Home | History | Annotate | Download | only in InstCombine
      1 //===- InstCombineLoadStoreAlloca.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 load, store and alloca.
     11 //
     12 //===----------------------------------------------------------------------===//
     13 
     14 #include "InstCombineInternal.h"
     15 #include "llvm/ADT/Statistic.h"
     16 #include "llvm/Analysis/Loads.h"
     17 #include "llvm/IR/DataLayout.h"
     18 #include "llvm/IR/LLVMContext.h"
     19 #include "llvm/IR/IntrinsicInst.h"
     20 #include "llvm/IR/MDBuilder.h"
     21 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
     22 #include "llvm/Transforms/Utils/Local.h"
     23 using namespace llvm;
     24 
     25 #define DEBUG_TYPE "instcombine"
     26 
     27 STATISTIC(NumDeadStore,    "Number of dead stores eliminated");
     28 STATISTIC(NumGlobalCopies, "Number of allocas copied from constant global");
     29 
     30 /// pointsToConstantGlobal - Return true if V (possibly indirectly) points to
     31 /// some part of a constant global variable.  This intentionally only accepts
     32 /// constant expressions because we can't rewrite arbitrary instructions.
     33 static bool pointsToConstantGlobal(Value *V) {
     34   if (GlobalVariable *GV = dyn_cast<GlobalVariable>(V))
     35     return GV->isConstant();
     36 
     37   if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V)) {
     38     if (CE->getOpcode() == Instruction::BitCast ||
     39         CE->getOpcode() == Instruction::AddrSpaceCast ||
     40         CE->getOpcode() == Instruction::GetElementPtr)
     41       return pointsToConstantGlobal(CE->getOperand(0));
     42   }
     43   return false;
     44 }
     45 
     46 /// isOnlyCopiedFromConstantGlobal - Recursively walk the uses of a (derived)
     47 /// pointer to an alloca.  Ignore any reads of the pointer, return false if we
     48 /// see any stores or other unknown uses.  If we see pointer arithmetic, keep
     49 /// track of whether it moves the pointer (with IsOffset) but otherwise traverse
     50 /// the uses.  If we see a memcpy/memmove that targets an unoffseted pointer to
     51 /// the alloca, and if the source pointer is a pointer to a constant global, we
     52 /// can optimize this.
     53 static bool
     54 isOnlyCopiedFromConstantGlobal(Value *V, MemTransferInst *&TheCopy,
     55                                SmallVectorImpl<Instruction *> &ToDelete) {
     56   // We track lifetime intrinsics as we encounter them.  If we decide to go
     57   // ahead and replace the value with the global, this lets the caller quickly
     58   // eliminate the markers.
     59 
     60   SmallVector<std::pair<Value *, bool>, 35> ValuesToInspect;
     61   ValuesToInspect.push_back(std::make_pair(V, false));
     62   while (!ValuesToInspect.empty()) {
     63     auto ValuePair = ValuesToInspect.pop_back_val();
     64     const bool IsOffset = ValuePair.second;
     65     for (auto &U : ValuePair.first->uses()) {
     66       Instruction *I = cast<Instruction>(U.getUser());
     67 
     68       if (LoadInst *LI = dyn_cast<LoadInst>(I)) {
     69         // Ignore non-volatile loads, they are always ok.
     70         if (!LI->isSimple()) return false;
     71         continue;
     72       }
     73 
     74       if (isa<BitCastInst>(I) || isa<AddrSpaceCastInst>(I)) {
     75         // If uses of the bitcast are ok, we are ok.
     76         ValuesToInspect.push_back(std::make_pair(I, IsOffset));
     77         continue;
     78       }
     79       if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(I)) {
     80         // If the GEP has all zero indices, it doesn't offset the pointer. If it
     81         // doesn't, it does.
     82         ValuesToInspect.push_back(
     83             std::make_pair(I, IsOffset || !GEP->hasAllZeroIndices()));
     84         continue;
     85       }
     86 
     87       if (auto CS = CallSite(I)) {
     88         // If this is the function being called then we treat it like a load and
     89         // ignore it.
     90         if (CS.isCallee(&U))
     91           continue;
     92 
     93         // Inalloca arguments are clobbered by the call.
     94         unsigned ArgNo = CS.getArgumentNo(&U);
     95         if (CS.isInAllocaArgument(ArgNo))
     96           return false;
     97 
     98         // If this is a readonly/readnone call site, then we know it is just a
     99         // load (but one that potentially returns the value itself), so we can
    100         // ignore it if we know that the value isn't captured.
    101         if (CS.onlyReadsMemory() &&
    102             (CS.getInstruction()->use_empty() || CS.doesNotCapture(ArgNo)))
    103           continue;
    104 
    105         // If this is being passed as a byval argument, the caller is making a
    106         // copy, so it is only a read of the alloca.
    107         if (CS.isByValArgument(ArgNo))
    108           continue;
    109       }
    110 
    111       // Lifetime intrinsics can be handled by the caller.
    112       if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {
    113         if (II->getIntrinsicID() == Intrinsic::lifetime_start ||
    114             II->getIntrinsicID() == Intrinsic::lifetime_end) {
    115           assert(II->use_empty() && "Lifetime markers have no result to use!");
    116           ToDelete.push_back(II);
    117           continue;
    118         }
    119       }
    120 
    121       // If this is isn't our memcpy/memmove, reject it as something we can't
    122       // handle.
    123       MemTransferInst *MI = dyn_cast<MemTransferInst>(I);
    124       if (!MI)
    125         return false;
    126 
    127       // If the transfer is using the alloca as a source of the transfer, then
    128       // ignore it since it is a load (unless the transfer is volatile).
    129       if (U.getOperandNo() == 1) {
    130         if (MI->isVolatile()) return false;
    131         continue;
    132       }
    133 
    134       // If we already have seen a copy, reject the second one.
    135       if (TheCopy) return false;
    136 
    137       // If the pointer has been offset from the start of the alloca, we can't
    138       // safely handle this.
    139       if (IsOffset) return false;
    140 
    141       // If the memintrinsic isn't using the alloca as the dest, reject it.
    142       if (U.getOperandNo() != 0) return false;
    143 
    144       // If the source of the memcpy/move is not a constant global, reject it.
    145       if (!pointsToConstantGlobal(MI->getSource()))
    146         return false;
    147 
    148       // Otherwise, the transform is safe.  Remember the copy instruction.
    149       TheCopy = MI;
    150     }
    151   }
    152   return true;
    153 }
    154 
    155 /// isOnlyCopiedFromConstantGlobal - Return true if the specified alloca is only
    156 /// modified by a copy from a constant global.  If we can prove this, we can
    157 /// replace any uses of the alloca with uses of the global directly.
    158 static MemTransferInst *
    159 isOnlyCopiedFromConstantGlobal(AllocaInst *AI,
    160                                SmallVectorImpl<Instruction *> &ToDelete) {
    161   MemTransferInst *TheCopy = nullptr;
    162   if (isOnlyCopiedFromConstantGlobal(AI, TheCopy, ToDelete))
    163     return TheCopy;
    164   return nullptr;
    165 }
    166 
    167 static Instruction *simplifyAllocaArraySize(InstCombiner &IC, AllocaInst &AI) {
    168   // Check for array size of 1 (scalar allocation).
    169   if (!AI.isArrayAllocation()) {
    170     // i32 1 is the canonical array size for scalar allocations.
    171     if (AI.getArraySize()->getType()->isIntegerTy(32))
    172       return nullptr;
    173 
    174     // Canonicalize it.
    175     Value *V = IC.Builder->getInt32(1);
    176     AI.setOperand(0, V);
    177     return &AI;
    178   }
    179 
    180   // Convert: alloca Ty, C - where C is a constant != 1 into: alloca [C x Ty], 1
    181   if (const ConstantInt *C = dyn_cast<ConstantInt>(AI.getArraySize())) {
    182     Type *NewTy = ArrayType::get(AI.getAllocatedType(), C->getZExtValue());
    183     AllocaInst *New = IC.Builder->CreateAlloca(NewTy, nullptr, AI.getName());
    184     New->setAlignment(AI.getAlignment());
    185 
    186     // Scan to the end of the allocation instructions, to skip over a block of
    187     // allocas if possible...also skip interleaved debug info
    188     //
    189     BasicBlock::iterator It = New;
    190     while (isa<AllocaInst>(*It) || isa<DbgInfoIntrinsic>(*It))
    191       ++It;
    192 
    193     // Now that I is pointing to the first non-allocation-inst in the block,
    194     // insert our getelementptr instruction...
    195     //
    196     Type *IdxTy = IC.getDataLayout().getIntPtrType(AI.getType());
    197     Value *NullIdx = Constant::getNullValue(IdxTy);
    198     Value *Idx[2] = {NullIdx, NullIdx};
    199     Instruction *GEP =
    200         GetElementPtrInst::CreateInBounds(New, Idx, New->getName() + ".sub");
    201     IC.InsertNewInstBefore(GEP, *It);
    202 
    203     // Now make everything use the getelementptr instead of the original
    204     // allocation.
    205     return IC.ReplaceInstUsesWith(AI, GEP);
    206   }
    207 
    208   if (isa<UndefValue>(AI.getArraySize()))
    209     return IC.ReplaceInstUsesWith(AI, Constant::getNullValue(AI.getType()));
    210 
    211   // Ensure that the alloca array size argument has type intptr_t, so that
    212   // any casting is exposed early.
    213   Type *IntPtrTy = IC.getDataLayout().getIntPtrType(AI.getType());
    214   if (AI.getArraySize()->getType() != IntPtrTy) {
    215     Value *V = IC.Builder->CreateIntCast(AI.getArraySize(), IntPtrTy, false);
    216     AI.setOperand(0, V);
    217     return &AI;
    218   }
    219 
    220   return nullptr;
    221 }
    222 
    223 Instruction *InstCombiner::visitAllocaInst(AllocaInst &AI) {
    224   if (auto *I = simplifyAllocaArraySize(*this, AI))
    225     return I;
    226 
    227   if (AI.getAllocatedType()->isSized()) {
    228     // If the alignment is 0 (unspecified), assign it the preferred alignment.
    229     if (AI.getAlignment() == 0)
    230       AI.setAlignment(DL.getPrefTypeAlignment(AI.getAllocatedType()));
    231 
    232     // Move all alloca's of zero byte objects to the entry block and merge them
    233     // together.  Note that we only do this for alloca's, because malloc should
    234     // allocate and return a unique pointer, even for a zero byte allocation.
    235     if (DL.getTypeAllocSize(AI.getAllocatedType()) == 0) {
    236       // For a zero sized alloca there is no point in doing an array allocation.
    237       // This is helpful if the array size is a complicated expression not used
    238       // elsewhere.
    239       if (AI.isArrayAllocation()) {
    240         AI.setOperand(0, ConstantInt::get(AI.getArraySize()->getType(), 1));
    241         return &AI;
    242       }
    243 
    244       // Get the first instruction in the entry block.
    245       BasicBlock &EntryBlock = AI.getParent()->getParent()->getEntryBlock();
    246       Instruction *FirstInst = EntryBlock.getFirstNonPHIOrDbg();
    247       if (FirstInst != &AI) {
    248         // If the entry block doesn't start with a zero-size alloca then move
    249         // this one to the start of the entry block.  There is no problem with
    250         // dominance as the array size was forced to a constant earlier already.
    251         AllocaInst *EntryAI = dyn_cast<AllocaInst>(FirstInst);
    252         if (!EntryAI || !EntryAI->getAllocatedType()->isSized() ||
    253             DL.getTypeAllocSize(EntryAI->getAllocatedType()) != 0) {
    254           AI.moveBefore(FirstInst);
    255           return &AI;
    256         }
    257 
    258         // If the alignment of the entry block alloca is 0 (unspecified),
    259         // assign it the preferred alignment.
    260         if (EntryAI->getAlignment() == 0)
    261           EntryAI->setAlignment(
    262               DL.getPrefTypeAlignment(EntryAI->getAllocatedType()));
    263         // Replace this zero-sized alloca with the one at the start of the entry
    264         // block after ensuring that the address will be aligned enough for both
    265         // types.
    266         unsigned MaxAlign = std::max(EntryAI->getAlignment(),
    267                                      AI.getAlignment());
    268         EntryAI->setAlignment(MaxAlign);
    269         if (AI.getType() != EntryAI->getType())
    270           return new BitCastInst(EntryAI, AI.getType());
    271         return ReplaceInstUsesWith(AI, EntryAI);
    272       }
    273     }
    274   }
    275 
    276   if (AI.getAlignment()) {
    277     // Check to see if this allocation is only modified by a memcpy/memmove from
    278     // a constant global whose alignment is equal to or exceeds that of the
    279     // allocation.  If this is the case, we can change all users to use
    280     // the constant global instead.  This is commonly produced by the CFE by
    281     // constructs like "void foo() { int A[] = {1,2,3,4,5,6,7,8,9...}; }" if 'A'
    282     // is only subsequently read.
    283     SmallVector<Instruction *, 4> ToDelete;
    284     if (MemTransferInst *Copy = isOnlyCopiedFromConstantGlobal(&AI, ToDelete)) {
    285       unsigned SourceAlign = getOrEnforceKnownAlignment(
    286           Copy->getSource(), AI.getAlignment(), DL, &AI, AC, DT);
    287       if (AI.getAlignment() <= SourceAlign) {
    288         DEBUG(dbgs() << "Found alloca equal to global: " << AI << '\n');
    289         DEBUG(dbgs() << "  memcpy = " << *Copy << '\n');
    290         for (unsigned i = 0, e = ToDelete.size(); i != e; ++i)
    291           EraseInstFromFunction(*ToDelete[i]);
    292         Constant *TheSrc = cast<Constant>(Copy->getSource());
    293         Constant *Cast
    294           = ConstantExpr::getPointerBitCastOrAddrSpaceCast(TheSrc, AI.getType());
    295         Instruction *NewI = ReplaceInstUsesWith(AI, Cast);
    296         EraseInstFromFunction(*Copy);
    297         ++NumGlobalCopies;
    298         return NewI;
    299       }
    300     }
    301   }
    302 
    303   // At last, use the generic allocation site handler to aggressively remove
    304   // unused allocas.
    305   return visitAllocSite(AI);
    306 }
    307 
    308 /// \brief Helper to combine a load to a new type.
    309 ///
    310 /// This just does the work of combining a load to a new type. It handles
    311 /// metadata, etc., and returns the new instruction. The \c NewTy should be the
    312 /// loaded *value* type. This will convert it to a pointer, cast the operand to
    313 /// that pointer type, load it, etc.
    314 ///
    315 /// Note that this will create all of the instructions with whatever insert
    316 /// point the \c InstCombiner currently is using.
    317 static LoadInst *combineLoadToNewType(InstCombiner &IC, LoadInst &LI, Type *NewTy) {
    318   Value *Ptr = LI.getPointerOperand();
    319   unsigned AS = LI.getPointerAddressSpace();
    320   SmallVector<std::pair<unsigned, MDNode *>, 8> MD;
    321   LI.getAllMetadata(MD);
    322 
    323   LoadInst *NewLoad = IC.Builder->CreateAlignedLoad(
    324       IC.Builder->CreateBitCast(Ptr, NewTy->getPointerTo(AS)),
    325       LI.getAlignment(), LI.getName());
    326   MDBuilder MDB(NewLoad->getContext());
    327   for (const auto &MDPair : MD) {
    328     unsigned ID = MDPair.first;
    329     MDNode *N = MDPair.second;
    330     // Note, essentially every kind of metadata should be preserved here! This
    331     // routine is supposed to clone a load instruction changing *only its type*.
    332     // The only metadata it makes sense to drop is metadata which is invalidated
    333     // when the pointer type changes. This should essentially never be the case
    334     // in LLVM, but we explicitly switch over only known metadata to be
    335     // conservatively correct. If you are adding metadata to LLVM which pertains
    336     // to loads, you almost certainly want to add it here.
    337     switch (ID) {
    338     case LLVMContext::MD_dbg:
    339     case LLVMContext::MD_tbaa:
    340     case LLVMContext::MD_prof:
    341     case LLVMContext::MD_fpmath:
    342     case LLVMContext::MD_tbaa_struct:
    343     case LLVMContext::MD_invariant_load:
    344     case LLVMContext::MD_alias_scope:
    345     case LLVMContext::MD_noalias:
    346     case LLVMContext::MD_nontemporal:
    347     case LLVMContext::MD_mem_parallel_loop_access:
    348       // All of these directly apply.
    349       NewLoad->setMetadata(ID, N);
    350       break;
    351 
    352     case LLVMContext::MD_nonnull:
    353       // This only directly applies if the new type is also a pointer.
    354       if (NewTy->isPointerTy()) {
    355         NewLoad->setMetadata(ID, N);
    356         break;
    357       }
    358       // If it's integral now, translate it to !range metadata.
    359       if (NewTy->isIntegerTy()) {
    360         auto *ITy = cast<IntegerType>(NewTy);
    361         auto *NullInt = ConstantExpr::getPtrToInt(
    362             ConstantPointerNull::get(cast<PointerType>(Ptr->getType())), ITy);
    363         auto *NonNullInt =
    364             ConstantExpr::getAdd(NullInt, ConstantInt::get(ITy, 1));
    365         NewLoad->setMetadata(LLVMContext::MD_range,
    366                              MDB.createRange(NonNullInt, NullInt));
    367       }
    368       break;
    369 
    370     case LLVMContext::MD_range:
    371       // FIXME: It would be nice to propagate this in some way, but the type
    372       // conversions make it hard. If the new type is a pointer, we could
    373       // translate it to !nonnull metadata.
    374       break;
    375     }
    376   }
    377   return NewLoad;
    378 }
    379 
    380 /// \brief Combine a store to a new type.
    381 ///
    382 /// Returns the newly created store instruction.
    383 static StoreInst *combineStoreToNewValue(InstCombiner &IC, StoreInst &SI, Value *V) {
    384   Value *Ptr = SI.getPointerOperand();
    385   unsigned AS = SI.getPointerAddressSpace();
    386   SmallVector<std::pair<unsigned, MDNode *>, 8> MD;
    387   SI.getAllMetadata(MD);
    388 
    389   StoreInst *NewStore = IC.Builder->CreateAlignedStore(
    390       V, IC.Builder->CreateBitCast(Ptr, V->getType()->getPointerTo(AS)),
    391       SI.getAlignment());
    392   for (const auto &MDPair : MD) {
    393     unsigned ID = MDPair.first;
    394     MDNode *N = MDPair.second;
    395     // Note, essentially every kind of metadata should be preserved here! This
    396     // routine is supposed to clone a store instruction changing *only its
    397     // type*. The only metadata it makes sense to drop is metadata which is
    398     // invalidated when the pointer type changes. This should essentially
    399     // never be the case in LLVM, but we explicitly switch over only known
    400     // metadata to be conservatively correct. If you are adding metadata to
    401     // LLVM which pertains to stores, you almost certainly want to add it
    402     // here.
    403     switch (ID) {
    404     case LLVMContext::MD_dbg:
    405     case LLVMContext::MD_tbaa:
    406     case LLVMContext::MD_prof:
    407     case LLVMContext::MD_fpmath:
    408     case LLVMContext::MD_tbaa_struct:
    409     case LLVMContext::MD_alias_scope:
    410     case LLVMContext::MD_noalias:
    411     case LLVMContext::MD_nontemporal:
    412     case LLVMContext::MD_mem_parallel_loop_access:
    413       // All of these directly apply.
    414       NewStore->setMetadata(ID, N);
    415       break;
    416 
    417     case LLVMContext::MD_invariant_load:
    418     case LLVMContext::MD_nonnull:
    419     case LLVMContext::MD_range:
    420       // These don't apply for stores.
    421       break;
    422     }
    423   }
    424 
    425   return NewStore;
    426 }
    427 
    428 /// \brief Combine loads to match the type of value their uses after looking
    429 /// through intervening bitcasts.
    430 ///
    431 /// The core idea here is that if the result of a load is used in an operation,
    432 /// we should load the type most conducive to that operation. For example, when
    433 /// loading an integer and converting that immediately to a pointer, we should
    434 /// instead directly load a pointer.
    435 ///
    436 /// However, this routine must never change the width of a load or the number of
    437 /// loads as that would introduce a semantic change. This combine is expected to
    438 /// be a semantic no-op which just allows loads to more closely model the types
    439 /// of their consuming operations.
    440 ///
    441 /// Currently, we also refuse to change the precise type used for an atomic load
    442 /// or a volatile load. This is debatable, and might be reasonable to change
    443 /// later. However, it is risky in case some backend or other part of LLVM is
    444 /// relying on the exact type loaded to select appropriate atomic operations.
    445 static Instruction *combineLoadToOperationType(InstCombiner &IC, LoadInst &LI) {
    446   // FIXME: We could probably with some care handle both volatile and atomic
    447   // loads here but it isn't clear that this is important.
    448   if (!LI.isSimple())
    449     return nullptr;
    450 
    451   if (LI.use_empty())
    452     return nullptr;
    453 
    454   Type *Ty = LI.getType();
    455   const DataLayout &DL = IC.getDataLayout();
    456 
    457   // Try to canonicalize loads which are only ever stored to operate over
    458   // integers instead of any other type. We only do this when the loaded type
    459   // is sized and has a size exactly the same as its store size and the store
    460   // size is a legal integer type.
    461   if (!Ty->isIntegerTy() && Ty->isSized() &&
    462       DL.isLegalInteger(DL.getTypeStoreSizeInBits(Ty)) &&
    463       DL.getTypeStoreSizeInBits(Ty) == DL.getTypeSizeInBits(Ty)) {
    464     if (std::all_of(LI.user_begin(), LI.user_end(), [&LI](User *U) {
    465           auto *SI = dyn_cast<StoreInst>(U);
    466           return SI && SI->getPointerOperand() != &LI;
    467         })) {
    468       LoadInst *NewLoad = combineLoadToNewType(
    469           IC, LI,
    470           Type::getIntNTy(LI.getContext(), DL.getTypeStoreSizeInBits(Ty)));
    471       // Replace all the stores with stores of the newly loaded value.
    472       for (auto UI = LI.user_begin(), UE = LI.user_end(); UI != UE;) {
    473         auto *SI = cast<StoreInst>(*UI++);
    474         IC.Builder->SetInsertPoint(SI);
    475         combineStoreToNewValue(IC, *SI, NewLoad);
    476         IC.EraseInstFromFunction(*SI);
    477       }
    478       assert(LI.use_empty() && "Failed to remove all users of the load!");
    479       // Return the old load so the combiner can delete it safely.
    480       return &LI;
    481     }
    482   }
    483 
    484   // Fold away bit casts of the loaded value by loading the desired type.
    485   if (LI.hasOneUse())
    486     if (auto *BC = dyn_cast<BitCastInst>(LI.user_back())) {
    487       LoadInst *NewLoad = combineLoadToNewType(IC, LI, BC->getDestTy());
    488       BC->replaceAllUsesWith(NewLoad);
    489       IC.EraseInstFromFunction(*BC);
    490       return &LI;
    491     }
    492 
    493   // FIXME: We should also canonicalize loads of vectors when their elements are
    494   // cast to other types.
    495   return nullptr;
    496 }
    497 
    498 // If we can determine that all possible objects pointed to by the provided
    499 // pointer value are, not only dereferenceable, but also definitively less than
    500 // or equal to the provided maximum size, then return true. Otherwise, return
    501 // false (constant global values and allocas fall into this category).
    502 //
    503 // FIXME: This should probably live in ValueTracking (or similar).
    504 static bool isObjectSizeLessThanOrEq(Value *V, uint64_t MaxSize,
    505                                      const DataLayout &DL) {
    506   SmallPtrSet<Value *, 4> Visited;
    507   SmallVector<Value *, 4> Worklist(1, V);
    508 
    509   do {
    510     Value *P = Worklist.pop_back_val();
    511     P = P->stripPointerCasts();
    512 
    513     if (!Visited.insert(P).second)
    514       continue;
    515 
    516     if (SelectInst *SI = dyn_cast<SelectInst>(P)) {
    517       Worklist.push_back(SI->getTrueValue());
    518       Worklist.push_back(SI->getFalseValue());
    519       continue;
    520     }
    521 
    522     if (PHINode *PN = dyn_cast<PHINode>(P)) {
    523       for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
    524         Worklist.push_back(PN->getIncomingValue(i));
    525       continue;
    526     }
    527 
    528     if (GlobalAlias *GA = dyn_cast<GlobalAlias>(P)) {
    529       if (GA->mayBeOverridden())
    530         return false;
    531       Worklist.push_back(GA->getAliasee());
    532       continue;
    533     }
    534 
    535     // If we know how big this object is, and it is less than MaxSize, continue
    536     // searching. Otherwise, return false.
    537     if (AllocaInst *AI = dyn_cast<AllocaInst>(P)) {
    538       if (!AI->getAllocatedType()->isSized())
    539         return false;
    540 
    541       ConstantInt *CS = dyn_cast<ConstantInt>(AI->getArraySize());
    542       if (!CS)
    543         return false;
    544 
    545       uint64_t TypeSize = DL.getTypeAllocSize(AI->getAllocatedType());
    546       // Make sure that, even if the multiplication below would wrap as an
    547       // uint64_t, we still do the right thing.
    548       if ((CS->getValue().zextOrSelf(128)*APInt(128, TypeSize)).ugt(MaxSize))
    549         return false;
    550       continue;
    551     }
    552 
    553     if (GlobalVariable *GV = dyn_cast<GlobalVariable>(P)) {
    554       if (!GV->hasDefinitiveInitializer() || !GV->isConstant())
    555         return false;
    556 
    557       uint64_t InitSize = DL.getTypeAllocSize(GV->getType()->getElementType());
    558       if (InitSize > MaxSize)
    559         return false;
    560       continue;
    561     }
    562 
    563     return false;
    564   } while (!Worklist.empty());
    565 
    566   return true;
    567 }
    568 
    569 // If we're indexing into an object of a known size, and the outer index is
    570 // not a constant, but having any value but zero would lead to undefined
    571 // behavior, replace it with zero.
    572 //
    573 // For example, if we have:
    574 // @f.a = private unnamed_addr constant [1 x i32] [i32 12], align 4
    575 // ...
    576 // %arrayidx = getelementptr inbounds [1 x i32]* @f.a, i64 0, i64 %x
    577 // ... = load i32* %arrayidx, align 4
    578 // Then we know that we can replace %x in the GEP with i64 0.
    579 //
    580 // FIXME: We could fold any GEP index to zero that would cause UB if it were
    581 // not zero. Currently, we only handle the first such index. Also, we could
    582 // also search through non-zero constant indices if we kept track of the
    583 // offsets those indices implied.
    584 static bool canReplaceGEPIdxWithZero(InstCombiner &IC, GetElementPtrInst *GEPI,
    585                                      Instruction *MemI, unsigned &Idx) {
    586   if (GEPI->getNumOperands() < 2)
    587     return false;
    588 
    589   // Find the first non-zero index of a GEP. If all indices are zero, return
    590   // one past the last index.
    591   auto FirstNZIdx = [](const GetElementPtrInst *GEPI) {
    592     unsigned I = 1;
    593     for (unsigned IE = GEPI->getNumOperands(); I != IE; ++I) {
    594       Value *V = GEPI->getOperand(I);
    595       if (const ConstantInt *CI = dyn_cast<ConstantInt>(V))
    596         if (CI->isZero())
    597           continue;
    598 
    599       break;
    600     }
    601 
    602     return I;
    603   };
    604 
    605   // Skip through initial 'zero' indices, and find the corresponding pointer
    606   // type. See if the next index is not a constant.
    607   Idx = FirstNZIdx(GEPI);
    608   if (Idx == GEPI->getNumOperands())
    609     return false;
    610   if (isa<Constant>(GEPI->getOperand(Idx)))
    611     return false;
    612 
    613   SmallVector<Value *, 4> Ops(GEPI->idx_begin(), GEPI->idx_begin() + Idx);
    614   Type *AllocTy = GetElementPtrInst::getIndexedType(
    615       cast<PointerType>(GEPI->getOperand(0)->getType()->getScalarType())
    616           ->getElementType(),
    617       Ops);
    618   if (!AllocTy || !AllocTy->isSized())
    619     return false;
    620   const DataLayout &DL = IC.getDataLayout();
    621   uint64_t TyAllocSize = DL.getTypeAllocSize(AllocTy);
    622 
    623   // If there are more indices after the one we might replace with a zero, make
    624   // sure they're all non-negative. If any of them are negative, the overall
    625   // address being computed might be before the base address determined by the
    626   // first non-zero index.
    627   auto IsAllNonNegative = [&]() {
    628     for (unsigned i = Idx+1, e = GEPI->getNumOperands(); i != e; ++i) {
    629       bool KnownNonNegative, KnownNegative;
    630       IC.ComputeSignBit(GEPI->getOperand(i), KnownNonNegative,
    631                         KnownNegative, 0, MemI);
    632       if (KnownNonNegative)
    633         continue;
    634       return false;
    635     }
    636 
    637     return true;
    638   };
    639 
    640   // FIXME: If the GEP is not inbounds, and there are extra indices after the
    641   // one we'll replace, those could cause the address computation to wrap
    642   // (rendering the IsAllNonNegative() check below insufficient). We can do
    643   // better, ignoring zero indicies (and other indicies we can prove small
    644   // enough not to wrap).
    645   if (Idx+1 != GEPI->getNumOperands() && !GEPI->isInBounds())
    646     return false;
    647 
    648   // Note that isObjectSizeLessThanOrEq will return true only if the pointer is
    649   // also known to be dereferenceable.
    650   return isObjectSizeLessThanOrEq(GEPI->getOperand(0), TyAllocSize, DL) &&
    651          IsAllNonNegative();
    652 }
    653 
    654 // If we're indexing into an object with a variable index for the memory
    655 // access, but the object has only one element, we can assume that the index
    656 // will always be zero. If we replace the GEP, return it.
    657 template <typename T>
    658 static Instruction *replaceGEPIdxWithZero(InstCombiner &IC, Value *Ptr,
    659                                           T &MemI) {
    660   if (GetElementPtrInst *GEPI = dyn_cast<GetElementPtrInst>(Ptr)) {
    661     unsigned Idx;
    662     if (canReplaceGEPIdxWithZero(IC, GEPI, &MemI, Idx)) {
    663       Instruction *NewGEPI = GEPI->clone();
    664       NewGEPI->setOperand(Idx,
    665         ConstantInt::get(GEPI->getOperand(Idx)->getType(), 0));
    666       NewGEPI->insertBefore(GEPI);
    667       MemI.setOperand(MemI.getPointerOperandIndex(), NewGEPI);
    668       return NewGEPI;
    669     }
    670   }
    671 
    672   return nullptr;
    673 }
    674 
    675 Instruction *InstCombiner::visitLoadInst(LoadInst &LI) {
    676   Value *Op = LI.getOperand(0);
    677 
    678   // Try to canonicalize the loaded type.
    679   if (Instruction *Res = combineLoadToOperationType(*this, LI))
    680     return Res;
    681 
    682   // Attempt to improve the alignment.
    683   unsigned KnownAlign = getOrEnforceKnownAlignment(
    684       Op, DL.getPrefTypeAlignment(LI.getType()), DL, &LI, AC, DT);
    685   unsigned LoadAlign = LI.getAlignment();
    686   unsigned EffectiveLoadAlign =
    687       LoadAlign != 0 ? LoadAlign : DL.getABITypeAlignment(LI.getType());
    688 
    689   if (KnownAlign > EffectiveLoadAlign)
    690     LI.setAlignment(KnownAlign);
    691   else if (LoadAlign == 0)
    692     LI.setAlignment(EffectiveLoadAlign);
    693 
    694   // Replace GEP indices if possible.
    695   if (Instruction *NewGEPI = replaceGEPIdxWithZero(*this, Op, LI)) {
    696       Worklist.Add(NewGEPI);
    697       return &LI;
    698   }
    699 
    700   // None of the following transforms are legal for volatile/atomic loads.
    701   // FIXME: Some of it is okay for atomic loads; needs refactoring.
    702   if (!LI.isSimple()) return nullptr;
    703 
    704   // Do really simple store-to-load forwarding and load CSE, to catch cases
    705   // where there are several consecutive memory accesses to the same location,
    706   // separated by a few arithmetic operations.
    707   BasicBlock::iterator BBI = &LI;
    708   if (Value *AvailableVal = FindAvailableLoadedValue(Op, LI.getParent(), BBI,6))
    709     return ReplaceInstUsesWith(
    710         LI, Builder->CreateBitOrPointerCast(AvailableVal, LI.getType(),
    711                                             LI.getName() + ".cast"));
    712 
    713   // load(gep null, ...) -> unreachable
    714   if (GetElementPtrInst *GEPI = dyn_cast<GetElementPtrInst>(Op)) {
    715     const Value *GEPI0 = GEPI->getOperand(0);
    716     // TODO: Consider a target hook for valid address spaces for this xform.
    717     if (isa<ConstantPointerNull>(GEPI0) && GEPI->getPointerAddressSpace() == 0){
    718       // Insert a new store to null instruction before the load to indicate
    719       // that this code is not reachable.  We do this instead of inserting
    720       // an unreachable instruction directly because we cannot modify the
    721       // CFG.
    722       new StoreInst(UndefValue::get(LI.getType()),
    723                     Constant::getNullValue(Op->getType()), &LI);
    724       return ReplaceInstUsesWith(LI, UndefValue::get(LI.getType()));
    725     }
    726   }
    727 
    728   // load null/undef -> unreachable
    729   // TODO: Consider a target hook for valid address spaces for this xform.
    730   if (isa<UndefValue>(Op) ||
    731       (isa<ConstantPointerNull>(Op) && LI.getPointerAddressSpace() == 0)) {
    732     // Insert a new store to null instruction before the load to indicate that
    733     // this code is not reachable.  We do this instead of inserting an
    734     // unreachable instruction directly because we cannot modify the CFG.
    735     new StoreInst(UndefValue::get(LI.getType()),
    736                   Constant::getNullValue(Op->getType()), &LI);
    737     return ReplaceInstUsesWith(LI, UndefValue::get(LI.getType()));
    738   }
    739 
    740   if (Op->hasOneUse()) {
    741     // Change select and PHI nodes to select values instead of addresses: this
    742     // helps alias analysis out a lot, allows many others simplifications, and
    743     // exposes redundancy in the code.
    744     //
    745     // Note that we cannot do the transformation unless we know that the
    746     // introduced loads cannot trap!  Something like this is valid as long as
    747     // the condition is always false: load (select bool %C, int* null, int* %G),
    748     // but it would not be valid if we transformed it to load from null
    749     // unconditionally.
    750     //
    751     if (SelectInst *SI = dyn_cast<SelectInst>(Op)) {
    752       // load (select (Cond, &V1, &V2))  --> select(Cond, load &V1, load &V2).
    753       unsigned Align = LI.getAlignment();
    754       if (isSafeToLoadUnconditionally(SI->getOperand(1), SI, Align) &&
    755           isSafeToLoadUnconditionally(SI->getOperand(2), SI, Align)) {
    756         LoadInst *V1 = Builder->CreateLoad(SI->getOperand(1),
    757                                            SI->getOperand(1)->getName()+".val");
    758         LoadInst *V2 = Builder->CreateLoad(SI->getOperand(2),
    759                                            SI->getOperand(2)->getName()+".val");
    760         V1->setAlignment(Align);
    761         V2->setAlignment(Align);
    762         return SelectInst::Create(SI->getCondition(), V1, V2);
    763       }
    764 
    765       // load (select (cond, null, P)) -> load P
    766       if (isa<ConstantPointerNull>(SI->getOperand(1)) &&
    767           LI.getPointerAddressSpace() == 0) {
    768         LI.setOperand(0, SI->getOperand(2));
    769         return &LI;
    770       }
    771 
    772       // load (select (cond, P, null)) -> load P
    773       if (isa<ConstantPointerNull>(SI->getOperand(2)) &&
    774           LI.getPointerAddressSpace() == 0) {
    775         LI.setOperand(0, SI->getOperand(1));
    776         return &LI;
    777       }
    778     }
    779   }
    780   return nullptr;
    781 }
    782 
    783 /// \brief Combine stores to match the type of value being stored.
    784 ///
    785 /// The core idea here is that the memory does not have any intrinsic type and
    786 /// where we can we should match the type of a store to the type of value being
    787 /// stored.
    788 ///
    789 /// However, this routine must never change the width of a store or the number of
    790 /// stores as that would introduce a semantic change. This combine is expected to
    791 /// be a semantic no-op which just allows stores to more closely model the types
    792 /// of their incoming values.
    793 ///
    794 /// Currently, we also refuse to change the precise type used for an atomic or
    795 /// volatile store. This is debatable, and might be reasonable to change later.
    796 /// However, it is risky in case some backend or other part of LLVM is relying
    797 /// on the exact type stored to select appropriate atomic operations.
    798 ///
    799 /// \returns true if the store was successfully combined away. This indicates
    800 /// the caller must erase the store instruction. We have to let the caller erase
    801 /// the store instruction sas otherwise there is no way to signal whether it was
    802 /// combined or not: IC.EraseInstFromFunction returns a null pointer.
    803 static bool combineStoreToValueType(InstCombiner &IC, StoreInst &SI) {
    804   // FIXME: We could probably with some care handle both volatile and atomic
    805   // stores here but it isn't clear that this is important.
    806   if (!SI.isSimple())
    807     return false;
    808 
    809   Value *V = SI.getValueOperand();
    810 
    811   // Fold away bit casts of the stored value by storing the original type.
    812   if (auto *BC = dyn_cast<BitCastInst>(V)) {
    813     V = BC->getOperand(0);
    814     combineStoreToNewValue(IC, SI, V);
    815     return true;
    816   }
    817 
    818   // FIXME: We should also canonicalize loads of vectors when their elements are
    819   // cast to other types.
    820   return false;
    821 }
    822 
    823 static bool unpackStoreToAggregate(InstCombiner &IC, StoreInst &SI) {
    824   // FIXME: We could probably with some care handle both volatile and atomic
    825   // stores here but it isn't clear that this is important.
    826   if (!SI.isSimple())
    827     return false;
    828 
    829   Value *V = SI.getValueOperand();
    830   Type *T = V->getType();
    831 
    832   if (!T->isAggregateType())
    833     return false;
    834 
    835   if (StructType *ST = dyn_cast<StructType>(T)) {
    836     // If the struct only have one element, we unpack.
    837     if (ST->getNumElements() == 1) {
    838       V = IC.Builder->CreateExtractValue(V, 0);
    839       combineStoreToNewValue(IC, SI, V);
    840       return true;
    841     }
    842   }
    843 
    844   return false;
    845 }
    846 
    847 /// equivalentAddressValues - Test if A and B will obviously have the same
    848 /// value. This includes recognizing that %t0 and %t1 will have the same
    849 /// value in code like this:
    850 ///   %t0 = getelementptr \@a, 0, 3
    851 ///   store i32 0, i32* %t0
    852 ///   %t1 = getelementptr \@a, 0, 3
    853 ///   %t2 = load i32* %t1
    854 ///
    855 static bool equivalentAddressValues(Value *A, Value *B) {
    856   // Test if the values are trivially equivalent.
    857   if (A == B) return true;
    858 
    859   // Test if the values come form identical arithmetic instructions.
    860   // This uses isIdenticalToWhenDefined instead of isIdenticalTo because
    861   // its only used to compare two uses within the same basic block, which
    862   // means that they'll always either have the same value or one of them
    863   // will have an undefined value.
    864   if (isa<BinaryOperator>(A) ||
    865       isa<CastInst>(A) ||
    866       isa<PHINode>(A) ||
    867       isa<GetElementPtrInst>(A))
    868     if (Instruction *BI = dyn_cast<Instruction>(B))
    869       if (cast<Instruction>(A)->isIdenticalToWhenDefined(BI))
    870         return true;
    871 
    872   // Otherwise they may not be equivalent.
    873   return false;
    874 }
    875 
    876 Instruction *InstCombiner::visitStoreInst(StoreInst &SI) {
    877   Value *Val = SI.getOperand(0);
    878   Value *Ptr = SI.getOperand(1);
    879 
    880   // Try to canonicalize the stored type.
    881   if (combineStoreToValueType(*this, SI))
    882     return EraseInstFromFunction(SI);
    883 
    884   // Attempt to improve the alignment.
    885   unsigned KnownAlign = getOrEnforceKnownAlignment(
    886       Ptr, DL.getPrefTypeAlignment(Val->getType()), DL, &SI, AC, DT);
    887   unsigned StoreAlign = SI.getAlignment();
    888   unsigned EffectiveStoreAlign =
    889       StoreAlign != 0 ? StoreAlign : DL.getABITypeAlignment(Val->getType());
    890 
    891   if (KnownAlign > EffectiveStoreAlign)
    892     SI.setAlignment(KnownAlign);
    893   else if (StoreAlign == 0)
    894     SI.setAlignment(EffectiveStoreAlign);
    895 
    896   // Try to canonicalize the stored type.
    897   if (unpackStoreToAggregate(*this, SI))
    898     return EraseInstFromFunction(SI);
    899 
    900   // Replace GEP indices if possible.
    901   if (Instruction *NewGEPI = replaceGEPIdxWithZero(*this, Ptr, SI)) {
    902       Worklist.Add(NewGEPI);
    903       return &SI;
    904   }
    905 
    906   // Don't hack volatile/atomic stores.
    907   // FIXME: Some bits are legal for atomic stores; needs refactoring.
    908   if (!SI.isSimple()) return nullptr;
    909 
    910   // If the RHS is an alloca with a single use, zapify the store, making the
    911   // alloca dead.
    912   if (Ptr->hasOneUse()) {
    913     if (isa<AllocaInst>(Ptr))
    914       return EraseInstFromFunction(SI);
    915     if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(Ptr)) {
    916       if (isa<AllocaInst>(GEP->getOperand(0))) {
    917         if (GEP->getOperand(0)->hasOneUse())
    918           return EraseInstFromFunction(SI);
    919       }
    920     }
    921   }
    922 
    923   // Do really simple DSE, to catch cases where there are several consecutive
    924   // stores to the same location, separated by a few arithmetic operations. This
    925   // situation often occurs with bitfield accesses.
    926   BasicBlock::iterator BBI = &SI;
    927   for (unsigned ScanInsts = 6; BBI != SI.getParent()->begin() && ScanInsts;
    928        --ScanInsts) {
    929     --BBI;
    930     // Don't count debug info directives, lest they affect codegen,
    931     // and we skip pointer-to-pointer bitcasts, which are NOPs.
    932     if (isa<DbgInfoIntrinsic>(BBI) ||
    933         (isa<BitCastInst>(BBI) && BBI->getType()->isPointerTy())) {
    934       ScanInsts++;
    935       continue;
    936     }
    937 
    938     if (StoreInst *PrevSI = dyn_cast<StoreInst>(BBI)) {
    939       // Prev store isn't volatile, and stores to the same location?
    940       if (PrevSI->isSimple() && equivalentAddressValues(PrevSI->getOperand(1),
    941                                                         SI.getOperand(1))) {
    942         ++NumDeadStore;
    943         ++BBI;
    944         EraseInstFromFunction(*PrevSI);
    945         continue;
    946       }
    947       break;
    948     }
    949 
    950     // If this is a load, we have to stop.  However, if the loaded value is from
    951     // the pointer we're loading and is producing the pointer we're storing,
    952     // then *this* store is dead (X = load P; store X -> P).
    953     if (LoadInst *LI = dyn_cast<LoadInst>(BBI)) {
    954       if (LI == Val && equivalentAddressValues(LI->getOperand(0), Ptr) &&
    955           LI->isSimple())
    956         return EraseInstFromFunction(SI);
    957 
    958       // Otherwise, this is a load from some other location.  Stores before it
    959       // may not be dead.
    960       break;
    961     }
    962 
    963     // Don't skip over loads or things that can modify memory.
    964     if (BBI->mayWriteToMemory() || BBI->mayReadFromMemory())
    965       break;
    966   }
    967 
    968   // store X, null    -> turns into 'unreachable' in SimplifyCFG
    969   if (isa<ConstantPointerNull>(Ptr) && SI.getPointerAddressSpace() == 0) {
    970     if (!isa<UndefValue>(Val)) {
    971       SI.setOperand(0, UndefValue::get(Val->getType()));
    972       if (Instruction *U = dyn_cast<Instruction>(Val))
    973         Worklist.Add(U);  // Dropped a use.
    974     }
    975     return nullptr;  // Do not modify these!
    976   }
    977 
    978   // store undef, Ptr -> noop
    979   if (isa<UndefValue>(Val))
    980     return EraseInstFromFunction(SI);
    981 
    982   // If this store is the last instruction in the basic block (possibly
    983   // excepting debug info instructions), and if the block ends with an
    984   // unconditional branch, try to move it to the successor block.
    985   BBI = &SI;
    986   do {
    987     ++BBI;
    988   } while (isa<DbgInfoIntrinsic>(BBI) ||
    989            (isa<BitCastInst>(BBI) && BBI->getType()->isPointerTy()));
    990   if (BranchInst *BI = dyn_cast<BranchInst>(BBI))
    991     if (BI->isUnconditional())
    992       if (SimplifyStoreAtEndOfBlock(SI))
    993         return nullptr;  // xform done!
    994 
    995   return nullptr;
    996 }
    997 
    998 /// SimplifyStoreAtEndOfBlock - Turn things like:
    999 ///   if () { *P = v1; } else { *P = v2 }
   1000 /// into a phi node with a store in the successor.
   1001 ///
   1002 /// Simplify things like:
   1003 ///   *P = v1; if () { *P = v2; }
   1004 /// into a phi node with a store in the successor.
   1005 ///
   1006 bool InstCombiner::SimplifyStoreAtEndOfBlock(StoreInst &SI) {
   1007   BasicBlock *StoreBB = SI.getParent();
   1008 
   1009   // Check to see if the successor block has exactly two incoming edges.  If
   1010   // so, see if the other predecessor contains a store to the same location.
   1011   // if so, insert a PHI node (if needed) and move the stores down.
   1012   BasicBlock *DestBB = StoreBB->getTerminator()->getSuccessor(0);
   1013 
   1014   // Determine whether Dest has exactly two predecessors and, if so, compute
   1015   // the other predecessor.
   1016   pred_iterator PI = pred_begin(DestBB);
   1017   BasicBlock *P = *PI;
   1018   BasicBlock *OtherBB = nullptr;
   1019 
   1020   if (P != StoreBB)
   1021     OtherBB = P;
   1022 
   1023   if (++PI == pred_end(DestBB))
   1024     return false;
   1025 
   1026   P = *PI;
   1027   if (P != StoreBB) {
   1028     if (OtherBB)
   1029       return false;
   1030     OtherBB = P;
   1031   }
   1032   if (++PI != pred_end(DestBB))
   1033     return false;
   1034 
   1035   // Bail out if all the relevant blocks aren't distinct (this can happen,
   1036   // for example, if SI is in an infinite loop)
   1037   if (StoreBB == DestBB || OtherBB == DestBB)
   1038     return false;
   1039 
   1040   // Verify that the other block ends in a branch and is not otherwise empty.
   1041   BasicBlock::iterator BBI = OtherBB->getTerminator();
   1042   BranchInst *OtherBr = dyn_cast<BranchInst>(BBI);
   1043   if (!OtherBr || BBI == OtherBB->begin())
   1044     return false;
   1045 
   1046   // If the other block ends in an unconditional branch, check for the 'if then
   1047   // else' case.  there is an instruction before the branch.
   1048   StoreInst *OtherStore = nullptr;
   1049   if (OtherBr->isUnconditional()) {
   1050     --BBI;
   1051     // Skip over debugging info.
   1052     while (isa<DbgInfoIntrinsic>(BBI) ||
   1053            (isa<BitCastInst>(BBI) && BBI->getType()->isPointerTy())) {
   1054       if (BBI==OtherBB->begin())
   1055         return false;
   1056       --BBI;
   1057     }
   1058     // If this isn't a store, isn't a store to the same location, or is not the
   1059     // right kind of store, bail out.
   1060     OtherStore = dyn_cast<StoreInst>(BBI);
   1061     if (!OtherStore || OtherStore->getOperand(1) != SI.getOperand(1) ||
   1062         !SI.isSameOperationAs(OtherStore))
   1063       return false;
   1064   } else {
   1065     // Otherwise, the other block ended with a conditional branch. If one of the
   1066     // destinations is StoreBB, then we have the if/then case.
   1067     if (OtherBr->getSuccessor(0) != StoreBB &&
   1068         OtherBr->getSuccessor(1) != StoreBB)
   1069       return false;
   1070 
   1071     // Okay, we know that OtherBr now goes to Dest and StoreBB, so this is an
   1072     // if/then triangle.  See if there is a store to the same ptr as SI that
   1073     // lives in OtherBB.
   1074     for (;; --BBI) {
   1075       // Check to see if we find the matching store.
   1076       if ((OtherStore = dyn_cast<StoreInst>(BBI))) {
   1077         if (OtherStore->getOperand(1) != SI.getOperand(1) ||
   1078             !SI.isSameOperationAs(OtherStore))
   1079           return false;
   1080         break;
   1081       }
   1082       // If we find something that may be using or overwriting the stored
   1083       // value, or if we run out of instructions, we can't do the xform.
   1084       if (BBI->mayReadFromMemory() || BBI->mayWriteToMemory() ||
   1085           BBI == OtherBB->begin())
   1086         return false;
   1087     }
   1088 
   1089     // In order to eliminate the store in OtherBr, we have to
   1090     // make sure nothing reads or overwrites the stored value in
   1091     // StoreBB.
   1092     for (BasicBlock::iterator I = StoreBB->begin(); &*I != &SI; ++I) {
   1093       // FIXME: This should really be AA driven.
   1094       if (I->mayReadFromMemory() || I->mayWriteToMemory())
   1095         return false;
   1096     }
   1097   }
   1098 
   1099   // Insert a PHI node now if we need it.
   1100   Value *MergedVal = OtherStore->getOperand(0);
   1101   if (MergedVal != SI.getOperand(0)) {
   1102     PHINode *PN = PHINode::Create(MergedVal->getType(), 2, "storemerge");
   1103     PN->addIncoming(SI.getOperand(0), SI.getParent());
   1104     PN->addIncoming(OtherStore->getOperand(0), OtherBB);
   1105     MergedVal = InsertNewInstBefore(PN, DestBB->front());
   1106   }
   1107 
   1108   // Advance to a place where it is safe to insert the new store and
   1109   // insert it.
   1110   BBI = DestBB->getFirstInsertionPt();
   1111   StoreInst *NewSI = new StoreInst(MergedVal, SI.getOperand(1),
   1112                                    SI.isVolatile(),
   1113                                    SI.getAlignment(),
   1114                                    SI.getOrdering(),
   1115                                    SI.getSynchScope());
   1116   InsertNewInstBefore(NewSI, *BBI);
   1117   NewSI->setDebugLoc(OtherStore->getDebugLoc());
   1118 
   1119   // If the two stores had AA tags, merge them.
   1120   AAMDNodes AATags;
   1121   SI.getAAMetadata(AATags);
   1122   if (AATags) {
   1123     OtherStore->getAAMetadata(AATags, /* Merge = */ true);
   1124     NewSI->setAAMetadata(AATags);
   1125   }
   1126 
   1127   // Nuke the old stores.
   1128   EraseInstFromFunction(SI);
   1129   EraseInstFromFunction(*OtherStore);
   1130   return true;
   1131 }
   1132