Home | History | Annotate | Download | only in InstCombine
      1 //===- InstCombineVectorOps.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 instcombine for ExtractElement, InsertElement and
     11 // ShuffleVector.
     12 //
     13 //===----------------------------------------------------------------------===//
     14 
     15 #include "InstCombineInternal.h"
     16 #include "llvm/ADT/DenseMap.h"
     17 #include "llvm/Analysis/InstructionSimplify.h"
     18 #include "llvm/Analysis/VectorUtils.h"
     19 #include "llvm/IR/PatternMatch.h"
     20 using namespace llvm;
     21 using namespace PatternMatch;
     22 
     23 #define DEBUG_TYPE "instcombine"
     24 
     25 /// Return true if the value is cheaper to scalarize than it is to leave as a
     26 /// vector operation. isConstant indicates whether we're extracting one known
     27 /// element. If false we're extracting a variable index.
     28 static bool cheapToScalarize(Value *V, bool isConstant) {
     29   if (Constant *C = dyn_cast<Constant>(V)) {
     30     if (isConstant) return true;
     31 
     32     // If all elts are the same, we can extract it and use any of the values.
     33     if (Constant *Op0 = C->getAggregateElement(0U)) {
     34       for (unsigned i = 1, e = V->getType()->getVectorNumElements(); i != e;
     35            ++i)
     36         if (C->getAggregateElement(i) != Op0)
     37           return false;
     38       return true;
     39     }
     40   }
     41   Instruction *I = dyn_cast<Instruction>(V);
     42   if (!I) return false;
     43 
     44   // Insert element gets simplified to the inserted element or is deleted if
     45   // this is constant idx extract element and its a constant idx insertelt.
     46   if (I->getOpcode() == Instruction::InsertElement && isConstant &&
     47       isa<ConstantInt>(I->getOperand(2)))
     48     return true;
     49   if (I->getOpcode() == Instruction::Load && I->hasOneUse())
     50     return true;
     51   if (BinaryOperator *BO = dyn_cast<BinaryOperator>(I))
     52     if (BO->hasOneUse() &&
     53         (cheapToScalarize(BO->getOperand(0), isConstant) ||
     54          cheapToScalarize(BO->getOperand(1), isConstant)))
     55       return true;
     56   if (CmpInst *CI = dyn_cast<CmpInst>(I))
     57     if (CI->hasOneUse() &&
     58         (cheapToScalarize(CI->getOperand(0), isConstant) ||
     59          cheapToScalarize(CI->getOperand(1), isConstant)))
     60       return true;
     61 
     62   return false;
     63 }
     64 
     65 // If we have a PHI node with a vector type that has only 2 uses: feed
     66 // itself and be an operand of extractelement at a constant location,
     67 // try to replace the PHI of the vector type with a PHI of a scalar type.
     68 Instruction *InstCombiner::scalarizePHI(ExtractElementInst &EI, PHINode *PN) {
     69   // Verify that the PHI node has exactly 2 uses. Otherwise return NULL.
     70   if (!PN->hasNUses(2))
     71     return nullptr;
     72 
     73   // If so, it's known at this point that one operand is PHI and the other is
     74   // an extractelement node. Find the PHI user that is not the extractelement
     75   // node.
     76   auto iu = PN->user_begin();
     77   Instruction *PHIUser = dyn_cast<Instruction>(*iu);
     78   if (PHIUser == cast<Instruction>(&EI))
     79     PHIUser = cast<Instruction>(*(++iu));
     80 
     81   // Verify that this PHI user has one use, which is the PHI itself,
     82   // and that it is a binary operation which is cheap to scalarize.
     83   // otherwise return NULL.
     84   if (!PHIUser->hasOneUse() || !(PHIUser->user_back() == PN) ||
     85       !(isa<BinaryOperator>(PHIUser)) || !cheapToScalarize(PHIUser, true))
     86     return nullptr;
     87 
     88   // Create a scalar PHI node that will replace the vector PHI node
     89   // just before the current PHI node.
     90   PHINode *scalarPHI = cast<PHINode>(InsertNewInstWith(
     91       PHINode::Create(EI.getType(), PN->getNumIncomingValues(), ""), *PN));
     92   // Scalarize each PHI operand.
     93   for (unsigned i = 0; i < PN->getNumIncomingValues(); i++) {
     94     Value *PHIInVal = PN->getIncomingValue(i);
     95     BasicBlock *inBB = PN->getIncomingBlock(i);
     96     Value *Elt = EI.getIndexOperand();
     97     // If the operand is the PHI induction variable:
     98     if (PHIInVal == PHIUser) {
     99       // Scalarize the binary operation. Its first operand is the
    100       // scalar PHI, and the second operand is extracted from the other
    101       // vector operand.
    102       BinaryOperator *B0 = cast<BinaryOperator>(PHIUser);
    103       unsigned opId = (B0->getOperand(0) == PN) ? 1 : 0;
    104       Value *Op = InsertNewInstWith(
    105           ExtractElementInst::Create(B0->getOperand(opId), Elt,
    106                                      B0->getOperand(opId)->getName() + ".Elt"),
    107           *B0);
    108       Value *newPHIUser = InsertNewInstWith(
    109           BinaryOperator::Create(B0->getOpcode(), scalarPHI, Op), *B0);
    110       scalarPHI->addIncoming(newPHIUser, inBB);
    111     } else {
    112       // Scalarize PHI input:
    113       Instruction *newEI = ExtractElementInst::Create(PHIInVal, Elt, "");
    114       // Insert the new instruction into the predecessor basic block.
    115       Instruction *pos = dyn_cast<Instruction>(PHIInVal);
    116       BasicBlock::iterator InsertPos;
    117       if (pos && !isa<PHINode>(pos)) {
    118         InsertPos = ++pos->getIterator();
    119       } else {
    120         InsertPos = inBB->getFirstInsertionPt();
    121       }
    122 
    123       InsertNewInstWith(newEI, *InsertPos);
    124 
    125       scalarPHI->addIncoming(newEI, inBB);
    126     }
    127   }
    128   return ReplaceInstUsesWith(EI, scalarPHI);
    129 }
    130 
    131 Instruction *InstCombiner::visitExtractElementInst(ExtractElementInst &EI) {
    132   if (Value *V = SimplifyExtractElementInst(
    133           EI.getVectorOperand(), EI.getIndexOperand(), DL, TLI, DT, AC))
    134     return ReplaceInstUsesWith(EI, V);
    135 
    136   // If vector val is constant with all elements the same, replace EI with
    137   // that element.  We handle a known element # below.
    138   if (Constant *C = dyn_cast<Constant>(EI.getOperand(0)))
    139     if (cheapToScalarize(C, false))
    140       return ReplaceInstUsesWith(EI, C->getAggregateElement(0U));
    141 
    142   // If extracting a specified index from the vector, see if we can recursively
    143   // find a previously computed scalar that was inserted into the vector.
    144   if (ConstantInt *IdxC = dyn_cast<ConstantInt>(EI.getOperand(1))) {
    145     unsigned IndexVal = IdxC->getZExtValue();
    146     unsigned VectorWidth = EI.getVectorOperandType()->getNumElements();
    147 
    148     // InstSimplify handles cases where the index is invalid.
    149     assert(IndexVal < VectorWidth);
    150 
    151     // This instruction only demands the single element from the input vector.
    152     // If the input vector has a single use, simplify it based on this use
    153     // property.
    154     if (EI.getOperand(0)->hasOneUse() && VectorWidth != 1) {
    155       APInt UndefElts(VectorWidth, 0);
    156       APInt DemandedMask(VectorWidth, 0);
    157       DemandedMask.setBit(IndexVal);
    158       if (Value *V = SimplifyDemandedVectorElts(EI.getOperand(0), DemandedMask,
    159                                                 UndefElts)) {
    160         EI.setOperand(0, V);
    161         return &EI;
    162       }
    163     }
    164 
    165     // If this extractelement is directly using a bitcast from a vector of
    166     // the same number of elements, see if we can find the source element from
    167     // it.  In this case, we will end up needing to bitcast the scalars.
    168     if (BitCastInst *BCI = dyn_cast<BitCastInst>(EI.getOperand(0))) {
    169       if (VectorType *VT = dyn_cast<VectorType>(BCI->getOperand(0)->getType()))
    170         if (VT->getNumElements() == VectorWidth)
    171           if (Value *Elt = findScalarElement(BCI->getOperand(0), IndexVal))
    172             return new BitCastInst(Elt, EI.getType());
    173     }
    174 
    175     // If there's a vector PHI feeding a scalar use through this extractelement
    176     // instruction, try to scalarize the PHI.
    177     if (PHINode *PN = dyn_cast<PHINode>(EI.getOperand(0))) {
    178       Instruction *scalarPHI = scalarizePHI(EI, PN);
    179       if (scalarPHI)
    180         return scalarPHI;
    181     }
    182   }
    183 
    184   if (Instruction *I = dyn_cast<Instruction>(EI.getOperand(0))) {
    185     // Push extractelement into predecessor operation if legal and
    186     // profitable to do so.
    187     if (BinaryOperator *BO = dyn_cast<BinaryOperator>(I)) {
    188       if (I->hasOneUse() &&
    189           cheapToScalarize(BO, isa<ConstantInt>(EI.getOperand(1)))) {
    190         Value *newEI0 =
    191           Builder->CreateExtractElement(BO->getOperand(0), EI.getOperand(1),
    192                                         EI.getName()+".lhs");
    193         Value *newEI1 =
    194           Builder->CreateExtractElement(BO->getOperand(1), EI.getOperand(1),
    195                                         EI.getName()+".rhs");
    196         return BinaryOperator::Create(BO->getOpcode(), newEI0, newEI1);
    197       }
    198     } else if (InsertElementInst *IE = dyn_cast<InsertElementInst>(I)) {
    199       // Extracting the inserted element?
    200       if (IE->getOperand(2) == EI.getOperand(1))
    201         return ReplaceInstUsesWith(EI, IE->getOperand(1));
    202       // If the inserted and extracted elements are constants, they must not
    203       // be the same value, extract from the pre-inserted value instead.
    204       if (isa<Constant>(IE->getOperand(2)) && isa<Constant>(EI.getOperand(1))) {
    205         Worklist.AddValue(EI.getOperand(0));
    206         EI.setOperand(0, IE->getOperand(0));
    207         return &EI;
    208       }
    209     } else if (ShuffleVectorInst *SVI = dyn_cast<ShuffleVectorInst>(I)) {
    210       // If this is extracting an element from a shufflevector, figure out where
    211       // it came from and extract from the appropriate input element instead.
    212       if (ConstantInt *Elt = dyn_cast<ConstantInt>(EI.getOperand(1))) {
    213         int SrcIdx = SVI->getMaskValue(Elt->getZExtValue());
    214         Value *Src;
    215         unsigned LHSWidth =
    216           SVI->getOperand(0)->getType()->getVectorNumElements();
    217 
    218         if (SrcIdx < 0)
    219           return ReplaceInstUsesWith(EI, UndefValue::get(EI.getType()));
    220         if (SrcIdx < (int)LHSWidth)
    221           Src = SVI->getOperand(0);
    222         else {
    223           SrcIdx -= LHSWidth;
    224           Src = SVI->getOperand(1);
    225         }
    226         Type *Int32Ty = Type::getInt32Ty(EI.getContext());
    227         return ExtractElementInst::Create(Src,
    228                                           ConstantInt::get(Int32Ty,
    229                                                            SrcIdx, false));
    230       }
    231     } else if (CastInst *CI = dyn_cast<CastInst>(I)) {
    232       // Canonicalize extractelement(cast) -> cast(extractelement).
    233       // Bitcasts can change the number of vector elements, and they cost
    234       // nothing.
    235       if (CI->hasOneUse() && (CI->getOpcode() != Instruction::BitCast)) {
    236         Value *EE = Builder->CreateExtractElement(CI->getOperand(0),
    237                                                   EI.getIndexOperand());
    238         Worklist.AddValue(EE);
    239         return CastInst::Create(CI->getOpcode(), EE, EI.getType());
    240       }
    241     } else if (SelectInst *SI = dyn_cast<SelectInst>(I)) {
    242       if (SI->hasOneUse()) {
    243         // TODO: For a select on vectors, it might be useful to do this if it
    244         // has multiple extractelement uses. For vector select, that seems to
    245         // fight the vectorizer.
    246 
    247         // If we are extracting an element from a vector select or a select on
    248         // vectors, create a select on the scalars extracted from the vector
    249         // arguments.
    250         Value *TrueVal = SI->getTrueValue();
    251         Value *FalseVal = SI->getFalseValue();
    252 
    253         Value *Cond = SI->getCondition();
    254         if (Cond->getType()->isVectorTy()) {
    255           Cond = Builder->CreateExtractElement(Cond,
    256                                                EI.getIndexOperand(),
    257                                                Cond->getName() + ".elt");
    258         }
    259 
    260         Value *V1Elem
    261           = Builder->CreateExtractElement(TrueVal,
    262                                           EI.getIndexOperand(),
    263                                           TrueVal->getName() + ".elt");
    264 
    265         Value *V2Elem
    266           = Builder->CreateExtractElement(FalseVal,
    267                                           EI.getIndexOperand(),
    268                                           FalseVal->getName() + ".elt");
    269         return SelectInst::Create(Cond,
    270                                   V1Elem,
    271                                   V2Elem,
    272                                   SI->getName() + ".elt");
    273       }
    274     }
    275   }
    276   return nullptr;
    277 }
    278 
    279 /// If V is a shuffle of values that ONLY returns elements from either LHS or
    280 /// RHS, return the shuffle mask and true. Otherwise, return false.
    281 static bool collectSingleShuffleElements(Value *V, Value *LHS, Value *RHS,
    282                                          SmallVectorImpl<Constant*> &Mask) {
    283   assert(LHS->getType() == RHS->getType() &&
    284          "Invalid CollectSingleShuffleElements");
    285   unsigned NumElts = V->getType()->getVectorNumElements();
    286 
    287   if (isa<UndefValue>(V)) {
    288     Mask.assign(NumElts, UndefValue::get(Type::getInt32Ty(V->getContext())));
    289     return true;
    290   }
    291 
    292   if (V == LHS) {
    293     for (unsigned i = 0; i != NumElts; ++i)
    294       Mask.push_back(ConstantInt::get(Type::getInt32Ty(V->getContext()), i));
    295     return true;
    296   }
    297 
    298   if (V == RHS) {
    299     for (unsigned i = 0; i != NumElts; ++i)
    300       Mask.push_back(ConstantInt::get(Type::getInt32Ty(V->getContext()),
    301                                       i+NumElts));
    302     return true;
    303   }
    304 
    305   if (InsertElementInst *IEI = dyn_cast<InsertElementInst>(V)) {
    306     // If this is an insert of an extract from some other vector, include it.
    307     Value *VecOp    = IEI->getOperand(0);
    308     Value *ScalarOp = IEI->getOperand(1);
    309     Value *IdxOp    = IEI->getOperand(2);
    310 
    311     if (!isa<ConstantInt>(IdxOp))
    312       return false;
    313     unsigned InsertedIdx = cast<ConstantInt>(IdxOp)->getZExtValue();
    314 
    315     if (isa<UndefValue>(ScalarOp)) {  // inserting undef into vector.
    316       // We can handle this if the vector we are inserting into is
    317       // transitively ok.
    318       if (collectSingleShuffleElements(VecOp, LHS, RHS, Mask)) {
    319         // If so, update the mask to reflect the inserted undef.
    320         Mask[InsertedIdx] = UndefValue::get(Type::getInt32Ty(V->getContext()));
    321         return true;
    322       }
    323     } else if (ExtractElementInst *EI = dyn_cast<ExtractElementInst>(ScalarOp)){
    324       if (isa<ConstantInt>(EI->getOperand(1))) {
    325         unsigned ExtractedIdx =
    326         cast<ConstantInt>(EI->getOperand(1))->getZExtValue();
    327         unsigned NumLHSElts = LHS->getType()->getVectorNumElements();
    328 
    329         // This must be extracting from either LHS or RHS.
    330         if (EI->getOperand(0) == LHS || EI->getOperand(0) == RHS) {
    331           // We can handle this if the vector we are inserting into is
    332           // transitively ok.
    333           if (collectSingleShuffleElements(VecOp, LHS, RHS, Mask)) {
    334             // If so, update the mask to reflect the inserted value.
    335             if (EI->getOperand(0) == LHS) {
    336               Mask[InsertedIdx % NumElts] =
    337               ConstantInt::get(Type::getInt32Ty(V->getContext()),
    338                                ExtractedIdx);
    339             } else {
    340               assert(EI->getOperand(0) == RHS);
    341               Mask[InsertedIdx % NumElts] =
    342               ConstantInt::get(Type::getInt32Ty(V->getContext()),
    343                                ExtractedIdx + NumLHSElts);
    344             }
    345             return true;
    346           }
    347         }
    348       }
    349     }
    350   }
    351 
    352   return false;
    353 }
    354 
    355 
    356 /// We are building a shuffle to create V, which is a sequence of insertelement,
    357 /// extractelement pairs. If PermittedRHS is set, then we must either use it or
    358 /// not rely on the second vector source. Return a std::pair containing the
    359 /// left and right vectors of the proposed shuffle (or 0), and set the Mask
    360 /// parameter as required.
    361 ///
    362 /// Note: we intentionally don't try to fold earlier shuffles since they have
    363 /// often been chosen carefully to be efficiently implementable on the target.
    364 typedef std::pair<Value *, Value *> ShuffleOps;
    365 
    366 static ShuffleOps collectShuffleElements(Value *V,
    367                                          SmallVectorImpl<Constant *> &Mask,
    368                                          Value *PermittedRHS) {
    369   assert(V->getType()->isVectorTy() && "Invalid shuffle!");
    370   unsigned NumElts = cast<VectorType>(V->getType())->getNumElements();
    371 
    372   if (isa<UndefValue>(V)) {
    373     Mask.assign(NumElts, UndefValue::get(Type::getInt32Ty(V->getContext())));
    374     return std::make_pair(
    375         PermittedRHS ? UndefValue::get(PermittedRHS->getType()) : V, nullptr);
    376   }
    377 
    378   if (isa<ConstantAggregateZero>(V)) {
    379     Mask.assign(NumElts, ConstantInt::get(Type::getInt32Ty(V->getContext()),0));
    380     return std::make_pair(V, nullptr);
    381   }
    382 
    383   if (InsertElementInst *IEI = dyn_cast<InsertElementInst>(V)) {
    384     // If this is an insert of an extract from some other vector, include it.
    385     Value *VecOp    = IEI->getOperand(0);
    386     Value *ScalarOp = IEI->getOperand(1);
    387     Value *IdxOp    = IEI->getOperand(2);
    388 
    389     if (ExtractElementInst *EI = dyn_cast<ExtractElementInst>(ScalarOp)) {
    390       if (isa<ConstantInt>(EI->getOperand(1)) && isa<ConstantInt>(IdxOp)) {
    391         unsigned ExtractedIdx =
    392           cast<ConstantInt>(EI->getOperand(1))->getZExtValue();
    393         unsigned InsertedIdx = cast<ConstantInt>(IdxOp)->getZExtValue();
    394 
    395         // Either the extracted from or inserted into vector must be RHSVec,
    396         // otherwise we'd end up with a shuffle of three inputs.
    397         if (EI->getOperand(0) == PermittedRHS || PermittedRHS == nullptr) {
    398           Value *RHS = EI->getOperand(0);
    399           ShuffleOps LR = collectShuffleElements(VecOp, Mask, RHS);
    400           assert(LR.second == nullptr || LR.second == RHS);
    401 
    402           if (LR.first->getType() != RHS->getType()) {
    403             // We tried our best, but we can't find anything compatible with RHS
    404             // further up the chain. Return a trivial shuffle.
    405             for (unsigned i = 0; i < NumElts; ++i)
    406               Mask[i] = ConstantInt::get(Type::getInt32Ty(V->getContext()), i);
    407             return std::make_pair(V, nullptr);
    408           }
    409 
    410           unsigned NumLHSElts = RHS->getType()->getVectorNumElements();
    411           Mask[InsertedIdx % NumElts] =
    412             ConstantInt::get(Type::getInt32Ty(V->getContext()),
    413                              NumLHSElts+ExtractedIdx);
    414           return std::make_pair(LR.first, RHS);
    415         }
    416 
    417         if (VecOp == PermittedRHS) {
    418           // We've gone as far as we can: anything on the other side of the
    419           // extractelement will already have been converted into a shuffle.
    420           unsigned NumLHSElts =
    421               EI->getOperand(0)->getType()->getVectorNumElements();
    422           for (unsigned i = 0; i != NumElts; ++i)
    423             Mask.push_back(ConstantInt::get(
    424                 Type::getInt32Ty(V->getContext()),
    425                 i == InsertedIdx ? ExtractedIdx : NumLHSElts + i));
    426           return std::make_pair(EI->getOperand(0), PermittedRHS);
    427         }
    428 
    429         // If this insertelement is a chain that comes from exactly these two
    430         // vectors, return the vector and the effective shuffle.
    431         if (EI->getOperand(0)->getType() == PermittedRHS->getType() &&
    432             collectSingleShuffleElements(IEI, EI->getOperand(0), PermittedRHS,
    433                                          Mask))
    434           return std::make_pair(EI->getOperand(0), PermittedRHS);
    435       }
    436     }
    437   }
    438 
    439   // Otherwise, we can't do anything fancy. Return an identity vector.
    440   for (unsigned i = 0; i != NumElts; ++i)
    441     Mask.push_back(ConstantInt::get(Type::getInt32Ty(V->getContext()), i));
    442   return std::make_pair(V, nullptr);
    443 }
    444 
    445 /// Try to find redundant insertvalue instructions, like the following ones:
    446 ///  %0 = insertvalue { i8, i32 } undef, i8 %x, 0
    447 ///  %1 = insertvalue { i8, i32 } %0,    i8 %y, 0
    448 /// Here the second instruction inserts values at the same indices, as the
    449 /// first one, making the first one redundant.
    450 /// It should be transformed to:
    451 ///  %0 = insertvalue { i8, i32 } undef, i8 %y, 0
    452 Instruction *InstCombiner::visitInsertValueInst(InsertValueInst &I) {
    453   bool IsRedundant = false;
    454   ArrayRef<unsigned int> FirstIndices = I.getIndices();
    455 
    456   // If there is a chain of insertvalue instructions (each of them except the
    457   // last one has only one use and it's another insertvalue insn from this
    458   // chain), check if any of the 'children' uses the same indices as the first
    459   // instruction. In this case, the first one is redundant.
    460   Value *V = &I;
    461   unsigned Depth = 0;
    462   while (V->hasOneUse() && Depth < 10) {
    463     User *U = V->user_back();
    464     auto UserInsInst = dyn_cast<InsertValueInst>(U);
    465     if (!UserInsInst || U->getOperand(0) != V)
    466       break;
    467     if (UserInsInst->getIndices() == FirstIndices) {
    468       IsRedundant = true;
    469       break;
    470     }
    471     V = UserInsInst;
    472     Depth++;
    473   }
    474 
    475   if (IsRedundant)
    476     return ReplaceInstUsesWith(I, I.getOperand(0));
    477   return nullptr;
    478 }
    479 
    480 Instruction *InstCombiner::visitInsertElementInst(InsertElementInst &IE) {
    481   Value *VecOp    = IE.getOperand(0);
    482   Value *ScalarOp = IE.getOperand(1);
    483   Value *IdxOp    = IE.getOperand(2);
    484 
    485   // Inserting an undef or into an undefined place, remove this.
    486   if (isa<UndefValue>(ScalarOp) || isa<UndefValue>(IdxOp))
    487     ReplaceInstUsesWith(IE, VecOp);
    488 
    489   // If the inserted element was extracted from some other vector, and if the
    490   // indexes are constant, try to turn this into a shufflevector operation.
    491   if (ExtractElementInst *EI = dyn_cast<ExtractElementInst>(ScalarOp)) {
    492     if (isa<ConstantInt>(EI->getOperand(1)) && isa<ConstantInt>(IdxOp)) {
    493       unsigned NumInsertVectorElts = IE.getType()->getNumElements();
    494       unsigned NumExtractVectorElts =
    495           EI->getOperand(0)->getType()->getVectorNumElements();
    496       unsigned ExtractedIdx =
    497         cast<ConstantInt>(EI->getOperand(1))->getZExtValue();
    498       unsigned InsertedIdx = cast<ConstantInt>(IdxOp)->getZExtValue();
    499 
    500       if (ExtractedIdx >= NumExtractVectorElts) // Out of range extract.
    501         return ReplaceInstUsesWith(IE, VecOp);
    502 
    503       if (InsertedIdx >= NumInsertVectorElts)  // Out of range insert.
    504         return ReplaceInstUsesWith(IE, UndefValue::get(IE.getType()));
    505 
    506       // If we are extracting a value from a vector, then inserting it right
    507       // back into the same place, just use the input vector.
    508       if (EI->getOperand(0) == VecOp && ExtractedIdx == InsertedIdx)
    509         return ReplaceInstUsesWith(IE, VecOp);
    510 
    511       // If this insertelement isn't used by some other insertelement, turn it
    512       // (and any insertelements it points to), into one big shuffle.
    513       if (!IE.hasOneUse() || !isa<InsertElementInst>(IE.user_back())) {
    514         SmallVector<Constant*, 16> Mask;
    515         ShuffleOps LR = collectShuffleElements(&IE, Mask, nullptr);
    516 
    517         // The proposed shuffle may be trivial, in which case we shouldn't
    518         // perform the combine.
    519         if (LR.first != &IE && LR.second != &IE) {
    520           // We now have a shuffle of LHS, RHS, Mask.
    521           if (LR.second == nullptr)
    522             LR.second = UndefValue::get(LR.first->getType());
    523           return new ShuffleVectorInst(LR.first, LR.second,
    524                                        ConstantVector::get(Mask));
    525         }
    526       }
    527     }
    528   }
    529 
    530   unsigned VWidth = cast<VectorType>(VecOp->getType())->getNumElements();
    531   APInt UndefElts(VWidth, 0);
    532   APInt AllOnesEltMask(APInt::getAllOnesValue(VWidth));
    533   if (Value *V = SimplifyDemandedVectorElts(&IE, AllOnesEltMask, UndefElts)) {
    534     if (V != &IE)
    535       return ReplaceInstUsesWith(IE, V);
    536     return &IE;
    537   }
    538 
    539   return nullptr;
    540 }
    541 
    542 /// Return true if we can evaluate the specified expression tree if the vector
    543 /// elements were shuffled in a different order.
    544 static bool CanEvaluateShuffled(Value *V, ArrayRef<int> Mask,
    545                                 unsigned Depth = 5) {
    546   // We can always reorder the elements of a constant.
    547   if (isa<Constant>(V))
    548     return true;
    549 
    550   // We won't reorder vector arguments. No IPO here.
    551   Instruction *I = dyn_cast<Instruction>(V);
    552   if (!I) return false;
    553 
    554   // Two users may expect different orders of the elements. Don't try it.
    555   if (!I->hasOneUse())
    556     return false;
    557 
    558   if (Depth == 0) return false;
    559 
    560   switch (I->getOpcode()) {
    561     case Instruction::Add:
    562     case Instruction::FAdd:
    563     case Instruction::Sub:
    564     case Instruction::FSub:
    565     case Instruction::Mul:
    566     case Instruction::FMul:
    567     case Instruction::UDiv:
    568     case Instruction::SDiv:
    569     case Instruction::FDiv:
    570     case Instruction::URem:
    571     case Instruction::SRem:
    572     case Instruction::FRem:
    573     case Instruction::Shl:
    574     case Instruction::LShr:
    575     case Instruction::AShr:
    576     case Instruction::And:
    577     case Instruction::Or:
    578     case Instruction::Xor:
    579     case Instruction::ICmp:
    580     case Instruction::FCmp:
    581     case Instruction::Trunc:
    582     case Instruction::ZExt:
    583     case Instruction::SExt:
    584     case Instruction::FPToUI:
    585     case Instruction::FPToSI:
    586     case Instruction::UIToFP:
    587     case Instruction::SIToFP:
    588     case Instruction::FPTrunc:
    589     case Instruction::FPExt:
    590     case Instruction::GetElementPtr: {
    591       for (Value *Operand : I->operands()) {
    592         if (!CanEvaluateShuffled(Operand, Mask, Depth-1))
    593           return false;
    594       }
    595       return true;
    596     }
    597     case Instruction::InsertElement: {
    598       ConstantInt *CI = dyn_cast<ConstantInt>(I->getOperand(2));
    599       if (!CI) return false;
    600       int ElementNumber = CI->getLimitedValue();
    601 
    602       // Verify that 'CI' does not occur twice in Mask. A single 'insertelement'
    603       // can't put an element into multiple indices.
    604       bool SeenOnce = false;
    605       for (int i = 0, e = Mask.size(); i != e; ++i) {
    606         if (Mask[i] == ElementNumber) {
    607           if (SeenOnce)
    608             return false;
    609           SeenOnce = true;
    610         }
    611       }
    612       return CanEvaluateShuffled(I->getOperand(0), Mask, Depth-1);
    613     }
    614   }
    615   return false;
    616 }
    617 
    618 /// Rebuild a new instruction just like 'I' but with the new operands given.
    619 /// In the event of type mismatch, the type of the operands is correct.
    620 static Value *buildNew(Instruction *I, ArrayRef<Value*> NewOps) {
    621   // We don't want to use the IRBuilder here because we want the replacement
    622   // instructions to appear next to 'I', not the builder's insertion point.
    623   switch (I->getOpcode()) {
    624     case Instruction::Add:
    625     case Instruction::FAdd:
    626     case Instruction::Sub:
    627     case Instruction::FSub:
    628     case Instruction::Mul:
    629     case Instruction::FMul:
    630     case Instruction::UDiv:
    631     case Instruction::SDiv:
    632     case Instruction::FDiv:
    633     case Instruction::URem:
    634     case Instruction::SRem:
    635     case Instruction::FRem:
    636     case Instruction::Shl:
    637     case Instruction::LShr:
    638     case Instruction::AShr:
    639     case Instruction::And:
    640     case Instruction::Or:
    641     case Instruction::Xor: {
    642       BinaryOperator *BO = cast<BinaryOperator>(I);
    643       assert(NewOps.size() == 2 && "binary operator with #ops != 2");
    644       BinaryOperator *New =
    645           BinaryOperator::Create(cast<BinaryOperator>(I)->getOpcode(),
    646                                  NewOps[0], NewOps[1], "", BO);
    647       if (isa<OverflowingBinaryOperator>(BO)) {
    648         New->setHasNoUnsignedWrap(BO->hasNoUnsignedWrap());
    649         New->setHasNoSignedWrap(BO->hasNoSignedWrap());
    650       }
    651       if (isa<PossiblyExactOperator>(BO)) {
    652         New->setIsExact(BO->isExact());
    653       }
    654       if (isa<FPMathOperator>(BO))
    655         New->copyFastMathFlags(I);
    656       return New;
    657     }
    658     case Instruction::ICmp:
    659       assert(NewOps.size() == 2 && "icmp with #ops != 2");
    660       return new ICmpInst(I, cast<ICmpInst>(I)->getPredicate(),
    661                           NewOps[0], NewOps[1]);
    662     case Instruction::FCmp:
    663       assert(NewOps.size() == 2 && "fcmp with #ops != 2");
    664       return new FCmpInst(I, cast<FCmpInst>(I)->getPredicate(),
    665                           NewOps[0], NewOps[1]);
    666     case Instruction::Trunc:
    667     case Instruction::ZExt:
    668     case Instruction::SExt:
    669     case Instruction::FPToUI:
    670     case Instruction::FPToSI:
    671     case Instruction::UIToFP:
    672     case Instruction::SIToFP:
    673     case Instruction::FPTrunc:
    674     case Instruction::FPExt: {
    675       // It's possible that the mask has a different number of elements from
    676       // the original cast. We recompute the destination type to match the mask.
    677       Type *DestTy =
    678           VectorType::get(I->getType()->getScalarType(),
    679                           NewOps[0]->getType()->getVectorNumElements());
    680       assert(NewOps.size() == 1 && "cast with #ops != 1");
    681       return CastInst::Create(cast<CastInst>(I)->getOpcode(), NewOps[0], DestTy,
    682                               "", I);
    683     }
    684     case Instruction::GetElementPtr: {
    685       Value *Ptr = NewOps[0];
    686       ArrayRef<Value*> Idx = NewOps.slice(1);
    687       GetElementPtrInst *GEP = GetElementPtrInst::Create(
    688           cast<GetElementPtrInst>(I)->getSourceElementType(), Ptr, Idx, "", I);
    689       GEP->setIsInBounds(cast<GetElementPtrInst>(I)->isInBounds());
    690       return GEP;
    691     }
    692   }
    693   llvm_unreachable("failed to rebuild vector instructions");
    694 }
    695 
    696 Value *
    697 InstCombiner::EvaluateInDifferentElementOrder(Value *V, ArrayRef<int> Mask) {
    698   // Mask.size() does not need to be equal to the number of vector elements.
    699 
    700   assert(V->getType()->isVectorTy() && "can't reorder non-vector elements");
    701   if (isa<UndefValue>(V)) {
    702     return UndefValue::get(VectorType::get(V->getType()->getScalarType(),
    703                                            Mask.size()));
    704   }
    705   if (isa<ConstantAggregateZero>(V)) {
    706     return ConstantAggregateZero::get(
    707                VectorType::get(V->getType()->getScalarType(),
    708                                Mask.size()));
    709   }
    710   if (Constant *C = dyn_cast<Constant>(V)) {
    711     SmallVector<Constant *, 16> MaskValues;
    712     for (int i = 0, e = Mask.size(); i != e; ++i) {
    713       if (Mask[i] == -1)
    714         MaskValues.push_back(UndefValue::get(Builder->getInt32Ty()));
    715       else
    716         MaskValues.push_back(Builder->getInt32(Mask[i]));
    717     }
    718     return ConstantExpr::getShuffleVector(C, UndefValue::get(C->getType()),
    719                                           ConstantVector::get(MaskValues));
    720   }
    721 
    722   Instruction *I = cast<Instruction>(V);
    723   switch (I->getOpcode()) {
    724     case Instruction::Add:
    725     case Instruction::FAdd:
    726     case Instruction::Sub:
    727     case Instruction::FSub:
    728     case Instruction::Mul:
    729     case Instruction::FMul:
    730     case Instruction::UDiv:
    731     case Instruction::SDiv:
    732     case Instruction::FDiv:
    733     case Instruction::URem:
    734     case Instruction::SRem:
    735     case Instruction::FRem:
    736     case Instruction::Shl:
    737     case Instruction::LShr:
    738     case Instruction::AShr:
    739     case Instruction::And:
    740     case Instruction::Or:
    741     case Instruction::Xor:
    742     case Instruction::ICmp:
    743     case Instruction::FCmp:
    744     case Instruction::Trunc:
    745     case Instruction::ZExt:
    746     case Instruction::SExt:
    747     case Instruction::FPToUI:
    748     case Instruction::FPToSI:
    749     case Instruction::UIToFP:
    750     case Instruction::SIToFP:
    751     case Instruction::FPTrunc:
    752     case Instruction::FPExt:
    753     case Instruction::Select:
    754     case Instruction::GetElementPtr: {
    755       SmallVector<Value*, 8> NewOps;
    756       bool NeedsRebuild = (Mask.size() != I->getType()->getVectorNumElements());
    757       for (int i = 0, e = I->getNumOperands(); i != e; ++i) {
    758         Value *V = EvaluateInDifferentElementOrder(I->getOperand(i), Mask);
    759         NewOps.push_back(V);
    760         NeedsRebuild |= (V != I->getOperand(i));
    761       }
    762       if (NeedsRebuild) {
    763         return buildNew(I, NewOps);
    764       }
    765       return I;
    766     }
    767     case Instruction::InsertElement: {
    768       int Element = cast<ConstantInt>(I->getOperand(2))->getLimitedValue();
    769 
    770       // The insertelement was inserting at Element. Figure out which element
    771       // that becomes after shuffling. The answer is guaranteed to be unique
    772       // by CanEvaluateShuffled.
    773       bool Found = false;
    774       int Index = 0;
    775       for (int e = Mask.size(); Index != e; ++Index) {
    776         if (Mask[Index] == Element) {
    777           Found = true;
    778           break;
    779         }
    780       }
    781 
    782       // If element is not in Mask, no need to handle the operand 1 (element to
    783       // be inserted). Just evaluate values in operand 0 according to Mask.
    784       if (!Found)
    785         return EvaluateInDifferentElementOrder(I->getOperand(0), Mask);
    786 
    787       Value *V = EvaluateInDifferentElementOrder(I->getOperand(0), Mask);
    788       return InsertElementInst::Create(V, I->getOperand(1),
    789                                        Builder->getInt32(Index), "", I);
    790     }
    791   }
    792   llvm_unreachable("failed to reorder elements of vector instruction!");
    793 }
    794 
    795 static void recognizeIdentityMask(const SmallVectorImpl<int> &Mask,
    796                                   bool &isLHSID, bool &isRHSID) {
    797   isLHSID = isRHSID = true;
    798 
    799   for (unsigned i = 0, e = Mask.size(); i != e; ++i) {
    800     if (Mask[i] < 0) continue;  // Ignore undef values.
    801     // Is this an identity shuffle of the LHS value?
    802     isLHSID &= (Mask[i] == (int)i);
    803 
    804     // Is this an identity shuffle of the RHS value?
    805     isRHSID &= (Mask[i]-e == i);
    806   }
    807 }
    808 
    809 // Returns true if the shuffle is extracting a contiguous range of values from
    810 // LHS, for example:
    811 //                 +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
    812 //   Input:        |AA|BB|CC|DD|EE|FF|GG|HH|II|JJ|KK|LL|MM|NN|OO|PP|
    813 //   Shuffles to:  |EE|FF|GG|HH|
    814 //                 +--+--+--+--+
    815 static bool isShuffleExtractingFromLHS(ShuffleVectorInst &SVI,
    816                                        SmallVector<int, 16> &Mask) {
    817   unsigned LHSElems =
    818       cast<VectorType>(SVI.getOperand(0)->getType())->getNumElements();
    819   unsigned MaskElems = Mask.size();
    820   unsigned BegIdx = Mask.front();
    821   unsigned EndIdx = Mask.back();
    822   if (BegIdx > EndIdx || EndIdx >= LHSElems || EndIdx - BegIdx != MaskElems - 1)
    823     return false;
    824   for (unsigned I = 0; I != MaskElems; ++I)
    825     if (static_cast<unsigned>(Mask[I]) != BegIdx + I)
    826       return false;
    827   return true;
    828 }
    829 
    830 Instruction *InstCombiner::visitShuffleVectorInst(ShuffleVectorInst &SVI) {
    831   Value *LHS = SVI.getOperand(0);
    832   Value *RHS = SVI.getOperand(1);
    833   SmallVector<int, 16> Mask = SVI.getShuffleMask();
    834   Type *Int32Ty = Type::getInt32Ty(SVI.getContext());
    835 
    836   bool MadeChange = false;
    837 
    838   // Undefined shuffle mask -> undefined value.
    839   if (isa<UndefValue>(SVI.getOperand(2)))
    840     return ReplaceInstUsesWith(SVI, UndefValue::get(SVI.getType()));
    841 
    842   unsigned VWidth = cast<VectorType>(SVI.getType())->getNumElements();
    843 
    844   APInt UndefElts(VWidth, 0);
    845   APInt AllOnesEltMask(APInt::getAllOnesValue(VWidth));
    846   if (Value *V = SimplifyDemandedVectorElts(&SVI, AllOnesEltMask, UndefElts)) {
    847     if (V != &SVI)
    848       return ReplaceInstUsesWith(SVI, V);
    849     LHS = SVI.getOperand(0);
    850     RHS = SVI.getOperand(1);
    851     MadeChange = true;
    852   }
    853 
    854   unsigned LHSWidth = cast<VectorType>(LHS->getType())->getNumElements();
    855 
    856   // Canonicalize shuffle(x    ,x,mask) -> shuffle(x, undef,mask')
    857   // Canonicalize shuffle(undef,x,mask) -> shuffle(x, undef,mask').
    858   if (LHS == RHS || isa<UndefValue>(LHS)) {
    859     if (isa<UndefValue>(LHS) && LHS == RHS) {
    860       // shuffle(undef,undef,mask) -> undef.
    861       Value *Result = (VWidth == LHSWidth)
    862                       ? LHS : UndefValue::get(SVI.getType());
    863       return ReplaceInstUsesWith(SVI, Result);
    864     }
    865 
    866     // Remap any references to RHS to use LHS.
    867     SmallVector<Constant*, 16> Elts;
    868     for (unsigned i = 0, e = LHSWidth; i != VWidth; ++i) {
    869       if (Mask[i] < 0) {
    870         Elts.push_back(UndefValue::get(Int32Ty));
    871         continue;
    872       }
    873 
    874       if ((Mask[i] >= (int)e && isa<UndefValue>(RHS)) ||
    875           (Mask[i] <  (int)e && isa<UndefValue>(LHS))) {
    876         Mask[i] = -1;     // Turn into undef.
    877         Elts.push_back(UndefValue::get(Int32Ty));
    878       } else {
    879         Mask[i] = Mask[i] % e;  // Force to LHS.
    880         Elts.push_back(ConstantInt::get(Int32Ty, Mask[i]));
    881       }
    882     }
    883     SVI.setOperand(0, SVI.getOperand(1));
    884     SVI.setOperand(1, UndefValue::get(RHS->getType()));
    885     SVI.setOperand(2, ConstantVector::get(Elts));
    886     LHS = SVI.getOperand(0);
    887     RHS = SVI.getOperand(1);
    888     MadeChange = true;
    889   }
    890 
    891   if (VWidth == LHSWidth) {
    892     // Analyze the shuffle, are the LHS or RHS and identity shuffles?
    893     bool isLHSID, isRHSID;
    894     recognizeIdentityMask(Mask, isLHSID, isRHSID);
    895 
    896     // Eliminate identity shuffles.
    897     if (isLHSID) return ReplaceInstUsesWith(SVI, LHS);
    898     if (isRHSID) return ReplaceInstUsesWith(SVI, RHS);
    899   }
    900 
    901   if (isa<UndefValue>(RHS) && CanEvaluateShuffled(LHS, Mask)) {
    902     Value *V = EvaluateInDifferentElementOrder(LHS, Mask);
    903     return ReplaceInstUsesWith(SVI, V);
    904   }
    905 
    906   // SROA generates shuffle+bitcast when the extracted sub-vector is bitcast to
    907   // a non-vector type. We can instead bitcast the original vector followed by
    908   // an extract of the desired element:
    909   //
    910   //   %sroa = shufflevector <16 x i8> %in, <16 x i8> undef,
    911   //                         <4 x i32> <i32 0, i32 1, i32 2, i32 3>
    912   //   %1 = bitcast <4 x i8> %sroa to i32
    913   // Becomes:
    914   //   %bc = bitcast <16 x i8> %in to <4 x i32>
    915   //   %ext = extractelement <4 x i32> %bc, i32 0
    916   //
    917   // If the shuffle is extracting a contiguous range of values from the input
    918   // vector then each use which is a bitcast of the extracted size can be
    919   // replaced. This will work if the vector types are compatible, and the begin
    920   // index is aligned to a value in the casted vector type. If the begin index
    921   // isn't aligned then we can shuffle the original vector (keeping the same
    922   // vector type) before extracting.
    923   //
    924   // This code will bail out if the target type is fundamentally incompatible
    925   // with vectors of the source type.
    926   //
    927   // Example of <16 x i8>, target type i32:
    928   // Index range [4,8):         v-----------v Will work.
    929   //                +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
    930   //     <16 x i8>: |  |  |  |  |  |  |  |  |  |  |  |  |  |  |  |  |
    931   //     <4 x i32>: |           |           |           |           |
    932   //                +-----------+-----------+-----------+-----------+
    933   // Index range [6,10):              ^-----------^ Needs an extra shuffle.
    934   // Target type i40:           ^--------------^ Won't work, bail.
    935   if (isShuffleExtractingFromLHS(SVI, Mask)) {
    936     Value *V = LHS;
    937     unsigned MaskElems = Mask.size();
    938     unsigned BegIdx = Mask.front();
    939     VectorType *SrcTy = cast<VectorType>(V->getType());
    940     unsigned VecBitWidth = SrcTy->getBitWidth();
    941     unsigned SrcElemBitWidth = DL.getTypeSizeInBits(SrcTy->getElementType());
    942     assert(SrcElemBitWidth && "vector elements must have a bitwidth");
    943     unsigned SrcNumElems = SrcTy->getNumElements();
    944     SmallVector<BitCastInst *, 8> BCs;
    945     DenseMap<Type *, Value *> NewBCs;
    946     for (User *U : SVI.users())
    947       if (BitCastInst *BC = dyn_cast<BitCastInst>(U))
    948         if (!BC->use_empty())
    949           // Only visit bitcasts that weren't previously handled.
    950           BCs.push_back(BC);
    951     for (BitCastInst *BC : BCs) {
    952       Type *TgtTy = BC->getDestTy();
    953       unsigned TgtElemBitWidth = DL.getTypeSizeInBits(TgtTy);
    954       if (!TgtElemBitWidth)
    955         continue;
    956       unsigned TgtNumElems = VecBitWidth / TgtElemBitWidth;
    957       bool VecBitWidthsEqual = VecBitWidth == TgtNumElems * TgtElemBitWidth;
    958       bool BegIsAligned = 0 == ((SrcElemBitWidth * BegIdx) % TgtElemBitWidth);
    959       if (!VecBitWidthsEqual)
    960         continue;
    961       if (!VectorType::isValidElementType(TgtTy))
    962         continue;
    963       VectorType *CastSrcTy = VectorType::get(TgtTy, TgtNumElems);
    964       if (!BegIsAligned) {
    965         // Shuffle the input so [0,NumElements) contains the output, and
    966         // [NumElems,SrcNumElems) is undef.
    967         SmallVector<Constant *, 16> ShuffleMask(SrcNumElems,
    968                                                 UndefValue::get(Int32Ty));
    969         for (unsigned I = 0, E = MaskElems, Idx = BegIdx; I != E; ++Idx, ++I)
    970           ShuffleMask[I] = ConstantInt::get(Int32Ty, Idx);
    971         V = Builder->CreateShuffleVector(V, UndefValue::get(V->getType()),
    972                                          ConstantVector::get(ShuffleMask),
    973                                          SVI.getName() + ".extract");
    974         BegIdx = 0;
    975       }
    976       unsigned SrcElemsPerTgtElem = TgtElemBitWidth / SrcElemBitWidth;
    977       assert(SrcElemsPerTgtElem);
    978       BegIdx /= SrcElemsPerTgtElem;
    979       bool BCAlreadyExists = NewBCs.find(CastSrcTy) != NewBCs.end();
    980       auto *NewBC =
    981           BCAlreadyExists
    982               ? NewBCs[CastSrcTy]
    983               : Builder->CreateBitCast(V, CastSrcTy, SVI.getName() + ".bc");
    984       if (!BCAlreadyExists)
    985         NewBCs[CastSrcTy] = NewBC;
    986       auto *Ext = Builder->CreateExtractElement(
    987           NewBC, ConstantInt::get(Int32Ty, BegIdx), SVI.getName() + ".extract");
    988       // The shufflevector isn't being replaced: the bitcast that used it
    989       // is. InstCombine will visit the newly-created instructions.
    990       ReplaceInstUsesWith(*BC, Ext);
    991       MadeChange = true;
    992     }
    993   }
    994 
    995   // If the LHS is a shufflevector itself, see if we can combine it with this
    996   // one without producing an unusual shuffle.
    997   // Cases that might be simplified:
    998   // 1.
    999   // x1=shuffle(v1,v2,mask1)
   1000   //  x=shuffle(x1,undef,mask)
   1001   //        ==>
   1002   //  x=shuffle(v1,undef,newMask)
   1003   // newMask[i] = (mask[i] < x1.size()) ? mask1[mask[i]] : -1
   1004   // 2.
   1005   // x1=shuffle(v1,undef,mask1)
   1006   //  x=shuffle(x1,x2,mask)
   1007   // where v1.size() == mask1.size()
   1008   //        ==>
   1009   //  x=shuffle(v1,x2,newMask)
   1010   // newMask[i] = (mask[i] < x1.size()) ? mask1[mask[i]] : mask[i]
   1011   // 3.
   1012   // x2=shuffle(v2,undef,mask2)
   1013   //  x=shuffle(x1,x2,mask)
   1014   // where v2.size() == mask2.size()
   1015   //        ==>
   1016   //  x=shuffle(x1,v2,newMask)
   1017   // newMask[i] = (mask[i] < x1.size())
   1018   //              ? mask[i] : mask2[mask[i]-x1.size()]+x1.size()
   1019   // 4.
   1020   // x1=shuffle(v1,undef,mask1)
   1021   // x2=shuffle(v2,undef,mask2)
   1022   //  x=shuffle(x1,x2,mask)
   1023   // where v1.size() == v2.size()
   1024   //        ==>
   1025   //  x=shuffle(v1,v2,newMask)
   1026   // newMask[i] = (mask[i] < x1.size())
   1027   //              ? mask1[mask[i]] : mask2[mask[i]-x1.size()]+v1.size()
   1028   //
   1029   // Here we are really conservative:
   1030   // we are absolutely afraid of producing a shuffle mask not in the input
   1031   // program, because the code gen may not be smart enough to turn a merged
   1032   // shuffle into two specific shuffles: it may produce worse code.  As such,
   1033   // we only merge two shuffles if the result is either a splat or one of the
   1034   // input shuffle masks.  In this case, merging the shuffles just removes
   1035   // one instruction, which we know is safe.  This is good for things like
   1036   // turning: (splat(splat)) -> splat, or
   1037   // merge(V[0..n], V[n+1..2n]) -> V[0..2n]
   1038   ShuffleVectorInst* LHSShuffle = dyn_cast<ShuffleVectorInst>(LHS);
   1039   ShuffleVectorInst* RHSShuffle = dyn_cast<ShuffleVectorInst>(RHS);
   1040   if (LHSShuffle)
   1041     if (!isa<UndefValue>(LHSShuffle->getOperand(1)) && !isa<UndefValue>(RHS))
   1042       LHSShuffle = nullptr;
   1043   if (RHSShuffle)
   1044     if (!isa<UndefValue>(RHSShuffle->getOperand(1)))
   1045       RHSShuffle = nullptr;
   1046   if (!LHSShuffle && !RHSShuffle)
   1047     return MadeChange ? &SVI : nullptr;
   1048 
   1049   Value* LHSOp0 = nullptr;
   1050   Value* LHSOp1 = nullptr;
   1051   Value* RHSOp0 = nullptr;
   1052   unsigned LHSOp0Width = 0;
   1053   unsigned RHSOp0Width = 0;
   1054   if (LHSShuffle) {
   1055     LHSOp0 = LHSShuffle->getOperand(0);
   1056     LHSOp1 = LHSShuffle->getOperand(1);
   1057     LHSOp0Width = cast<VectorType>(LHSOp0->getType())->getNumElements();
   1058   }
   1059   if (RHSShuffle) {
   1060     RHSOp0 = RHSShuffle->getOperand(0);
   1061     RHSOp0Width = cast<VectorType>(RHSOp0->getType())->getNumElements();
   1062   }
   1063   Value* newLHS = LHS;
   1064   Value* newRHS = RHS;
   1065   if (LHSShuffle) {
   1066     // case 1
   1067     if (isa<UndefValue>(RHS)) {
   1068       newLHS = LHSOp0;
   1069       newRHS = LHSOp1;
   1070     }
   1071     // case 2 or 4
   1072     else if (LHSOp0Width == LHSWidth) {
   1073       newLHS = LHSOp0;
   1074     }
   1075   }
   1076   // case 3 or 4
   1077   if (RHSShuffle && RHSOp0Width == LHSWidth) {
   1078     newRHS = RHSOp0;
   1079   }
   1080   // case 4
   1081   if (LHSOp0 == RHSOp0) {
   1082     newLHS = LHSOp0;
   1083     newRHS = nullptr;
   1084   }
   1085 
   1086   if (newLHS == LHS && newRHS == RHS)
   1087     return MadeChange ? &SVI : nullptr;
   1088 
   1089   SmallVector<int, 16> LHSMask;
   1090   SmallVector<int, 16> RHSMask;
   1091   if (newLHS != LHS)
   1092     LHSMask = LHSShuffle->getShuffleMask();
   1093   if (RHSShuffle && newRHS != RHS)
   1094     RHSMask = RHSShuffle->getShuffleMask();
   1095 
   1096   unsigned newLHSWidth = (newLHS != LHS) ? LHSOp0Width : LHSWidth;
   1097   SmallVector<int, 16> newMask;
   1098   bool isSplat = true;
   1099   int SplatElt = -1;
   1100   // Create a new mask for the new ShuffleVectorInst so that the new
   1101   // ShuffleVectorInst is equivalent to the original one.
   1102   for (unsigned i = 0; i < VWidth; ++i) {
   1103     int eltMask;
   1104     if (Mask[i] < 0) {
   1105       // This element is an undef value.
   1106       eltMask = -1;
   1107     } else if (Mask[i] < (int)LHSWidth) {
   1108       // This element is from left hand side vector operand.
   1109       //
   1110       // If LHS is going to be replaced (case 1, 2, or 4), calculate the
   1111       // new mask value for the element.
   1112       if (newLHS != LHS) {
   1113         eltMask = LHSMask[Mask[i]];
   1114         // If the value selected is an undef value, explicitly specify it
   1115         // with a -1 mask value.
   1116         if (eltMask >= (int)LHSOp0Width && isa<UndefValue>(LHSOp1))
   1117           eltMask = -1;
   1118       } else
   1119         eltMask = Mask[i];
   1120     } else {
   1121       // This element is from right hand side vector operand
   1122       //
   1123       // If the value selected is an undef value, explicitly specify it
   1124       // with a -1 mask value. (case 1)
   1125       if (isa<UndefValue>(RHS))
   1126         eltMask = -1;
   1127       // If RHS is going to be replaced (case 3 or 4), calculate the
   1128       // new mask value for the element.
   1129       else if (newRHS != RHS) {
   1130         eltMask = RHSMask[Mask[i]-LHSWidth];
   1131         // If the value selected is an undef value, explicitly specify it
   1132         // with a -1 mask value.
   1133         if (eltMask >= (int)RHSOp0Width) {
   1134           assert(isa<UndefValue>(RHSShuffle->getOperand(1))
   1135                  && "should have been check above");
   1136           eltMask = -1;
   1137         }
   1138       } else
   1139         eltMask = Mask[i]-LHSWidth;
   1140 
   1141       // If LHS's width is changed, shift the mask value accordingly.
   1142       // If newRHS == NULL, i.e. LHSOp0 == RHSOp0, we want to remap any
   1143       // references from RHSOp0 to LHSOp0, so we don't need to shift the mask.
   1144       // If newRHS == newLHS, we want to remap any references from newRHS to
   1145       // newLHS so that we can properly identify splats that may occur due to
   1146       // obfuscation across the two vectors.
   1147       if (eltMask >= 0 && newRHS != nullptr && newLHS != newRHS)
   1148         eltMask += newLHSWidth;
   1149     }
   1150 
   1151     // Check if this could still be a splat.
   1152     if (eltMask >= 0) {
   1153       if (SplatElt >= 0 && SplatElt != eltMask)
   1154         isSplat = false;
   1155       SplatElt = eltMask;
   1156     }
   1157 
   1158     newMask.push_back(eltMask);
   1159   }
   1160 
   1161   // If the result mask is equal to one of the original shuffle masks,
   1162   // or is a splat, do the replacement.
   1163   if (isSplat || newMask == LHSMask || newMask == RHSMask || newMask == Mask) {
   1164     SmallVector<Constant*, 16> Elts;
   1165     for (unsigned i = 0, e = newMask.size(); i != e; ++i) {
   1166       if (newMask[i] < 0) {
   1167         Elts.push_back(UndefValue::get(Int32Ty));
   1168       } else {
   1169         Elts.push_back(ConstantInt::get(Int32Ty, newMask[i]));
   1170       }
   1171     }
   1172     if (!newRHS)
   1173       newRHS = UndefValue::get(newLHS->getType());
   1174     return new ShuffleVectorInst(newLHS, newRHS, ConstantVector::get(Elts));
   1175   }
   1176 
   1177   // If the result mask is an identity, replace uses of this instruction with
   1178   // corresponding argument.
   1179   bool isLHSID, isRHSID;
   1180   recognizeIdentityMask(newMask, isLHSID, isRHSID);
   1181   if (isLHSID && VWidth == LHSOp0Width) return ReplaceInstUsesWith(SVI, newLHS);
   1182   if (isRHSID && VWidth == RHSOp0Width) return ReplaceInstUsesWith(SVI, newRHS);
   1183 
   1184   return MadeChange ? &SVI : nullptr;
   1185 }
   1186