)
   1149     if (I.isExact() && Op1C->isNonNegative() && Op1C->isPowerOf2()) {
   1150       Value *ShAmt = ConstantInt::get(Op1->getType(), Op1C->exactLogBase2());
   1151       return BinaryOperator::CreateExactAShr(Op0, ShAmt, I.getName());
   1152     }
   1153 
   1154     // If the dividend is sign-extended and the constant divisor is small enough
   1155     // to fit in the source type, shrink the division to the narrower type:
   1156     // (sext X) sdiv C --> sext (X sdiv C)
   1157     Value *Op0Src;
   1158     if (match(Op0, m_OneUse(m_SExt(m_Value(Op0Src)))) &&
   1159         Op0Src->getType()->getScalarSizeInBits() >= Op1C->getMinSignedBits()) {
   1160 
   1161       // In the general case, we need to make sure that the dividend is not the
   1162       // minimum signed value because dividing that by -1 is UB. But here, we
   1163       // know that the -1 divisor case is already handled above.
   1164 
   1165       Constant *NarrowDivisor =
   1166           ConstantExpr::getTrunc(cast<Constant>(Op1), Op0Src->getType());
   1167       Value *NarrowOp = Builder->CreateSDiv(Op0Src, NarrowDivisor);
   1168       return new SExtInst(NarrowOp, Op0->getType());
   1169     }
   1170   }
   1171 
   1172   if (Constant *RHS = dyn_cast<Constant>(Op1)) {
   1173     // X/INT_MIN -> X == INT_MIN
   1174     if (RHS->isMinSignedValue())
   1175       return new ZExtInst(Builder->CreateICmpEQ(Op0, Op1), I.getType());
   1176 
   1177     // -X/C  -->  X/-C  provided the negation doesn't overflow.
   1178     Value *X;
   1179     if (match(Op0, m_NSWSub(m_Zero(), m_Value(X)))) {
   1180       auto *BO = BinaryOperator::CreateSDiv(X, ConstantExpr::getNeg(RHS));
   1181       BO->setIsExact(I.isExact());
   1182       return BO;
   1183     }
   1184   }
   1185 
   1186   // If the sign bits of both operands are zero (i.e. we can prove they are
   1187   // unsigned inputs), turn this into a udiv.
   1188   if (I.getType()->isIntegerTy()) {
   1189     APInt Mask(APInt::getSignBit(I.getType()->getPrimitiveSizeInBits()));
   1190     if (MaskedValueIsZero(Op0, Mask, 0, &I)) {
   1191       if (MaskedValueIsZero(Op1, Mask, 0, &I)) {
   1192         // X sdiv Y -> X udiv Y, iff X and Y don't have sign bit set
   1193         auto *BO = BinaryOperator::CreateUDiv(Op0, Op1, I.getName());
   1194         BO->setIsExact(I.isExact());
   1195         return BO;
   1196       }
   1197 
   1198       if (isKnownToBeAPowerOfTwo(Op1, DL, /*OrZero*/ true, 0, AC, &I, DT)) {
   1199         // X sdiv (1 << Y) -> X udiv (1 << Y) ( -> X u>> Y)
   1200         // Safe because the only negative value (1 << Y) can take on is
   1201         // INT_MIN, and X sdiv INT_MIN == X udiv INT_MIN == 0 if X doesn't have
   1202         // the sign bit set.
   1203         auto *BO = BinaryOperator::CreateUDiv(Op0, Op1, I.getName());
   1204         BO->setIsExact(I.isExact());
   1205         return BO;
   1206       }
   1207     }
   1208   }
   1209 
   1210   return nullptr;
   1211 }
   1212 
   1213 /// CvtFDivConstToReciprocal tries to convert X/C into X*1/C if C not a special
   1214 /// FP value and:
   1215 ///    1) 1/C is exact, or
   1216 ///    2) reciprocal is allowed.
   1217 /// If the conversion was successful, the simplified expression "X * 1/C" is
   1218 /// returned; otherwise, NULL is returned.
   1219 ///
   1220 static Instruction *CvtFDivConstToReciprocal(Value *Dividend, Constant *Divisor,
   1221                                              bool AllowReciprocal) {
   1222   if (!isa<ConstantFP>(Divisor)) // TODO: handle vectors.
   1223     return nullptr;
   1224 
   1225   const APFloat &FpVal = cast<ConstantFP>(Divisor)->getValueAPF();
   1226   APFloat Reciprocal(FpVal.getSemantics());
   1227   bool Cvt = FpVal.getExactInverse(&Reciprocal);
   1228 
   1229   if (!Cvt && AllowReciprocal && FpVal.isFiniteNonZero()) {
   1230     Reciprocal = APFloat(FpVal.getSemantics(), 1.0f);
   1231     (void)Reciprocal.divide(FpVal, APFloat::rmNearestTiesToEven);
   1232     Cvt = !Reciprocal.isDenormal();
   1233   }
   1234 
   1235   if (!Cvt)
   1236     return nullptr;
   1237 
   1238   ConstantFP *R;
   1239   R = ConstantFP::get(Dividend->getType()->getContext(), Reciprocal);
   1240   return BinaryOperator::CreateFMul(Dividend, R);
   1241 }
   1242 
   1243 Instruction *InstCombiner::visitFDiv(BinaryOperator &I) {
   1244   Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);
   1245 
   1246   if (Value *V = SimplifyVectorOp(I))
   1247     return replaceInstUsesWith(I, V);
   1248 
   1249   if (Value *V = SimplifyFDivInst(Op0, Op1, I.getFastMathFlags(),
   1250                                   DL, TLI, DT, AC))
   1251     return replaceInstUsesWith(I, V);
   1252 
   1253   if (isa<Constant>(Op0))
   1254     if (SelectInst *SI = dyn_cast<SelectInst>(Op1))
   1255       if (Instruction *R = FoldOpIntoSelect(I, SI))
   1256         return R;
   1257 
   1258   bool AllowReassociate = I.hasUnsafeAlgebra();
   1259   bool AllowReciprocal = I.hasAllowReciprocal();
   1260 
   1261   if (Constant *Op1C = dyn_cast<Constant>(Op1)) {
   1262     if (SelectInst *SI = dyn_cast<SelectInst>(Op0))
   1263       if (Instruction *R = FoldOpIntoSelect(I, SI))
   1264         return R;
   1265 
   1266     if (AllowReassociate) {
   1267       Constant *C1 = nullptr;
   1268       Constant *C2 = Op1C;
   1269       Value *X;
   1270       Instruction *Res = nullptr;
   1271 
   1272       if (match(Op0, m_FMul(m_Value(X), m_Constant(C1)))) {
   1273         // (X*C1)/C2 => X * (C1/C2)
   1274         //
   1275         Constant *C = ConstantExpr::getFDiv(C1, C2);
   1276         if (isNormalFp(C))
   1277           Res = BinaryOperator::CreateFMul(X, C);
   1278       } else if (match(Op0, m_FDiv(m_Value(X), m_Constant(C1)))) {
   1279         // (X/C1)/C2 => X /(C2*C1) [=> X * 1/(C2*C1) if reciprocal is allowed]
   1280         //
   1281         Constant *C = ConstantExpr::getFMul(C1, C2);
   1282         if (isNormalFp(C)) {
   1283           Res = CvtFDivConstToReciprocal(X, C, AllowReciprocal);
   1284           if (!Res)
   1285             Res = BinaryOperator::CreateFDiv(X, C);
   1286         }
   1287       }
   1288 
   1289       if (Res) {
   1290         Res->setFastMathFlags(I.getFastMathFlags());
   1291         return Res;
   1292       }
   1293     }
   1294 
   1295     // X / C => X * 1/C
   1296     if (Instruction *T = CvtFDivConstToReciprocal(Op0, Op1C, AllowReciprocal)) {
   1297       T->copyFastMathFlags(&I);
   1298       return T;
   1299     }
   1300 
   1301     return nullptr;
   1302   }
   1303 
   1304   if (AllowReassociate && isa<Constant>(Op0)) {
   1305     Constant *C1 = cast<Constant>(Op0), *C2;
   1306     Constant *Fold = nullptr;
   1307     Value *X;
   1308     bool CreateDiv = true;
   1309 
   1310     // C1 / (X*C2) => (C1/C2) / X
   1311     if (match(Op1, m_FMul(m_Value(X), m_Constant(C2))))
   1312       Fold = ConstantExpr::getFDiv(C1, C2);
   1313     else if (match(Op1, m_FDiv(m_Value(X), m_Constant(C2)))) {
   1314       // C1 / (X/C2) => (C1*C2) / X
   1315       Fold = ConstantExpr::getFMul(C1, C2);
   1316     } else if (match(Op1, m_FDiv(m_Constant(C2), m_Value(X)))) {
   1317       // C1 / (C2/X) => (C1/C2) * X
   1318       Fold = ConstantExpr::getFDiv(C1, C2);
   1319       CreateDiv = false;
   1320     }
   1321 
   1322     if (Fold && isNormalFp(Fold)) {
   1323       Instruction *R = CreateDiv ? BinaryOperator::CreateFDiv(Fold, X)
   1324                                  : BinaryOperator::CreateFMul(X, Fold);
   1325       R->setFastMathFlags(I.getFastMathFlags());
   1326       return R;
   1327     }
   1328     return nullptr;
   1329   }
   1330 
   1331   if (AllowReassociate) {
   1332     Value *X, *Y;
   1333     Value *NewInst = nullptr;
   1334     Instruction *SimpR = nullptr;
   1335 
   1336     if (Op0->hasOneUse() && match(Op0, m_FDiv(m_Value(X), m_Value(Y)))) {
   1337       // (X/Y) / Z => X / (Y*Z)
   1338       //
   1339       if (!isa<Constant>(Y) || !isa<Constant>(Op1)) {
   1340         NewInst = Builder->CreateFMul(Y, Op1);
   1341         if (Instruction *RI = dyn_cast<Instruction>(NewInst)) {
   1342           FastMathFlags Flags = I.getFastMathFlags();
   1343           Flags &= cast<Instruction>(Op0)->getFastMathFlags();
   1344           RI->setFastMathFlags(Flags);
   1345         }
   1346         SimpR = BinaryOperator::CreateFDiv(X, NewInst);
   1347       }
   1348     } else if (Op1->hasOneUse() && match(Op1, m_FDiv(m_Value(X), m_Value(Y)))) {
   1349       // Z / (X/Y) => Z*Y / X
   1350       //
   1351       if (!isa<Constant>(Y) || !isa<Constant>(Op0)) {
   1352         NewInst = Builder->CreateFMul(Op0, Y);
   1353         if (Instruction *RI = dyn_cast<Instruction>(NewInst)) {
   1354           FastMathFlags Flags = I.getFastMathFlags();
   1355           Flags &= cast<Instruction>(Op1)->getFastMathFlags();
   1356           RI->setFastMathFlags(Flags);
   1357         }
   1358         SimpR = BinaryOperator::CreateFDiv(NewInst, X);
   1359       }
   1360     }
   1361 
   1362     if (NewInst) {
   1363       if (Instruction *T = dyn_cast<Instruction>(NewInst))
   1364         T->setDebugLoc(I.getDebugLoc());
   1365       SimpR->setFastMathFlags(I.getFastMathFlags());
   1366       return SimpR;
   1367     }
   1368   }
   1369 
   1370   return nullptr;
   1371 }
   1372 
   1373 /// This function implements the transforms common to both integer remainder
   1374 /// instructions (urem and srem). It is called by the visitors to those integer
   1375 /// remainder instructions.
   1376 /// @brief Common integer remainder transforms
   1377 Instruction *InstCombiner::commonIRemTransforms(BinaryOperator &I) {
   1378   Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);
   1379 
   1380   // The RHS is known non-zero.
   1381   if (Value *V = simplifyValueKnownNonZero(I.getOperand(1), *this, I)) {
   1382     I.setOperand(1, V);
   1383     return &I;
   1384   }
   1385 
   1386   // Handle cases involving: rem X, (select Cond, Y, Z)
   1387   if (isa<SelectInst>(Op1) && SimplifyDivRemOfSelect(I))
   1388     return &I;
   1389 
   1390   if (isa<Constant>(Op1)) {
   1391     if (Instruction *Op0I = dyn_cast<Instruction>(Op0)) {
   1392       if (SelectInst *SI = dyn_cast<SelectInst>(Op0I)) {
   1393         if (Instruction *R = FoldOpIntoSelect(I, SI))
   1394           return R;
   1395       } else if (isa<PHINode>(Op0I)) {
   1396         using namespace llvm::PatternMatch;
   1397         const APInt *Op1Int;
   1398         if (match(Op1, m_APInt(Op1Int)) && !Op1Int->isMinValue() &&
   1399             (I.getOpcode() == Instruction::URem ||
   1400              !Op1Int->isMinSignedValue())) {
   1401           // FoldOpIntoPhi will speculate instructions to the end of the PHI's
   1402           // predecessor blocks, so do this only if we know the srem or urem
   1403           // will not fault.
   1404           if (Instruction *NV = FoldOpIntoPhi(I))
   1405             return NV;
   1406         }
   1407       }
   1408 
   1409       // See if we can fold away this rem instruction.
   1410       if (SimplifyDemandedInstructionBits(I))
   1411         return &I;
   1412     }
   1413   }
   1414 
   1415   return nullptr;
   1416 }
   1417 
   1418 Instruction *InstCombiner::visitURem(BinaryOperator &I) {
   1419   Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);
   1420 
   1421   if (Value *V = SimplifyVectorOp(I))
   1422     return replaceInstUsesWith(I, V);
   1423 
   1424   if (Value *V = SimplifyURemInst(Op0, Op1, DL, TLI, DT, AC))
   1425     return replaceInstUsesWith(I, V);
   1426 
   1427   if (Instruction *common = commonIRemTransforms(I))
   1428     return common;
   1429 
   1430   // (zext A) urem (zext B) --> zext (A urem B)
   1431   if (ZExtInst *ZOp0 = dyn_cast<ZExtInst>(Op0))
   1432     if (Value *ZOp1 = dyn_castZExtVal(Op1, ZOp0->getSrcTy()))
   1433       return new ZExtInst(Builder->CreateURem(ZOp0->getOperand(0), ZOp1),
   1434                           I.getType());
   1435 
   1436   // X urem Y -> X and Y-1, where Y is a power of 2,
   1437   if (isKnownToBeAPowerOfTwo(Op1, DL, /*OrZero*/ true, 0, AC, &I, DT)) {
   1438     Constant *N1 = Constant::getAllOnesValue(I.getType());
   1439     Value *Add = Builder->CreateAdd(Op1, N1);
   1440     return BinaryOperator::CreateAnd(Op0, Add);
   1441   }
   1442 
   1443   // 1 urem X -> zext(X != 1)
   1444   if (match(Op0, m_One())) {
   1445     Value *Cmp = Builder->CreateICmpNE(Op1, Op0);
   1446     Value *Ext = Builder->CreateZExt(Cmp, I.getType());
   1447     return replaceInstUsesWith(I, Ext);
   1448   }
   1449 
   1450   return nullptr;
   1451 }
   1452 
   1453 Instruction *InstCombiner::visitSRem(BinaryOperator &I) {
   1454   Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);
   1455 
   1456   if (Value *V = SimplifyVectorOp(I))
   1457     return replaceInstUsesWith(I, V);
   1458 
   1459   if (Value *V = SimplifySRemInst(Op0, Op1, DL, TLI, DT, AC))
   1460     return replaceInstUsesWith(I, V);
   1461 
   1462   // Handle the integer rem common cases
   1463   if (Instruction *Common = commonIRemTransforms(I))
   1464     return Common;
   1465 
   1466   {
   1467     const APInt *Y;
   1468     // X % -Y -> X % Y
   1469     if (match(Op1, m_APInt(Y)) && Y->isNegative() && !Y->isMinSignedValue()) {
   1470       Worklist.AddValue(I.getOperand(1));
   1471       I.setOperand(1, ConstantInt::get(I.getType(), -*Y));
   1472       return &I;
   1473     }
   1474   }
   1475 
   1476   // If the sign bits of both operands are zero (i.e. we can prove they are
   1477   // unsigned inputs), turn this into a urem.
   1478   if (I.getType()->isIntegerTy()) {
   1479     APInt Mask(APInt::getSignBit(I.getType()->getPrimitiveSizeInBits()));
   1480     if (MaskedValueIsZero(Op1, Mask, 0, &I) &&
   1481         MaskedValueIsZero(Op0, Mask, 0, &I)) {
   1482       // X srem Y -> X urem Y, iff X and Y don't have sign bit set
   1483       return BinaryOperator::CreateURem(Op0, Op1, I.getName());
   1484     }
   1485   }
   1486 
   1487   // If it's a constant vector, flip any negative values positive.
   1488   if (isa<ConstantVector>(Op1) || isa<ConstantDataVector>(Op1)) {
   1489     Constant *C = cast<Constant>(Op1);
   1490     unsigned VWidth = C->getType()->getVectorNumElements();
   1491 
   1492     bool hasNegative = false;
   1493     bool hasMissing = false;
   1494     for (unsigned i = 0; i != VWidth; ++i) {
   1495       Constant *Elt = C->getAggregateElement(i);
   1496       if (!Elt) {
   1497         hasMissing = true;
   1498         break;
   1499       }
   1500 
   1501       if (ConstantInt *RHS = dyn_cast<ConstantInt>(Elt))
   1502         if (RHS->isNegative())
   1503           hasNegative = true;
   1504     }
   1505 
   1506     if (hasNegative && !hasMissing) {
   1507       SmallVector<Constant *, 16> Elts(VWidth);
   1508       for (unsigned i = 0; i != VWidth; ++i) {
   1509         Elts[i] = C->getAggregateElement(i);  // Handle undef, etc.
   1510         if (ConstantInt *RHS = dyn_cast<ConstantInt>(Elts[i])) {
   1511           if (RHS->isNegative())
   1512             Elts[i] = cast<ConstantInt>(ConstantExpr::getNeg(RHS));
   1513         }
   1514       }
   1515 
   1516       Constant *NewRHSV = ConstantVector::get(Elts);
   1517       if (NewRHSV != C) {  // Don't loop on -MININT
   1518         Worklist.AddValue(I.getOperand(1));
   1519         I.setOperand(1, NewRHSV);
   1520         return &I;
   1521       }
   1522     }
   1523   }
   1524 
   1525   return nullptr;
   1526 }
   1527 
   1528 Instruction *InstCombiner::visitFRem(BinaryOperator &I) {
   1529   Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);
   1530 
   1531   if (Value *V = SimplifyVectorOp(I))
   1532     return replaceInstUsesWith(I, V);
   1533 
   1534   if (Value *V = SimplifyFRemInst(Op0, Op1, I.getFastMathFlags(),
   1535                                   DL, TLI, DT, AC))
   1536     return replaceInstUsesWith(I, V);
   1537 
   1538   // Handle cases involving: rem X, (select Cond, Y, Z)
   1539   if (isa<SelectInst>(Op1) && SimplifyDivRemOfSelect(I))
   1540     return &I;
   1541 
   1542   return nullptr;
   1543 }
   1544